top of page

Minutes That Get Done: The Research on Writing Up Decisions and Actions People Actually Act On.

13 hours ago
45 min read

The meeting ended at twenty past four and everybody left agreeing. By the following Tuesday there were three versions of the agreement in the building. One person believed the supplier review had been approved and would start in October. Another believed it had been approved in principle, subject to the finance figures landing first. A third had spent Monday writing the brief for it, because that was what they had heard. A fourth had left the room without registering that anything had been decided at all, and would say so, politely, in the next meeting.


The notes had gone round on the Friday and they were accurate. They were also eleven paragraphs of careful summary in which the word "agreed" appeared twice and nobody's name appeared once. I have written up a great many meetings, and the thing I did not understand for years is that the document everybody argues about, the minute, is rarely the document that makes anything happen. The research on how much of a conversation survives a week, on the wording that turns an intention into an action, and on what a record has to do to still be useful in March, is more precise about this than any template I was ever handed.


A small polished brass desk bell with a ring handle standing on a dark walnut desk beside a single folded cream card, a diagonal band of daylight falling across the pale wall behind them and both objects throwing long shadows across the grain of the wood, the Connected Assistants wordmark at the foot.
Writing it down is the easy half. The bell is the other one.



What you will get out of this


  • What two people who had the same conversation both remember a week later, measured, and why the answer is under five per cent

  • The wording rule that roughly trebled completion of difficult intentions, and the three words most action lists are missing

  • Why the formal minute and the action note are two different documents doing two different jobs, and what each one has to contain

  • What happens to an agreement when the person who owns it never says it out loud in the room

  • The laptop and longhand note-taking studies, what replicated and what did not, given straight

  • Why a record kept in one person's inbox is not a record, from forty years of research on how organisations remember

  • The honest state of the evidence on reminders, including the meta-analysis that cut the effect of nudges to almost nothing

  • The action note check: a tool that reads one action the way the research would and tells you what to change

  • The Action Note Pages: three pages for the note, the log and the follow-up rhythm




What the room actually keeps


Start with the thing nobody plans for, which is how little of a conversation survives contact with the following week.


In one study, pairs of people had a conversation and then, a week later, were asked to recall what had been said. One member of each pair wrote a memorandum of the conversation shortly afterwards, in the way a civil servant or a diplomat records a call. The people who wrote one recalled more of the conversation than the people who did not, and what both of them recalled was about equally accurate. Here is the number worth carrying around: of all the details of the conversation, only 4.7 per cent were recalled by both partners after a week. Two people, one conversation, and less than a twentieth of it held in common seven days later.


The older work in this area found the same shape. In a study of natural conversations, people could recall roughly a tenth of what had been said immediately afterwards, and about four per cent of it a month later, with recollections drifting steadily away from content and towards description and inference. The plain forgetting curve behind all of this has been replicated with modern methods and holds up, with a notable drop over the first twenty-four hours.


Then the part that matters if you are the one taking the record. Your position in the conversation changes what you keep. People remember what they said better than what they heard, and memory for who said what is weaker than memory for what was said. A study of people recalling conversations found that those who were overhearing rather than taking part recalled significantly less of the content, and attributed what they did recall to the wrong speaker more often. If you are in the corner writing while other people talk, you are closer to the overhearer than to the participant for at least part of the meeting, and the research says that is the position with the weakest memory for who said what.


There is an honest disagreement in this literature worth flagging. Several studies find people remember their own contributions better than their partner's, and one older study of natural conversations found the reverse, with people recalling more of their partner's side than their own. The likeliest reading is that attention and involvement shift the balance, which is precisely the variable that changes from one meeting to the next.


Two people had the same conversation. A week later, less than a twentieth of it was in both of their heads.

The other finding people do not expect is that remembering together does not reliably fix this. When groups recall something collectively, they tend to produce less than the same individuals would have produced working separately and having their answers combined, an effect known as collaborative inhibition. Discussion also spreads errors between the people having it. In one experiment, witnesses who discussed an event before being tested went on to report items that they could only have picked up from the other person, and this happened in 71 per cent of them. Socially encountered misinformation turned out to be more misleading than the same misinformation read on a page. And what gets said aloud actively pushes out what does not: selectively recounting part of a shared experience makes both the speaker and the listener more likely to forget the related parts that went unmentioned.


Read that last one twice, because it is the argument for writing your note properly. The summary you circulate does not simply describe what the room remembers. It becomes what the room remembers, and the items you left out become harder for everyone to retrieve.


A graphic titled How much of a conversation is still there later, with the eyebrow Measured recall of real conversations, from studies of natural talk and of conversation pairs. Standfirst: Nobody plans a meeting on the assumption that almost none of it will survive the week, and yet that is what the measurements keep showing. One hundred would mean every detail of the conversation was still available. Here is what people actually had. A cream card holds a dot plot headed Share of the conversation still recalled, out of 100, with an axis running from 0 to 100 in steps of 25. Three rows, each a single copper dot sitting close to the left-hand edge. Straight after the conversation, 1987 study of natural conversations, about 10. One month later, the same study, same people, about 4. Held by both people after a week, 2023 study of conversation pairs, 4.7. A footnote reads: The third line is the one that ends arguments. When two people who had the same conversation were asked a week later what was said, under five per cent of the details appeared in both accounts. Everything your meeting is relying on people to remember is somewhere in the white space to the right. Sources at the foot: Stafford, Burggraf and Sharkey (1987); Brown-Schmidt and colleagues (2023).
Everything your meeting is relying on people to remember lives in the white space on the right.

The practical fix the research points to is about format. When people reported a conversation using a timeline structure, laying out who said what and in what order, they reported more verbatim detail, made fewer sequencing errors and attributed more statements to the right speaker, without any loss of overall accuracy. Structure does not merely tidy the record. It changes how much of the record there is.


Try this. Write the note the same day, and write it in the shape the timeline studies used rather than as narrative. One line per decision, the name of the person who owns it against it, and the order things were settled in. If it has to wait until tomorrow, spend five minutes now listing the decisions alone and fill in the wording later, because the decisions are the part that gets quietly rewritten overnight.


Brown-Schmidt and colleagues (2023) had pairs of participants hold a conversation, with one member of each pair writing a memorandum of it shortly afterwards, and tested recall a week later; memorandum writers recalled more details than their partners while recall was equally and largely accurate for both, and only 4.7 per cent of the details of the conversation were recalled by both partners after the week's delay, indicating that contemporaneous note-taking increases the amount remembered rather than its accuracy. Stafford, Burggraf and Sharkey (1987) found that participants could recall only about ten per cent of their conversations immediately afterwards and about four per cent a month later, that delayed recall contained less content and more descriptive statements and inferences and was less accurate, and that participants recalled more of their partner's contributions than their own. Miller (1996) found in problem-solving conversations that partners had better memory for their own contributions than for each other's, with the difference exacerbated by topic importance, and that accurate recall correlated positively with social competence. McKinley, Brown-Schmidt and Benjamin (2017) found that developing common ground was positively related to both item memory for what was said and context memory for who said it, and that content that was spoken was remembered better than content that was heard. Brown-Schmidt and colleagues (2025) had 60 participants recall 60 conversations and found that overhearers recalled significantly less content than speakers or addressees and sourced what they recalled to the actual contributor less accurately. Brown-Schmidt and Benjamin (2018) reviewed distant memory for conversations for legal settings, covering how much can be recalled and the kinds of errors likely. Diachek and Brown-Schmidt (2024) analysed 59 conversational dyads and found that disfluencies and backchannelling promoted recall while disagreements and the use of "like" did not, with self-generated material better remembered overall. Murre and Dros (2015) replicated Ebbinghaus's forgetting curve by the method of savings across delays from twenty minutes to thirty-one days and concluded that the curve is not completely smooth, most probably showing a jump upwards starting at the twenty-four hour point. Weldon and Bellinger (1997) and Basden and colleagues (1997) established collaborative inhibition, in which collaborating groups recall less than the pooled recall of the same number of individuals working alone, with Basden and colleagues attributing it to disruption of individual retrieval strategies. Rajaram and Pereira-Pasarin (2010) reviewed collaborative memory and set out both the costs, including reduced recall and increased errors, and the conditions under which collaboration improves memory. Greeley and Rajaram (2023) reviewed how collaborative recall synchronises what and how group members later remember. Gabbert, Memon and Allan (2003) had members of dyads watch different videos of the same event and found that 71 per cent of witnesses who discussed the event beforehand went on to mistakenly recall items they could only have acquired during the discussion, with no age differences in susceptibility. Gabbert and colleagues (2004) compared misinformation encountered socially with the same misinformation embedded in a written narrative and found the socially encountered version significantly more misleading for both younger and older adults. Roediger, Meade and Bergman (2001) demonstrated the social contagion of memory, in which items falsely suggested by a partner during collaborative recall were later recalled as though seen. Cuc, Koppel and Hirst (2007) demonstrated socially shared retrieval-induced forgetting, in which listening to a speaker selectively recall material induced forgetting of related unmentioned material in the listener, including in free-flowing conversation. Coman, Manier and Hirst (2009) extended the effect to well-rehearsed, emotionally intense memories of the September 11 attacks. Hirst and Echterhoff (2012) reviewed remembering in conversation, covering collaborative facilitation and inhibition, social contagion, audience tuning and retrieval-induced forgetting. Harris, Paterson and Kemp (2008) reviewed group remembering across paradigms. Marsh and Tversky (2004) asked participants to record how they told others about events from their own lives and found that participants labelled 61 per cent of their retellings as distorted through exaggeration, omission, minimisation or addition, while labelling only 42 per cent as inaccurate. Bourgeois, Salapska-Gelleri and McElroy (2026) presented seven memory tasks to participants recalling individually or in groups of two or three and found that group discussion increased actual memory for information and also increased misattribution and persistence while decreasing transience. Hope and colleagues (2019) found across three experiments, with a large effect size and 134 participants in total, that a timeline reporting format produced more verbatim information, fewer sequencing errors and more correct speaker attributions than free recall, without compromising accuracy.




Two documents, and why one cannot do both jobs


Most of the frustration around minutes comes from asking one document to do two incompatible jobs at once.


The first job is the record. In the United Kingdom, a company is required to keep minutes of its directors' meetings and to hold them for ten years, and the same formality applies to boards of trustees, committees with delegated authority and anything that might later be read by an auditor, a regulator or a court. A practitioner account of minute-taking sets out the four purposes such a document serves: it is a communication tool, a legal document, an instrument of management practice and a historic record. That document needs the date and place, who was present and who sent apologies, whether the meeting was quorate, the motions put and the outcomes of any votes, declarations of interest, and the decisions taken in words that will still be defensible when nobody involved can be found. It is deliberately impersonal, and it summarises discussion rather than reproducing it.


The second job is entirely different. Somebody has to leave the room knowing what they are doing on Wednesday morning. That is the action note, and it wants the opposite qualities: names, verbs, dates, and no summary at all.


The reason the two get muddled is partly that the models most people are taught come from the wrong place. An analysis of professional writing textbooks against actual workplace practice found that the textbooks, where they mentioned minutes at all, offered parliamentary-style examples, while the norm in workplaces was an action-oriented style whose function was to solidify consensus and hold named individuals accountable. The style you were taught was built for a motion and a seconder. The style your team needs was built for a Tuesday.


The research that has gone inside real boardrooms is quietly encouraging about how much a record can carry. Researchers have used the actual minutes of board and committee meetings as their raw data, reconstructing what boards spend their time on, which tells you that a well-made minute is a rich enough account of an organisation's decisions to support serious research. Video studies of board meetings add the detail that discussion varies enormously between agenda items within a single meeting, with different directors leading on different items, and that how evenly directors participated was associated with how effective the board was thought to be. A single flat summary of a meeting erases exactly that texture. Across thirty-five organisations, the use of formal mechanisms for storing and distributing information, such as agendas and supporting documents, varied by meeting format rather than being applied consistently.


A graphic titled The three documents a meeting should produce, with the eyebrow One meeting, two jobs, and a third that pays for itself six months later. Standfirst: Most of the frustration about minutes comes from asking one file to protect the organisation and get the work done at once. They want opposite qualities, so write them separately and let each of them be good at the one job it has. Three cream cards side by side. Document one, The formal minute: For the record, and for whoever reads it when nobody involved can be found. Date, place, who was there, apologies, whether it was quorate, declared interests, motions and votes, and the decisions in defensible words. Impersonal by design. Circulated for approval at the next meeting. Document two, The action note: For the people who have to do something on Wednesday. Names, verbs, triggers and dates, one line each, and no summary at all. The wording carries the research: when this happens, this person will do this, by this date. Sent the same day, to everyone it names. Document three, The decision log: For the version of you that gets asked in March why on earth this was agreed. What was decided, why, what was ruled out, what has to stay true, and when it gets looked at again. Four minutes to write, and kept where anyone on the team can find it later. Sources at the foot: Smith (2013); Wolfe (2006); Onditi and colleagues (2004); Bujar and colleagues (2020).
One file cannot protect the organisation and get the work done. Give the jobs to different documents and both get easier.

The formal minute protects the organisation. The action note protects the work. Asking one document to do both produces a document that does neither well.

So separate them, and let each be good at its own job. The minute is circulated for approval at the next meeting, because approval is what makes it a record. The action note goes out within hours and belongs to whoever has to act, because speed is what makes it useful. Both can come from the same set of notes, and they should never be the same file.


Try this. Take the last meeting you wrote up and mark every sentence M or A: minute or action. If more than a third of it is neither, you have found where the time goes. Send the A sentences within the day in a table with a name against each row, and let the M sentences take the week they need.


Smith (2013) set out the four major purposes of well-documented meeting minutes as communication tools, legal documents, instruments of good management practice and historic records, and listed the essential contents, including the meeting's location, time and date, purpose and officiating officer, attendance and whether a quorum was achieved, approval of previously written minutes, unfinished and new business, a record of all motions and the outcomes of all votes, key discussion items and agreed actions, and adjournment. Wolfe (2006) compared the treatment of meeting minutes in professional writing textbooks with workplace practice, finding that textbooks rarely mention minutes in chapters on collaboration and provide examples in a parliamentary style rather than the action-oriented style that is the norm in most workplace settings, and argued from interviews with three engineering managers and the professional communication literature that minutes project a team forward by solidifying consensus and holding individuals accountable for actions. Schwartz-Ziv and Weisbach (2013) used the actual minutes of board and board-committee meetings as data to reconstruct what boards do. Bezemer, Nicholson and Pugliese (2014) analysed videotaped board meetings at two Australian companies and found that while board meetings appear similar at a surface level, interaction varies significantly at a deeper level, with different directors participating during different stages of discussion, pointing towards the agenda item as the appropriate unit of analysis. Pugliese, Nicholson and Bezemer (2015) observed three board meetings at each of two corporations and found that a few directors led the discussion while the contributing directors changed across agenda items, and that directors' inclusiveness and evenness of participation were associated with higher perceptions of board effectiveness. Forbes and Milliken (1999) set out boards of directors as strategic decision-making groups whose effectiveness depends on group processes. Volkema and Niederman (1995) observed meetings within thirty-five different organisations and found that the use of formal mechanisms for storing and distributing information, such as agendas and supporting documents, varied by meeting format, and that the availability and use of meeting technologies was limited.




How to word a commitment so it happens


This is the part of the research I wish somebody had shown me in my first year, because it is specific, it is old, and it changes an action list in about a minute.


The finding is called the implementation intention, and the idea is that an intention becomes far more likely to be carried out when it is furnished in advance with a when, a where and a how, in the form "if situation Y arises, then I will do X". In the founding studies, difficult intentions were completed about three times as often when people had given them that structure. A meta-analysis of ninety-four independent tests found a medium-to-large effect on goal attainment, with a d of .65, and identified the mechanism: forming the plan makes the specified moment more noticeable, and makes the response more automatic once that moment arrives. A much larger synthesis published recently covered 642 independent tests and found the technique effective across thinking, feeling and behaviour, with effects ranging from .27 to .66, and with three conditions raising the effect. Plans worked better when they had a contingent if-then format, when the person was motivated to pursue the goal, and when the plan had been rehearsed.


Hold on to those three conditions, because they map exactly onto what a meeting can and cannot do. You can supply the format. You can check the motivation by asking. And you can supply the rehearsal by having the person say the line out loud before they leave.


The field evidence is unusually good, because plan-making has been tested at scale outside the laboratory. In a field experiment during the 2008 United States presidential election with 287,228 people, a telephone script that helped the voter form a plan, covering when they would vote, where they would be coming from and what they would be doing beforehand, raised turnout by 4.1 percentage points among those contacted, rising to 9.1 points in single-voter households. A standard encouragement call raised nothing measurable, and neither did asking people to predict whether they would vote. Encouragement and prediction are what most meetings do. The plan is what they usually skip.


In health settings, a trial with 211 cardiac patients found that combining an action plan covering when, where and how with a coping plan for the anticipated obstacle produced significantly more exercise two months after discharge than either standard care or the action plan alone. Reviews of the planning literature recommend if-then format, cues that are salient and relevant to the person, examples of what a cue looks like, plans that are guided rather than left entirely to the person to invent, and boosters afterwards. A specific version has been tested for groups as well: collective plans phrased as "we" rather than "I" changed how groups performed, and worked through cooperation in a way that individual plans did not.


There are two counterweights, and they both make the advice sharper.


The first is that plans are only as good as the planner. In a pair of studies, action plans that people generated spontaneously predicted their later behaviour only when the person scored well on a standardised test of planning skill. Poor planners' own plans did not help them at all. The useful half of that finding is the other half: a supplied planning intervention improved goal attainment specifically in the poor planners. Writing the plan for somebody, in the note, is the condition under which planning works for the people it does not come naturally to, rather than a piece of bureaucracy.


The second is that this does not scale infinitely. Research on multiple goals found that the benefit of implemental planning for one goal did not extend to several at once, because planning many things at once draws attention to how hard the whole set will be and undermines commitment to all of it. A meeting that produces fourteen planned actions per person has not produced fourteen actions.


A graphic titled Six lines from a real action list, rewritten, with the eyebrow The wording rule, from sixty years of research on turning an intention into an action. Standfirst: On the left is the phrasing most action lists use. On the right is what the research on if-then planning, identifiability and cues would have you write instead. The rewrite takes about eleven seconds a line. Six rows, each with a dark panel on the left headed Instead of and a cream panel on the right joined by a copper arrow. Instead of Finance to review the supplier list: Name the person, not the function. Effort drops measurably when a contribution cannot be traced to a particular someone, and a team is nobody. Instead of We will look into the venue options: Use a verb you could watch happen. Looking into something is not an act, whereas ringing three venues and sending one page of costs by Friday plainly is. Instead of Sam will do this soon: Give it a date, and a moment that starts it. If-then wording roughly trebled completion of difficult intentions, because the cue does the remembering on your behalf. Instead of Agreed to move the review to October: Say what has to stay true. A commitment with no recorded condition is one nobody can reopen without losing face. Instead of Eleven actions, one per person, every week: Pick three. Planning several goals at once drew attention to how hard the whole set would be and undermined commitment to all of it. Instead of Send it and hope: Read each one aloud and let the owner confirm it in their own words. Rehearsed plans worked better, and so did commitments other people heard. Sources at the foot: Gollwitzer and Brandstatter (1997); Sheeran, Listrom and Gollwitzer (2025); Karau and Williams (1993); Dalton and Spiller (2012).
Eleven seconds a line, and about three times the chance of anybody doing it.

So the wording rule is small and it is worth being strict about. Every action gets a named person, a verb describing an observable act, a moment that triggers it, and a date. "Finance to review the supplier list" fails on three counts. "When the October figures land on the fifteenth, Sam will compare the top six suppliers on unit cost and send one page to the group by the twentieth" passes on all four, and takes eleven seconds longer to write.


Try this. Go through your last action list and rewrite each line as: when [trigger], [name] will [observable act], by [date]. Then cut the list to the three that matter most this fortnight and move the rest to a parked list with a review date. The research on multiple goals says the three will go better for the others being out of sight.


Gollwitzer (1999) set out implementation intentions as plans of the form "if situation Y is encountered, then I will initiate goal-directed behaviour X", which delegate control of the response to an anticipated cue. Gollwitzer and Brandstätter (1997) found across three studies that difficult goal intentions were completed about three times more often when participants had furnished them with implementation intentions, and that implementation intentions facilitated the immediate initiation of goal-directed action when the intended opportunity was encountered. Gollwitzer and Sheeran (2006) meta-analysed ninety-four independent tests and found implementation intentions had a positive effect of medium-to-large magnitude, d = .65, on goal attainment, working through heightened accessibility of the specified opportunity and automation of the goal-directed response. Sheeran, Listrom and Gollwitzer (2025) meta-analysed 642 independent tests and found implementation intentions effective for cognitive, affective and behavioural outcomes, with effect sizes between .27 and .66, larger when plans had a contingent if-then format, when participants were highly motivated to pursue the goal, and when plans were rehearsed, and developed a taxonomy of the cues and responses that plans specify. Nickerson and Rogers (2010) ran a field experiment during the 2008 United States presidential election with 287,228 people and found that helping voters form a voting plan increased turnout by 4.1 percentage points among those contacted, rising to 9.1 points among single-eligible-voter households, while a standard encouragement call and a self-prediction question had no significant impact. Sniehotta, Scholz and Schwarzer (2006) randomised 211 cardiac rehabilitation patients and found that those given combined action planning and coping planning did significantly more physical exercise two months after discharge than those given action planning alone or standard care. Hagger and Luszczynska (2014) reviewed planning interventions in health behaviour and recommended that effective plans adopt if-then formats, account for salient and relevant cues, include examples of cues, be guided rather than user-defined, and include boosters. Bélanger-Gravel, Godin and Amireault (2013) meta-analysed the effect of implementation intentions on physical activity. Thürmer, Wieber and Gollwitzer (2017) found across two experiments with three-member groups that collective implementation intentions phrased as "we" plans improved performance through cooperation, whereas individual "I" plans did not rely on cooperation. Trötschel and Gollwitzer (2007) found that implementation intentions supported the pursuit of prosocial goals in negotiation. Allan, Sniehotta and Johnston (2013) found in a prospective study of 72 people that spontaneously generated action plans explained significantly more variance in later behaviour when produced by a skilled rather than a poor planner, and in an experiment with 144 people that a supplied planning intervention significantly improved goal attainment only among poor planners. Dalton and Spiller (2012) found that the benefits of implementation intentions for a single goal did not typically extend to multiple goals, because implemental planning drew attention to the difficulty of executing several goals and undermined commitment to them. Lehmann-Willenbrock, Allen and Kauffeld (2013) coded 59 team meetings from 19 organisations and found that procedural communication, meaning verbal behaviour that structures discussion, promoted proactive statements about who will do what and when and significantly inhibited dysfunctional meeting behaviour, and that the more evenly procedural behaviour was distributed across members the more satisfied they were with the outcome.




Saying it out loud, and what that adds


Writing the action down is half of it. The other half is whether the person who owns it ever said so in front of anybody.


The oldest strand of this work is on commitment. In a pair of field experiments in the 1970s, homeowners who agreed to try to save energy and expected to be named publicly used measurably less gas and electricity over the following month than homeowners who agreed under the same conditions but expected to remain anonymous. The energy saved came from nothing more than the expectation of being identified. A later meta-analysis of environmental studies containing a commitment manipulation concluded that commitment making does produce behaviour change in both the short and the long term compared with control conditions, while being careful to say that the mechanisms underneath are still not well understood.


The second strand is accountability, and it is more nuanced than the word usually implies. The major review of the field found that accountability is not a general improver. What it does depends on the ground rules: who you are answerable to, whether their views are known, and whether you are answerable for the process you followed or only for how it turned out. Being accountable for the process tends to sharpen judgement. Being accountable purely for the outcome can make judgement worse, which is an argument for a note that records what somebody undertook to do rather than one that only records whether it worked.


The third strand is about naming, and it is the one that changes meetings most. When individual contributions cannot be identified, people reliably put in less effort, an effect established across a long line of studies and confirmed by meta-analysis: making contributions identifiable is one of the conditions that reduces social loafing. The related finding from the bystander literature is that responsibility spreads thin when several people could act and no one has been singled out. "The team will pick this up" is a sentence that describes nobody, and the literature has been telling us so since 1968.


The economics of commitment devices adds a practical note. Voluntary commitments with a real cost attached work, sometimes remarkably well: a savings-account device for smokers produced lasting cessation, still visible in surprise tests at twelve months, and workplace commitment contracts around gym attendance produced changes detectable years later. You are not going to ask a colleague to stake money on a supplier review. What you can borrow is the structure, which is a commitment that is specific, voluntary and witnessed.


Now the counterweight, because public commitment has a dark side and the literature names it plainly. Once people have publicly committed to a course of action, they become more likely to keep pouring resources into it after the evidence has turned. The classic demonstration is nearly fifty years old and a meta-analysis has since mapped the conditions that make it worse. If the note that records the commitment never records the assumption it rested on, then the commitment is the only thing left to be loyal to.


A commitment nobody witnessed is a hope. A commitment nobody can revisit is a trap.

Which gives you the shape of a good closing minute of a meeting. Say each action back, in the owner's hearing, in the owner's words, with the date. Then write beside it the condition under which it should be reopened.


Try this. In the last two minutes of your next meeting, read the actions aloud and ask each owner to confirm theirs in their own words, and write down the words they use. It takes ninety seconds, it turns four people's assumptions into one agreement, and the phrasing you capture is the phrasing that will make sense to them on Wednesday.


Pallak and Cummings (1976) interviewed homeowners who expected either to be personally identified as having agreed to attempt energy conservation or not to be identified, and found through utility-meter readings for the following month that those under public commitment showed a lower rate of increase in both natural gas and electricity use than those under private commitment or in the no-interview control. Lokhorst and colleagues (2013) meta-analysed environmental studies containing a commitment manipulation and concluded that commitment making leads to behaviour change in the short and long term, especially compared with control conditions, while noting that the underlying mechanisms need better understanding. Cialdini and Goldstein (2004) reviewed compliance and conformity research and the goals that make people susceptible to influence, including the maintenance of a consistent self-concept. Deutsch and Gerard (1955) distinguished normative from informational social influence and examined the effect of commitment to a publicly announced judgement. Lerner and Tetlock (1999) reviewed the accountability literature and found that the effects of accountability depend on the ground rules, that accountability attenuates some cognitive biases and amplifies others, and that accountability for process differs in its effects from accountability for outcomes. Siegel-Jacobs and Yates (1996) compared procedural and outcome accountability experimentally and found that the two had different effects on judgement quality, with outcome accountability capable of degrading performance. Latané, Williams and Harkins (1979) demonstrated social loafing, in which individual effort declines when contributions are pooled and cannot be identified. Karau and Williams (1993) meta-analysed social loafing and found it robust across tasks and populations, and moderated by factors including the identifiability of individual contributions. Darley and Latané (1968) demonstrated diffusion of responsibility, in which the presence of others who could act reduced the likelihood that any individual acted. Bryan, Karlan and Nelson (2010) reviewed the evidence on hard and soft commitment devices and the demand for them. Giné, Karlan and Zinman (2010) offered smokers a voluntary commitment savings account forfeited on failing a nicotine test; eleven per cent took it up, those offered it were three percentage points more likely to pass the six-month test, and the effect persisted in surprise tests at twelve months. Royer, Stehr and Sydnor (2015) ran a large workplace field experiment and found that a four-week gym incentive produced only small lasting effects, while adding a self-funded commitment contract at the end produced significant long-run changes still detectable several years later. Staw (1976) demonstrated escalation of commitment, in which decision makers allocated more resources to a previously chosen course of action after negative results, particularly when personally responsible for it. Sleesman and colleagues (2012) meta-analysed the determinants of escalation of commitment and mapped the psychological, social, project and structural factors that increase it.




What to write while it is happening


Note-taking research has two halves that are worth keeping separate, because they are usually collapsed into one argument about laptops.


The first half is the encoding function: writing something down helps the writer understand and remember it. The second is the external storage function: the note exists so that somebody can consult it later. For a student revising for an examination, these compete. For you, writing up a meeting for eleven other people, the external storage function is the entire point, and this changes which findings apply.


On encoding, the honest position is that the evidence is more mixed than the famous headline suggests. The well-known 2014 studies found that students taking notes on laptops performed worse on conceptual questions than students writing longhand, and that the mechanism was verbatim transcription: laptop users typed more of the lecturer's actual words rather than reframing the material in their own. It is a lovely finding and it has had a hard time since. A large replication and extension in 2019 failed to reproduce the central result. A direct replication in 2021, with 74 laptop and 68 longhand participants, found the same difference in note-taking behaviour, with laptop users recording more words overall and more verbatim words, and found no difference at all in quiz performance; exploratory meta-analyses across eight similar studies echoed that pattern. That same replication found that a higher word count went with better quiz performance, while the association between verbatim overlap and worse performance was not robust.


What survives all of that is two findings, and they happen to be the two that matter most for a minute-taker. More words is generally better for what you can reconstruct afterwards. And review is what converts notes into knowledge: note-taking improved understanding at the deeper level of the situation model in experiments testing both immediately and a week later, and studies of revising lecture notes found that pauses, revision and working with a partner all improved the notes and what people got from them. A meta-analysis of the encoding effect of note-taking mapped where it is limited, and the practical constraint is capacity. Recording notes while comprehending new material is demanding work, measurably more effortful than reading or learning the same material, and one of the strongest predictors of who takes good lecture notes is how fluently they can get words down.


That last sentence explains something many people quietly worry about. If you feel you cannot follow the argument and write it at the same time, the research says that is a real cognitive constraint rather than a personal failing, and the answer is to change the format rather than to concentrate harder.


Here is where the conversation research rejoins the note-taking research, and it is the most useful thing in this section. The study where one person in each pair wrote up the conversation afterwards found that writing it up increased how much they remembered a week later, without making them more accurate. Writing the note buys you volume, not truth. Which means the check on accuracy has to come from somewhere else, and the only somewhere else available is the other people who were in the room.


The newer work on automated recaps points the same way. When researchers built and tested two kinds of machine-generated meeting recap, a set of highlights and a structured hierarchical minutes view, both were valuable in different situations, and what users did with them, adding, editing and deleting, was where the value appeared. Studies of how people actually use meeting records found that they return to them for specific, narrow purposes rather than reading them through. Nobody sits down and rereads a meeting; they search it for the one thing they need.


Try this. Capture generously in the room and shape afterwards. Take more words than feels elegant, with a mark in the margin every time a decision or an owner appears, then spend fifteen minutes turning the marks into the action note and the rest into the minute. Send the action note to the people named in it with the line, "this is what I have you down for, tell me if it is wrong", which is the accuracy check the research says you cannot perform alone.


Kiewra and colleagues (1991) distinguished the encoding and external storage functions of note-taking and tested note-taking and review techniques against them. Mueller and Oppenheimer (2014) found across three studies that students who took notes on laptops performed worse on conceptual questions than students taking notes longhand, and attributed this to laptop users transcribing lectures verbatim rather than reframing material in their own words. Morehead, Dunlosky and Rawson (2019) conducted a replication and extension and did not reproduce the central advantage of longhand note-taking. Urry and colleagues (2021) ran a direct replication of the original Study 1 with 74 laptop and 68 longhand participants and found that laptop participants took notes containing more verbatim words and more words overall, that laptop participants did not perform worse on the quiz, and that exploratory meta-analyses of eight similar studies echoed this pattern; higher word count was associated with better quiz performance, while the association between verbatim overlap and worse performance was not robust in the replication. Kobayashi (2005) meta-analysed the encoding effect of note-taking and examined what limits it. Bohay and colleagues (2011) found across two experiments, with tests immediately and one week later, that note-taking improved performance at the situation model level of understanding, with review of notes manipulated separately. Luo, Kiewra and Samuelson (2016) found that revision, pauses and working with a partner affected the completeness of lecture notes and subsequent achievement. Piolat, Olive and Kellogg (2005) reviewed the working memory demands of note-taking and showed through dual-task measures that note-taking demands more cognitive effort than reading or learning, though less than original written composition. Peverly and colleagues (2007) found that transcription fluency was a leading predictor of skill in lecture note-taking. Brown-Schmidt and colleagues (2023) found that writing a memorandum of a conversation increased the number of details recalled a week later without increasing the accuracy of that recall. Asthana and colleagues (2025) designed and evaluated two machine-generated meeting recap representations, important highlights and a structured hierarchical minutes view, with users in the context of their own work meetings, and found both valuable in different contexts with users' additions, edits and deletions revealing what they needed from a recap. Whittaker, Laban and Tucker (2006) studied how people use meeting records ethnographically and drew out the implications for systems designed to support retrieval. Whittaker and colleagues (2008) designed and evaluated systems for capturing and retrieving meeting interaction and found that retrieval needs were specific and targeted.




The record as the team's shared memory


There is a body of research that explains why the location of a note matters as much as its contents, and it starts with a laboratory study about assembling radios.


Groups whose members had been trained together recalled more of the assembly procedure and produced better radios a week later than groups whose members had been trained separately. The analysis showed the advantage came from a transactive memory system: the group had developed a shared sense of who knew what, so that each person could rely on the others to hold part of the task. A follow-up ruled out the simpler explanation that group training merely improved communication. Since then the idea has been extended from small groups to whole organisations, where it describes overlapping networks of people using each other as external memory, and reviewed comprehensively.


The organisational half of this literature is where the warning sits. An organisation's memory lives in several places at once: in people, in routines, in culture, in physical structures, and in records. A study of a multinational consultancy identified several distinct memory systems running in parallel, including social networks, formal knowledge centres and computer-based systems, each with its own reach and its own failure mode. The broader field has since been reviewed as organisational memory studies, which treats forgetting as an active organisational process rather than an accident.


And organisations forget faster than anyone budgets for. Two studies of learning inside real organisations, one across manufacturing plants and one across thirty-six pizza franchises tracked weekly for a year and a half, found that knowledge acquired through doing the work depreciates rapidly, so that cumulative experience substantially overstates how much a group still knows. The franchise study found something sharper still: knowledge transferred between stores owned by the same franchisee and did not transfer between stores owned by different ones. The ties, not the industry, carried the learning.


Put those two findings together and you get the practical rule. What one person remembers decays quickly, and what a group knows is held in the pattern of who knows what, which breaks the moment somebody leaves. The written record is what stops a departure from taking a year of decisions with it, and a record that lives in one person's sent items cannot do that job. It is the same information, sitting somewhere the system cannot reach.


None of which requires a platform. It requires one agreed place, one naming convention and one rule about what goes there, which is a great deal less work than the alternative and is, in my experience, the single most valuable thing a person in this seat quietly installs.


Try this. Pick the place where the action notes and decisions for one recurring meeting will live, name the files by date and meeting rather than by topic, and write a single line at the top of each one saying who to ask about it. Then tell the group where it is, twice, because telling them once is telling them nothing.


Liang, Moreland and Argote (1995) trained participants to assemble transistor radios either in groups or individually and tested them a week later in three-person groups; groups whose members had trained together recalled more of the procedure and produced better-quality radios, and analysis of videotape indicated the effect worked primarily through the development of transactive memory systems. Moreland and Myaskovsky (2000) tested whether the group training advantage came from transactive memory or from improved communication and found evidence for transactive memory. Ren and Argote (2011) reviewed twenty-five years of transactive memory systems research and set out an integrative framework of its dimensions, antecedents and consequences. Peltokorpi (2012) extended the transactive memory concept to organisations, defining organisational transactive memory as overlapping networks of interdependent work groups using each other as external cognitive aids. Walsh and Ungson (1991) set out organisational memory as information stored across individuals, culture, transformations, structures, ecology and external archives. Stein (1995) reviewed the conceptual foundations of organisational memory and the processes of acquisition, retention, maintenance and retrieval. Olivera (2000) studied a multinational business consulting organisation and identified multiple parallel memory systems, including social networks, knowledge centres and computer-based systems, assessing the characteristics and perceived effectiveness of each. Foroughi and colleagues (2020) reviewed organisational memory studies and the ways remembering and forgetting shape organising. Argote, Beckman and Epple (1990) found across multiple organisations that knowledge acquired through production depreciates rapidly and that cumulative output significantly overstates the persistence of learning. Darr, Argote and Epple (1995) tracked thirty-six pizza stores weekly over eighteen months and found that unit costs fell with experience, that knowledge acquired through learning by doing depreciated rapidly, and that knowledge transferred across stores owned by the same franchisee but not across stores owned by different franchisees.




The follow-up that does the work


An action note with no follow-up date is a wish with a font.


The cognitive research explains why. Remembering to do something in the future is a distinct kind of memory, and the evidence says people use two different routes to it. They either monitor the environment for the moment to act, which is effortful and competes with everything else they are doing, or they rely on an anticipated cue to bring the intention back automatically. The theoretical account of goal memory says the same thing in different language: pending goals sit in memory at a level of activation that decays, and cues are what retrieve them. This is exactly why the if-then wording works, and it is exactly why an action with no cue attached dies quietly.


It is also why interruptions are so expensive for the person holding the list. In a controlled study, interruptions averaging 4.4 seconds tripled the rate of sequence errors on the following trials, and interruptions of 2.8 seconds doubled them. Four and a half seconds of somebody's head round the door. The person trying to hold six pending actions in their head across a day of interruptions is fighting a battle the research says they will lose, and the note is the offloading that makes the day survivable rather than a tidiness measure.


Reminders work, and the field evidence is decent. A field experiment with gym members found that simple weekly reminders substantially raised attendance over a long period, that people responded within hours, and that the response recurred for every subsequent reminder rather than fading. A Cochrane review of mobile phone reminders for healthcare appointments found they improve attendance. Both of those are about a specific action with a specific time, which is the kind of action a meeting produces.


A graphic titled What actually moves an action, with the eyebrow Five things that change whether an action on a page becomes a thing that happened. Standfirst: None of these is about willpower, and all of them are things the person writing the note controls. The left-hand column is the move, and the right-hand column is what the evidence says it buys you. Five rows, each a copper tile on the left and a cream panel on the right. A cue, if-then plans, 642 tests: Pending intentions sit in memory at a level that fades, and a specific anticipated moment is what brings them back. Effects were larger when plans had an if-then format, when the person wanted the outcome, and when the plan had been rehearsed. Write the moment that starts it, not only the deadline. One name, social loafing: Effort reliably drops when individual contributions cannot be told apart, and responsibility spreads thin when several people could act and nobody was singled out. One owner per line, even when three people will help. Said out loud, public commitment: Householders who expected to be named publicly used less gas and electricity than those promised anonymity, and commitment making changed behaviour in the short and the long run against controls. Ninety seconds of reading the actions back. One place, shared memory: What a group knows sits in the pattern of who knows what, and it breaks when somebody leaves. Knowledge gained by doing the work depreciates fast. One agreed place and one naming convention beats any amount of goodwill. One chase, reminders, and cost: Weekly reminders raised gym attendance within hours and kept working for each new one. They also raised unsubscribes sharply in a charity experiment, so every reminder spends a little goodwill. Few, precise, and attached to something already agreed. Sources at the foot: Sheeran, Listrom and Gollwitzer (2025); Karau and Williams (1993); Pallak and Cummings (1976); Damgaard and Gravert (2018).
None of these is willpower, and all of them are things the person holding the pen decides.

Now the honesty, because this is the area where the evidence has been most contested in the past few years. A large meta-analysis of choice architecture interventions concluded that nudges reliably promote behaviour change across domains. A response published in the same journal argued that the same data show substantial publication bias, and that once corrected, the evidence for nudging largely disappears, with the true effect possibly as low as d = 0.08. That argument is still live, and I am not going to pretend it is settled. My reading is that the strongest, most replicated cases are the narrow ones, a specific person being reminded about a specific commitment at a specific time, and the weakest are the broad claims about changing behaviour in general.


There is one more cost worth knowing, because it applies directly to the person who sends the follow-ups. In a large field experiment with a charity, reminders did raise donations and they also substantially raised the rate at which people unsubscribed from the mailing list. Reminding has a price. Every reminder you send spends a small amount of the recipient's goodwill, which is an argument for few reminders, precisely targeted, attached to something the person already agreed to.


Try this. Build the follow-up into the note rather than into your memory. Against each action, write the date you will check and the event that triggers the check, then put one recurring fifteen-minute slot in your week for all of them together. One person chasing six things once a week costs the team far less goodwill than six separate messages sent whenever they occur to you.


McDaniel and Einstein (2000) set out a multiprocess view of prospective memory in which remembering a delayed intention is supported either by strategic monitoring of the environment or by relatively automatic retrieval triggered by an anticipated cue, with the balance depending on the task and the cue. Altmann and Trafton (2002) modelled goal-directed cognition in terms of activation and associative priming rather than a goal stack, with an interference constraint from residual memory for old goals, a strengthening constraint on encoding a new goal, and a priming constraint describing the role of cues in retrieving pending goals. Altmann, Trafton and Hambrick (2014) found that interruptions averaging 4.4 seconds tripled the rate of sequence errors on post-interruption trials relative to baseline, and that interruptions averaging 2.8 seconds doubled them, while non-sequence errors showed no interruption effect. Calzolari and Nardotto (2017) ran a field experiment with gym users and found that simple weekly reminders substantially increased attendance over an extended period, that the response was immediate within hours and recurred for every subsequent reminder, consistent with limited attention, with some evidence of habit formation persisting after treatment. Gurol-Urganci and colleagues (2013) conducted a Cochrane review of mobile phone messaging reminders for attendance at healthcare appointments and assessed their effects on attendance, costs and health outcomes. Mertens and colleagues (2022) meta-analysed over a decade of choice architecture research and concluded that such interventions successfully promote behaviour change across behavioural domains, populations and locations. Maier and colleagues (2022) reanalysed that meta-analysis and argued that once the moderate publication bias it reports is appropriately corrected for, no evidence for the effectiveness of nudges remains, with the true effect size possibly as low as d = 0.08. Damgaard and Gravert (2018) ran a large-scale field experiment with a charity and found that reminders increased donations and also substantially increased unsubscriptions from the mailing list, estimating a welfare loss of $2.35 per reminder for non-givers against a net benefit of $0.18 per reminder to the charity.




The log you will be grateful for in six months


The last piece is the one nobody budgets time for, and it is the one that quietly decides how good next year's decisions are.


A study of how decisions are tracked from documentation such as minutes found that decisions, their reasoning and the relationships between them are frequently not recorded at all, or are buried deep in prose, and that the consequence is later decisions being delayed or made on the wrong basis. A survey of how organisations document their major decisions found that where systems exist, they are used mainly to record outcomes rather than the process, that the information recorded is not always used afterwards, and that feedback loops comparing what was expected with what happened are largely absent. Work on decision reuse in software architecture found that practitioners reuse past decisions frequently and are confined to the decisions they already happen to know about, and concluded that improving documentation and discoverability has a large potential to increase reuse. An expert survey of architects and team leads across twenty-two companies mapped what kinds of decisions get captured in practice and what influences them.


There is a whole tradition of tools built to capture the reasoning behind a decision, going back to hypertext systems in the late 1980s that linked issues, positions and arguments, and continuing through architectures for organisational memory that record the decision, the rationale and who was in the room. The tools have come and gone, and the requirement behind them has stayed exactly where it was.


Why does the reasoning matter so much more than the decision? Because of one of the most reliable findings in psychology. Once you know how something turned out, you cannot reconstruct what you believed beforehand. Hindsight bias has been meta-analysed and reviewed extensively, and it is made of memory distortion, changed beliefs about how likely the outcome was, and changed beliefs about your own foresight. Its consequences include narrowing attention onto a single causal story and general overconfidence. The intervention that reduces it is being asked to consider alternative explanations, which is more or less the same thing as having written down at the time what else you considered.


The same literature offers the trick for writing a good log entry. Prospective hindsight means describing a future event as though it has already happened and explaining why. The research shows that whether the outcome is treated as certain changes the kind of explanation people produce. In practice it turns a bland note into a useful one: rather than "we agreed to move the supplier review to October", you write what you expect to be true by then and what would make you think again.


And the reason this is worth doing now rather than later is that later does not come. A survey of research and development projects found that only one in five received a post-project review at all, and that where they did happen they were constrained by lack of time, attention, personal interest and ability, and focused on technical output rather than process. Reviews of project learning methods say much the same. The log has to be written while the meeting is still warm, because the review you are relying on to reconstruct it is probably not going to happen.


A good entry needs five things and takes four minutes: what was decided, why, what was ruled out and why, what has to be true for it to hold, and when it will be looked at again.


Try this. Add one line to your action note called "what we ruled out", and one called "revisit when". You will feel silly writing them for about a fortnight. The first time somebody in a meeting asks why on earth the decision was made that way, you will have the answer in front of you, and the person who asked will assume you have a better memory than anybody in the building.


Onditi and colleagues (2004) studied how decisions are tracked from documentation such as meeting minutes and found that decisions, their rationale and the relationships between decisions are frequently not recorded or are buried deeply in text, with the consequence that subsequent decisions are delayed or misinformed. Bujar and colleagues (2020) surveyed twelve pharmaceutical companies and eleven regulatory agencies about documenting major decisions and found that where systems existed they were used primarily to document outcomes rather than the process, that information from documentation is not always used, and that feedback loops comparing expected with actual outcomes are not in place. Manteuffel, Avgeriou and Hamberg (2018) conducted a qualitative case study in the software-intensive systems industry and found that architects frequently reuse past decisions but are confined to decisions they already know or have heard about, reuse them in an ad hoc manner, and that improving the documentation and discoverability of decisions holds large potential. Weinreich, Groher and Miesbauer (2015) surveyed twenty-five software architects, team leads and senior developers from twenty-two companies in ten countries on the kinds, influence factors and documentation of design decisions in practice. Conklin and Begeman (1988) built and tested a hypertext tool for capturing the issues, positions and arguments behind policy decisions. Alvarado, Bañares-Alcántara and Trujillo (2005) set out an organisational memory architecture that explicitly represents decision-making rationale alongside workflow and participant profiles so that recorded experience can be reused. Roese and Vohs (2012) reviewed hindsight bias as a combination of memory distortion, changed beliefs about an event's objective likelihood and changed beliefs about one's own predictive ability, arising from cognitive, metacognitive and motivational inputs, with consequences including myopic attention to a single causal understanding and overconfidence, and noted that an intervention encouraging people to consider alternative causal explanations reduces it. Guilbault and colleagues (2004) meta-analysed research on hindsight bias. Mitchell, Russo and Pennington (1989) examined prospective hindsight, in which an explanation is generated for a future event as though it had already happened, and found that whether the event's occurrence was certain or uncertain strongly affected the type of explanation produced. Von Zedtwitz (2002) surveyed research and development projects and found that only one in five received a post-project review, that reviews were constrained by lack of time, attention, personal interest and ability, and that they focused mostly on technical output and bureaucratic measurement rather than on process factors such as project management. Schindler and Eppler (2003) reviewed project learning methods and the factors that make harvesting project knowledge succeed or fail.




The action note check


Everything above comes down to one small object: a single line in a document, describing one thing one person is going to do. So here is a tool for that line.


Put in the action as it is currently written, who owns it, how the commitment is worded, whether the owner said it out loud, what it depends on, where the note lives and what the follow-up is. It will tell you how likely the research thinks this action is to happen, where the weakness is, what to change in the wording, what to check afterwards, and it gives you a rewritten action note to copy.





Where to start


If you take one thing from all of this, make it the sentence shape. Every action gets a name, an observable act, a trigger and a date, in that order, and the person whose name it is says it out loud before they leave the room. That one change carries most of the research in this post: the if-then format that raised completion of difficult intentions roughly threefold, the identifiability that stops effort leaking away, and the cue that lets a pending intention come back on its own instead of being carried around all week.


After that, there are three moves worth making.


Split the minute from the action note and send the action note the same day, because the record needs approval and the work needs speed. Put both in one agreed place with one naming convention, because what a team knows decays quickly and scatters when somebody leaves. And add two lines to every decision, what you ruled out and when you will look at it again, because you will not be able to reconstruct either of them once you know how it turned out.


I have put the three pages I would use into a download: an action note built around the wording rule, a decision log with the two lines most logs are missing, and a follow-up rhythm you can run in fifteen minutes a week.



A graphic titled The action note, on a page, with the eyebrow One action, written the way the research would write it, with the evidence beside each line. Standfirst: This is one row of a real action note. Every field is there because something in the literature says an action without it is likelier to go missing, and the whole row takes about eleven seconds longer to write than the version most of us send. A large cream card holds six labelled rows, each with the wording in the middle and the evidence in a tinted box on the right. The decision, in the room's words: We are moving the supplier review to October rather than running it alongside the audit. Evidence: Decisions and their reasons are often unrecorded or buried in prose, and later decisions then go wrong. The owner, one name, always: Sam, by name rather than by department. Confirmed out loud in the meeting. Evidence: Effort drops when a contribution cannot be traced to one person, and the commitments that hold are the ones other people heard. The action, when, who, what, by: When the October figures land on the fifteenth, Sam will compare the top six suppliers on unit cost and send one page to the group by the twentieth. Evidence: Difficult intentions were completed about three times as often when they carried a when, a where and a how. Ruled out, and why: Running it in July, because the audit takes the same two people for three weeks. Evidence: Once you know how something turned out you cannot reconstruct what you believed before it. Revisit when, a condition, not a mood: If the audit moves, or if two of the six suppliers change their terms. Evidence: Public commitments can lock people into a course after the evidence has turned, and a named condition is the way back out. Where it lives, one place, one rule: Shared folder, filed by date and meeting, with the person to ask named at the top. Evidence: A record in one person's sent items is the same information somewhere the rest of the team cannot look. Sources at the foot: Onditi and colleagues (2004); Gollwitzer and Brandstatter (1997); Roese and Vohs (2012); Staw (1976).
Six fields, each of them there because something goes missing without it.

The best minute-takers I have worked with were never the fastest typists. They were the people who understood that a meeting produces two things, a record and a set of promises, and that the second one is fragile in ways nobody in the room can see. Getting that right is a quiet, unglamorous kind of influence, and it is one of the few places where a single well-built document changes what an organisation actually does. If you are the person who writes the note, you are holding more of the organisation's memory than your job title suggests, and it is worth holding it well.


The part of this that shows up in how you are seen at work sits inside what the Connected Leader Assessment calls Communication and Presence: whether what you write lands the way you meant it, whether the people who read it know what to do next, and whether the record you leave behind still makes sense to somebody who was not there. It is the dimension that decides whether your work travels beyond the room you were in.


The assessment is free, takes about twelve minutes and gives you a read across all five dimensions. If you want to go further on this one, the Skill Sprints are short, focused programmes on a single dimension, and the Irreplaceable Audit is the place to look at how much of your organisation's memory is currently sitting in your head rather than in its records, which is usually a larger number than anyone expects.


Stay savvy.


Meg ✌️




Free tools


  • The action note check, the interactive tool in this post, free

  • The Action Note Pages, the download in this post, free

  • The Connected Leader Assessment, twelve minutes, free, with a read on Communication and Presence, the dimension this post sits in

  • The Irreplaceable Audit, forty-eight activities across four kinds of work, free, for seeing how much of the organisation's memory is currently held in your head




P.S. Want to Go Deeper on Minutes and Action Notes?


The papers behind this piece are below, each with a line on what it found, for anyone who likes to see the working.


📃 Brown-Schmidt and colleagues (2023) Pairs who had a conversation. A week later only 4.7% of the details were recalled by both of them. https://doi.org/10.1111/cogs.13271


📃 Stafford, Burggraf and Sharkey (1987) People recalled about a tenth of a conversation straight afterwards, and about four per cent a month later. https://doi.org/10.1111/j.1468-2958.1987.tb00127.x


📃 Brown-Schmidt and colleagues (2025) 60 people, 60 conversations. Overhearers recalled less, and got who said it wrong more often. https://doi.org/10.3758/s13421-024-01630-8


📃 McKinley, Brown-Schmidt and Benjamin (2017) What was said and who said it are two different memories, and the spoken is remembered better than the heard. https://doi.org/10.3758/s13421-017-0730-3


📃 Brown-Schmidt and Benjamin (2018) A review of what can be recalled from a conversation long afterwards, and the errors to expect. https://doi.org/10.1177/2372732218786975


📃 Diachek and Brown-Schmidt (2024) 59 pairs. Hesitations and small acknowledgements made the surrounding words more likely to be recalled. https://doi.org/10.3758/s13423-023-02440-w


📃 Hope and colleagues (2019) A timeline format produced more verbatim detail, fewer sequencing errors and better speaker attribution. https://doi.org/10.1037/lhb0000329


📃 Murre and Dros (2015) The classic forgetting curve, replicated, with a jump upwards around the twenty-four hour mark. https://doi.org/10.1371/journal.pone.0120644


📃 Weldon and Bellinger (1997) Collaborative inhibition: groups recalling together produced less than the same people recalling separately. https://doi.org/10.1037/0278-7393.23.5.1160


📃 Basden and colleagues (1997) Why remembering together costs you: collaboration disrupts each person's own retrieval strategy. https://doi.org/10.1037/0278-7393.23.5.1176


📃 Rajaram and Pereira-Pasarin (2010) Review of collaborative memory, its costs, its errors, and the conditions under which it helps. https://doi.org/10.1177/1745691610388763


📃 Greeley and Rajaram (2023) Recalling together synchronises what a group remembers and the order it remembers it in. https://doi.org/10.1002/wcs.1641


📃 Roediger, Meade and Bergman (2001) The social contagion of memory: items a partner made up were later recalled as though seen. https://doi.org/10.3758/bf03196174


📃 Gabbert, Memon and Allan (2003) 71% of witnesses who discussed the event first went on to report things only the other person had seen. https://doi.org/10.1002/acp.885


📃 Gabbert and colleagues (2004) Misinformation picked up in conversation was more misleading than the same misinformation read on a page. https://doi.org/10.1348/1355325041719428


📃 Hirst and Echterhoff (2012) Review of how conversation reshapes memory, through contagion, audience tuning and induced forgetting. https://doi.org/10.1146/annurev-psych-120710-100340


📃 Cuc, Koppel and Hirst (2007) Saying part of a shared memory aloud made the listener forget the related parts that went unmentioned. https://doi.org/10.1111/j.1467-9280.2007.01967.x


📃 Coman, Manier and Hirst (2009) The same selective-recounting effect on well-rehearsed, emotionally intense shared memories. https://doi.org/10.1111/j.1467-9280.2009.02343.x


📃 Marsh and Tversky (2004) People labelled 61% of their own retellings as distorted and only 42% as inaccurate. https://doi.org/10.1002/acp.1001


📃 Harris, Paterson and Kemp (2008) A review drawing together what happens when people remember in groups. https://doi.org/10.1080/09658210701811862


📃 Bourgeois, Salapska-Gelleri and McElroy (2026) Group discussion raised how much was remembered and also raised misattribution and persistence. https://doi.org/10.1080/09658211.2026.2615982


📃 Miller (1996) In problem-solving conversations, people remembered their own contributions better than their partner's. https://doi.org/10.1080/741940999


📃 Wolfe (2006) Textbooks teach a parliamentary style of minutes; workplaces run on an action-oriented one. https://doi.org/10.1109/tpc.2006.885837


📃 Schwartz-Ziv and Weisbach (2013) A study that used the actual minutes of board and committee meetings to reconstruct what boards do. https://doi.org/10.1016/j.jfineco.2012.04.011


📃 Bezemer, Nicholson and Pugliese (2014) Videotaped board meetings. Discussion varies enormously between agenda items inside one meeting. https://doi.org/10.1108/qram-02-2013-0005


📃 Pugliese, Nicholson and Bezemer (2015) Directors' inclusiveness and evenness of participation went with higher perceived board effectiveness. https://doi.org/10.1111/1467-8551.12074


📃 Forbes and Milliken (1999) Boards of directors understood as strategic decision-making groups, and what makes them work. https://doi.org/10.2307/259138


📃 Smith (2013) The four purposes of formal minutes: communication, legal record, management instrument, history. https://doi.org/10.1097/01.numa.0000427898.70872.21


📃 Volkema and Niederman (1995) 35 organisations. Whether agendas and supporting documents were used at all varied by meeting format. https://doi.org/10.1177/1046496495261001


📃 Gollwitzer (1999) The implementation intention: if situation Y arises, then I will do X. https://doi.org/10.1037/0003-066x.54.7.493


📃 Gollwitzer and Sheeran (2006) 94 tests. If-then plans had a medium-to-large effect on goal attainment, d = .65. https://doi.org/10.1016/s0065-2601(06)38002-1


📃 Sheeran, Listrom and Gollwitzer (2025) 642 tests. Effects were larger for if-then format, high motivation and rehearsed plans. https://doi.org/10.1080/10463283.2024.2334563


📃 Gollwitzer and Brandstätter (1997) Difficult intentions were completed about three times as often when given a when, where and how. https://doi.org/10.1037/0022-3514.73.1.186


📃 Thürmer, Wieber and Gollwitzer (2017) Group plans phrased as we rather than I worked, and worked through cooperation. https://doi.org/10.3389/fpsyg.2017.00603


📃 Allan, Sniehotta and Johnston (2013) Plans are only as good as the planner, and a supplied plan helped the poor planners most. https://doi.org/10.1007/s12160-013-9483-9


📃 Dalton and Spiller (2012) Planning one goal helps; planning many at once undermines commitment to all of them. https://doi.org/10.1086/664500


📃 Trötschel and Gollwitzer (2007) If-then plans used to hold to a prosocial goal under the pressure of a negotiation. https://doi.org/10.1016/j.jesp.2006.06.002


📃 Nickerson and Rogers (2010) 287,228 people. Making a voting plan raised turnout 4.1 points; encouragement raised nothing. https://doi.org/10.1177/0956797609359326


📃 Lehmann-Willenbrock, Allen and Kauffeld (2013) 59 meetings in 19 organisations. Procedural talk produced more who-will-do-what-and-when statements. https://doi.org/10.1080/00909882.2013.844847


📃 Sniehotta, Scholz and Schwarzer (2006) 211 cardiac patients. An action plan plus a plan for the obstacle beat the action plan alone. https://doi.org/10.1348/135910705x43804


📃 Hagger and Luszczynska (2014) What makes a planning intervention work: if-then format, salient cues, guided plans and boosters. https://doi.org/10.1111/aphw.12017


📃 Bélanger-Gravel, Godin and Amireault (2013) Meta-analysis of implementation intentions applied to physical activity. https://doi.org/10.1080/17437199.2011.560095


📃 Lerner and Tetlock (1999) Accountability is not a general improver; what it does depends on the ground rules. https://doi.org/10.1037/0033-2909.125.2.255


📃 Siegel-Jacobs and Yates (1996) Being accountable for your process and being accountable for the outcome do different things to judgement. https://doi.org/10.1006/obhd.1996.0001


📃 Lokhorst and colleagues (2013) Meta-analysis. Making a commitment changed behaviour in the short and the long term against controls. https://doi.org/10.1177/0013916511411477


📃 Cialdini and Goldstein (2004) Review of compliance and conformity, including the pull of staying consistent with what you said. https://doi.org/10.1146/annurev.psych.55.090902.142015


📃 Deutsch and Gerard (1955) The founding distinction between normative and informational influence, and the weight of a public answer. https://doi.org/10.1037/h0046408


📃 Bryan, Karlan and Nelson (2010) A review of commitment devices, hard and soft, and when people ask for them. https://doi.org/10.1146/annurev.economics.102308.124324


📃 Giné, Karlan and Zinman (2010) A voluntary commitment account for smokers. The effect was still there in surprise tests at a year. https://doi.org/10.1257/app.2.4.213


📃 Royer, Stehr and Sydnor (2015) Gym incentives faded; adding a commitment contract produced changes visible years later. https://doi.org/10.1257/app.20130327


📃 Karau and Williams (1993) Meta-analysis of social loafing. Making individual contributions identifiable is one of the cures. https://doi.org/10.1037/0022-3514.65.4.681


📃 Latané, Williams and Harkins (1979) Many hands make light the work: effort drops when nobody can tell who did what. https://doi.org/10.1037/0022-3514.37.6.822


📃 Darley and Latané (1968) Diffusion of responsibility. The more people who could act, the less likely any one of them did. https://doi.org/10.1037/h0025589


📃 Staw (1976) Knee-deep in the big muddy: people put more in after the evidence turned, especially if they chose it. https://doi.org/10.1016/0030-5073(76)90005-2


📃 Sleesman and colleagues (2012) Meta-analysis of what makes escalation of commitment worse. https://doi.org/10.5465/amj.2010.0696


📃 Pallak and Cummings (1976) Homeowners who expected to be named publicly used less gas and electricity than those who did not. https://doi.org/10.1177/014616727600200105


📃 Mueller and Oppenheimer (2014) Three studies. Laptop note takers transcribed more verbatim and did worse on conceptual questions. https://doi.org/10.1177/0956797614524581


📃 Morehead, Dunlosky and Rawson (2019) A replication and extension that did not reproduce the longhand advantage. https://doi.org/10.1007/s10648-019-09468-2


📃 Urry and colleagues (2021) A direct replication. Same difference in the notes, no difference in the quiz; more words went with better scores. https://doi.org/10.1177/0956797620965541


📃 Kobayashi (2005) Meta-analysis of the encoding effect of note-taking, and what limits it. https://doi.org/10.1016/j.cedpsych.2004.10.001


📃 Luo, Kiewra and Samuelson (2016) Revision, pauses and a partner all improved the notes people took and what they got from them. https://doi.org/10.1007/s11251-016-9370-4


📃 Peverly and colleagues (2007) How fast you can get words down is one of the strongest predictors of good lecture notes. https://doi.org/10.1037/0022-0663.99.1.167


📃 Bohay and colleagues (2011) Note-taking improved deeper understanding, tested immediately and a week later. https://doi.org/10.5406/amerjpsyc.124.1.0063


📃 Piolat, Olive and Kellogg (2005) Taking notes costs more cognitive effort than reading or learning the same material. https://doi.org/10.1002/acp.1086


📃 Kiewra and colleagues (1991) The two jobs notes do: encoding, which helps the writer, and external storage, which helps everyone else. https://doi.org/10.1037//0022-0663.83.2.240


📃 Asthana and colleagues (2025) Two machine-generated recap designs tested in real work meetings, and what users added, edited and deleted. https://doi.org/10.1145/3711074


📃 Whittaker, Laban and Tucker (2006) How people actually use meeting records, and what that means for the systems built to hold them. https://doi.org/10.1007/11677482_9


📃 Whittaker and colleagues (2008) People return to a meeting record for narrow, specific things rather than reading it through. https://doi.org/10.1007/s00779-007-0146-3


📃 Liang, Moreland and Argote (1995) Groups trained together built a shared sense of who knew what, and built better radios a week later. https://doi.org/10.1177/0146167295214009


📃 Moreland and Myaskovsky (2000) A test of whether the group-training advantage was transactive memory or merely better communication. https://doi.org/10.1006/obhd.2000.2891


📃 Ren and Argote (2011) Twenty-five years of transactive memory research, gathered into one framework. https://doi.org/10.5465/19416520.2011.590300


📃 Olivera (2000) A multinational consultancy ran several memory systems at once, each with its own failure mode. https://doi.org/10.1111/1467-6486.00205


📃 Walsh and Ungson (1991) Organisational memory: where an organisation's knowledge is actually stored, and how it is retrieved. https://doi.org/10.2307/258607


📃 Stein (1995) A review of organisational memory and the processes of acquiring, keeping and retrieving it. https://doi.org/10.1016/0268-4012(94)00003-c


📃 Argote, Beckman and Epple (1990) Knowledge gained by doing the work depreciates rapidly, and cumulative experience overstates what is left. https://doi.org/10.1287/mnsc.36.2.140


📃 Darr, Argote and Epple (1995) 36 pizza stores over eighteen months. Knowledge moved between stores with the same owner, not between owners. https://doi.org/10.1287/mnsc.41.11.1750


📃 Foroughi and colleagues (2020) A review of organisational memory studies, treating forgetting as something organisations do. https://doi.org/10.1177/0170840620974338


📃 Peltokorpi (2012) Transactive memory scaled up: overlapping networks of groups using each other as external memory. https://doi.org/10.1027/1016-9040/a000044


📃 Altmann and Trafton (2002) Pending goals sit in memory at an activation level that decays, and cues are what bring them back. https://doi.org/10.1207/s15516709cog2601_2


📃 McDaniel and Einstein (2000) Remembering to do something later runs either on effortful monitoring or on an anticipated cue. https://doi.org/10.1002/acp.775


📃 Calzolari and Nardotto (2017) Weekly reminders raised gym attendance, worked within hours, and kept working for each new reminder. https://doi.org/10.1287/mnsc.2016.2499


📃 Gurol-Urganci and colleagues (2013) Cochrane review of mobile phone reminders for healthcare appointments. https://doi.org/10.1002/14651858.cd007458.pub3


📃 Damgaard and Gravert (2018) Reminders raised donations and substantially raised unsubscribes. Reminding is not free. https://doi.org/10.1016/j.jpubeco.2017.11.005


📃 Mertens and colleagues (2022) Meta-analysis concluding that choice architecture interventions promote behaviour change across domains. https://doi.org/10.1073/pnas.2107346118


📃 Maier and colleagues (2022) The reply. Corrected for publication bias, the effect of nudges could be as low as d = 0.08. https://doi.org/10.1073/pnas.2200300119


📃 Altmann, Trafton and Hambrick (2014) Interruptions averaging 4.4 seconds tripled the rate of sequence errors afterwards. https://doi.org/10.1037/a0030986


📃 Onditi and colleagues (2004) Decisions and their reasons are often unrecorded or buried in prose, so later decisions go wrong. https://doi.org/10.1007/978-3-540-27779-8_13


📃 Manteuffel, Avgeriou and Hamberg (2018) Practitioners reuse past decisions, and are confined to the ones they already happen to know about. https://doi.org/10.1016/j.jss.2018.05.064


📃 Weinreich, Groher and Miesbauer (2015) A survey of twenty-five architects in twenty-two companies on what decisions get written down. https://doi.org/10.1016/j.future.2014.12.002


📃 Bujar and colleagues (2020) Where decision records exist they capture outcomes rather than process, and are often never read again. https://doi.org/10.1007/s43441-020-00167-7


📃 Roese and Vohs (2012) Hindsight bias, what it is made of, and the one intervention that reliably reduces it. https://doi.org/10.1177/1745691612454303


📃 Guilbault and colleagues (2004) Meta-analysis of research on hindsight bias. https://doi.org/10.1207/s15324834basp2602&3_1


📃 Mitchell, Russo and Pennington (1989) Prospective hindsight: explaining a future event as though it had already happened. https://doi.org/10.1002/bdm.3960020103


📃 Von Zedtwitz (2002) Only one in five research and development projects got a post-project review at all. https://doi.org/10.1111/1467-9310.00258


📃 Schindler and Eppler (2003) A review of project learning methods and what makes harvesting project knowledge succeed. https://doi.org/10.1016/s0263-7863(02)00096-0


📃 Conklin and Begeman (1988) An early tool for capturing the issues, positions and arguments behind a decision. https://doi.org/10.1145/62266.62278


📃 Alvarado, Bañares-Alcántara and Trujillo (2005) An organisational memory that records the decision, the reasoning and who was in the room. https://doi.org/10.1016/j.petrol.2004.11.009




P.P.S. Why Do I Even Have the Nerve to Write This?


Fair question.


I am someone who has spent years working alongside remarkable executives, navigating chaos, translating vision into structure and figuring things out, sometimes beautifully and sometimes the very hard way. I have had the privilege of being the steady right hand to leaders who move fast, take risks and expect a lot. That environment has shaped me more than any classroom ever could.


I also happen to have an MBA, and I am studying psychology because people fascinate me. How we work. Why we disconnect. What actually holds us together when the pressure rises.


But really, none of that is the point.


I am here because I have lived the erosion and rebuilt from it. I have seen the cost of carrying too much and the relief that arrives when you finally stop disappearing inside your own competence.


So I share what I have learned in case it helps someone else. Take what is useful. Leave the rest. And remember this one truth that professionals like you often forget:


You are probably doing far better than you give yourself credit for.


Comments


Commenting on this post isn't available anymore. Contact the site owner for more info.
Post: Blog2_Post
bottom of page