Research overview

Better Meetings

A meeting's purpose should shape everything about it, yet in our studies many meetings ran without one stated. Help with meeting goals should move people from wherever they start, with depth that fits the meeting.

This page sets out the evidence from nineteen papers published from 2020 to 2026. Eight take up meeting purpose directly, and two more, from Better Thinking With AI, test interfaces that AI generates. Nine papers on video and hybrid meetings, most of them earlier, led to them. The background tells how my colleagues and I came to the question.

› Background

Meetings aren't the problem. Meetings nobody can explain are.

Much of my research has been about how people meet through screens. Calls, chat, hybrid rooms, avatars, and robots all kept running into the same question. Why are we meeting? This is the story of how my colleagues and I came to that question and followed it, from 2020 to 2026.

We started where meeting research in HCI usually starts, with the technology. Our early studies asked how to make remote and hybrid meetings more engaging and inclusive. We classified what attention in video meetings is for, from where people look to whether they are engaged. Interviewees described low engagement as a deliberate practice, and we mapped email multitasking across tens of thousands of meeting attendances. We studied how meeting chat lets people contribute without interrupting, and how it can distract. We argued for meeting-centred design, which treats the whole meeting, with its people, technology, and setting, as the thing to configure.

COVID showed that the problem was bigger than the technology. When the pandemic moved work onto video, our diary study suggested meetings had become containers for almost any kind of contact, social or task. In a survey that year, failing to stick to the agenda topped the complaints that were not about the technology. Email multitasking was more common in long and recurring meetings. Even our chat study ended by arguing for a more intentional stance, so that organisational purpose is more accountable in our meeting tools. Better audio and video could help people feel present, but they could not supply a reason to be there.

So we pivoted to purpose. Meeting intentionality is knowing why a meeting is occurring and focusing activity on that purpose. A meeting's purpose is its design brief. It should decide whether to meet at all, who comes, how the meeting runs, which tools and features it needs, and what carries on afterwards. As far as I know, this is one of the most sustained research programs on meeting goals. It is also one of the few to follow the whole meeting lifecycle, from deciding to meet to what carries into the next meeting.

Many meetings run without that brief. When we asked people what their meetings were for, many named topics, not outcomes. If goals are the destination, agendas are only the route. The people we interviewed held two mental models of a meeting, as a means to an end or as an end in itself, for connection or discovery. Both are legitimate, and we argued that for some meetings goals are best left implicit, as long as people reflect on why they meet. In our recordings, only 4 of 12 workplace meetings said what they were meant to achieve.

Then we tried to help people change, in two ways. The first was the simpler approach, the same light nudge for everyone. For two weeks, 361 people took part in a randomised field experiment. A control group rated how effective each of their meetings had been, just after it ended. A treatment group did the same, and also got a chat survey five minutes before their meetings, asking what success would look like. By their own account, both groups got clearer about goals and better at communicating them. Of those who left comments, 57% in one group and 60% in the other described a change in awareness or behaviour. Some organisers said they put goals in invitations and opened meetings with them, and some people said they declined or moved meetings. We think two weeks of repeatedly focusing on meeting goals may have worked as micro-training. The extra reflection before meetings added no measurable gain in rated meeting effectiveness.

The second was new, an AI coach. The Meeting Purpose Assistant talked 18 people through their own upcoming meetings, in sessions guided by a researcher. It asked why each meeting was happening, what success would look like, and what could get in the way, and it did not volunteer answers. Several people said putting their goals into words helped them focus and arrive with a plan. Some shared the summary with attendees, and three later said that asking attendees to add to the agenda improved attendance and engagement. Participants saw the coach as best for important or uncertain meetings, and we suggest a quicker version for routine ones.

What we look for is movement, not a pass mark. Much meeting research sorts meetings into types, or looks for one simple measure of whether a meeting worked. Those taxonomies are useful, and our field experiment drew its list of goals from them. But meetings are too varied for one bar. The question is whether people get clearer, communicate more, and make better choices, from wherever each person and each meeting starts. The field experiment's surveys reached 361 people. Of those who left comments, 57% in one group and 60% in the other described some change in awareness or behaviour. The AI coach went deeper with 18 volunteers, who described clarifying goals, preparing, and rethinking recurring meetings.

Along the way, we learned what people wanted from AI. In one study, 26 employees were guided through a simulated meeting with CoExplorer. This prototype used generative AI to draft a meeting's goal and phases from the invitation, poll attendees to find where they disagreed, and open files and apps as talk changed phase. 24 of 26 found it helpful to gather attendees' views in advance. But 14 wanted the organiser to have the final say on that input, and 21 wanted the system to ask before showing files or apps.

In another study, 15 employees replayed recordings of their own meetings alongside two AI prototypes. One asked them questions about the meeting's goals as it went, which we call active reflection. The other quietly displayed the goals and topics the AI had inferred, a passive display. With the questions, 13 of 15 began by asking what the meeting was trying to achieve. With the passive display, 10 of 15 started from the AI's version of the goals and built on it, and 4 named goals the AI had missed. The display helped people keep focus but was easy to overlook, and the questions led people to say they would act but could interrupt. So we suggest support that stays in the background until a meeting drifts, then steps up.

Purpose is social. Organisers and attendees wait on each other. In our field experiment, 21% of participants said that as attendees, meeting goals were not their responsibility. Answering as organisers, 27% doubted attendees would read or understand a goal. Asking why can feel risky, and voice is uneven too, since in our telemetry study women chatted more and men spoke more, and junior staff who felt psychologically safe sent more chats. We think one person's reflection can break that loop. AI could also act as a neutral party that lets anyone raise the question.

And purpose has to outlast the meeting. In eight weekly meetings of two research teams, people looked back and ahead throughout, and we argue those links are often unnoticed and soon lost. We designed recaps and handoffs to pass purpose from one meeting to the next, a chain of intentionality, but have not yet tested them with users. The next step, which Lev Tankelevitch and I have argued for, is to make goals first-class objects that people and AI agents share. Our design for a workspace that remembers how work happened lets AI turn a meeting's talk and edits into tools the team builds as it works. That is where better meetings meet better thinking with AI.

I lead this research program, built with Lev Tankelevitch, Ava Elizabeth Scott, Payod Panda, Rishi Vanukuru, Xinyue Chen, Gun Woo Park and Nagaravind Challakere, and with the colleagues named on each study.

How a meeting's purpose gets lost, or kept

What the studies show together

We used interviews, prototypes people tried, an AI coach used before real meetings, recordings, telemetry, surveys, diaries, a preregistered field experiment, a lab study, an online experiment, a system demonstration, and a position paper. Nearly all the studies with participants were at one global technology company. Each study has a research brief with its findings, quotes, and figures, checked against the paper.

Strongest evidence Also points this way Complicates it

ClaimInterviewsmental models of meeting goalsGuided walkthroughAI that plans and runs meetingsParticipatory promptingAI-assisted prospective reflectionProbe studyAI goal reflection during meetingsRecordings and designlooking back and ahead in meetingsField experimentnudging attention to meeting goalsPosition papergoals in human-AI workSystem designAI-mediated shared workspacesConversation analysismeeting-centred designInterviews and narrationwhat attention in video is forInterviewslow engagement as a choiceResearch synthesiscollaboration and meetingsTelemetry and diariesmultitasking in remote meetingsSurvey and dashboard studyfeedback on inclusive meetingsSurvey and diarieschat alongside video meetingsDiary studyvideo meetings in the pandemicSurvey and telemetrychat and voice in meetingsLab studyprompt refinement controlsRandomised experimentscaffolds for AI-assisted writing
Meetings often run on purposes nobody has statedInterviews: mental models of meeting goals, strongest evidenceParticipatory prompting: AI-assisted prospective reflection, also points this wayProbe study: AI goal reflection during meetings, strongest evidenceTelemetry and diaries: multitasking in remote meetings, also points this waySurvey and dashboard study: feedback on inclusive meetings, also points this wayDiary study: video meetings in the pandemic, also points this way
AI can help people work out what a meeting is for and keep it in viewGuided walkthrough: AI that plans and runs meetings, also points this wayParticipatory prompting: AI-assisted prospective reflection, strongest evidenceProbe study: AI goal reflection during meetings, strongest evidenceField experiment: nudging attention to meeting goals, also points this wayPosition paper: goals in human-AI work, also points this wayDiary study: video meetings in the pandemic, also points this way
Help with meeting goals should start where people are, and go deeper only for meetings that need itInterviews: mental models of meeting goals, also points this wayGuided walkthrough: AI that plans and runs meetings, also points this wayParticipatory prompting: AI-assisted prospective reflection, strongest evidenceProbe study: AI goal reflection during meetings, strongest evidenceField experiment: nudging attention to meeting goals, strongest evidenceResearch synthesis: collaboration and meetings, complicates itSurvey and dashboard study: feedback on inclusive meetings, also points this wayDiary study: video meetings in the pandemic, complicates it
A meeting's purpose is everyone's job, but asking what it is can feel risky, especially for junior staffInterviews: mental models of meeting goals, strongest evidenceGuided walkthrough: AI that plans and runs meetings, complicates itParticipatory prompting: AI-assisted prospective reflection, also points this wayProbe study: AI goal reflection during meetings, strongest evidenceField experiment: nudging attention to meeting goals, strongest evidenceConversation analysis: meeting-centred design, also points this wayInterviews and narration: what attention in video is for, also points this wayInterviews: low engagement as a choice, also points this wayTelemetry and diaries: multitasking in remote meetings, also points this waySurvey and diaries: chat alongside video meetings, also points this waySurvey and telemetry: chat and voice in meetings, also points this way
A meeting's purpose is easily lost between meetings unless something carries it forwardInterviews: mental models of meeting goals, also points this wayGuided walkthrough: AI that plans and runs meetings, also points this wayParticipatory prompting: AI-assisted prospective reflection, strongest evidenceProbe study: AI goal reflection during meetings, also points this wayRecordings and design: looking back and ahead in meetings, also points this wayField experiment: nudging attention to meeting goals, also points this wayPosition paper: goals in human-AI work, also points this wayResearch synthesis: collaboration and meetings, also points this waySurvey and dashboard study: feedback on inclusive meetings, also points this waySurvey and diaries: chat alongside video meetings, also points this way
AI can generate interfaces around what people intend, as long as people stay in controlGuided walkthrough: AI that plans and runs meetings, strongest evidenceRecordings and design: looking back and ahead in meetings, also points this waySystem design: AI-mediated shared workspaces, also points this wayLab study: prompt refinement controls, strongest evidenceRandomised experiment: scaffolds for AI-assisted writing, strongest evidence

Meetings often run on purposes nobody has stated

The employees we interviewed treated a meeting as a means to an end or as the point in itself, and many let topic lists stand in for goals. Only 4 of 12 recorded workplace meetings stated what they should achieve, and only 1 referred back to its goals. An earlier survey had already flagged meetings failing to stick to their agendas, telemetry linked email multitasking to longer and recurring meetings, and diaries pointed to attention drifting from meetings that felt irrelevant.

Strongest evidence

  • Interviews: mental models of meeting goals. In interviews with 21 employees, we identified two mental models of meeting goals, meetings as a means to an end or as ends in themselves. Many interviewees treated a list of discussion topics as the goal.
  • Probe study: AI goal reflection during meetings. Only 4 of 12 recorded workplace meetings stated what they should achieve, though 8 set out their topics at the start, and only 1 referred back to its goals.

Also points this way

AI can help people work out what a meeting is for and keep it in view

Our AI studies explore how AI can support meeting intentionality, by helping people work out what a meeting is for and keep it in view. Before real upcoming meetings, 18 employees talked with our Meeting Purpose Assistant, which asked why each meeting was needed, what success would look like, and what could get in the way. Several likened it to a coach that made them think for themselves, though solution-oriented people wanted it to offer recommendations too. When 15 employees watched their own recorded meetings with AI that asked about goals, 13 began by asking what the meeting was trying to achieve, though some found the questions intrusive. When AI drafted a goal and phases for a simulated meeting, 24 of 26 people valued gathering attendees' views beforehand, so that discussion could focus where they disagreed. Discussing our field experiment, we argue that people stating their own goals remains a prerequisite, because current AI is limited at inferring intent.

Strongest evidence

  • Probe study: AI goal reflection during meetings. With 15 employees, we explored two AI designs for reflecting on goals during meetings, a passive display of topics and goals and active questions, using recordings of their own meetings. Participants said clear goals helped them see when a meeting was off track. The display helped them keep focus without intruding, and the questions prompted them to say they would act, though the questions risked disrupting the conversation.
  • Participatory prompting: AI-assisted prospective reflection. With a researcher guiding each session, the AI coach asked questions and offered suggestions only when asked. Five of 18 people likened it to a coach or therapist that made them think for themselves.

Also points this way

Help with meeting goals should start where people are, and go deeper only for meetings that need it

Two generations of help with meeting goals sit at opposite ends of a scale. People who rated their meetings for two weeks reported clearer goals, with or without the extra checklist. Of those who commented, 57% with the checklist and 60% without described changes in awareness or behaviour. An AI coach before chosen meetings went deeper with fewer people. A goal checklist before each meeting added no measurable gain on top, and those who got it were less keen on future prompts. So match depth to stakes, ask early, and step in during meetings only when they drift.

Simpler: the same light check for everyone

Nudging Attention to Workplace Meeting Goals. 361 people, 7,196 meetings, two weeks.

Everyone rated their meetings afterwards for two weeks. A randomly assigned group also got a chat message five minutes before each meeting, asking what success would look like, with a list of goals to tick.

Both groups reported clearer, better-communicated goals, and of those who commented, 57% with the checklist and 60% without described a change in awareness or behaviour. Some said they put goals in invitations or declined meetings. The extra checklist added no measurable gain on top, and 55% of checklists were answered more than 15 minutes after they were sent.

New: an AI coach for chosen meetings

What Does Success Look Like? 18 people, up to three real upcoming meetings each.

For about 10 minutes per meeting, with a researcher guiding the session, the Meeting Purpose Assistant asked why the meeting was happening, what success would look like, and what could get in the way. It then wrote a short summary under those three questions.

Afterwards, people said naming the goal helped them focus and prepare. Some shared the summary with attendees, and three said that asking attendees to add to the agenda improved attendance and engagement. It is deeper and slower, and participants saw it as best for important or uncertain meetings.

“I sent [the summary] to the attendees, and they all brought topics of discussion, which made it very effective.”

A participant, after the meeting

Strongest evidence

  • Participatory prompting: AI-assisted prospective reflection. Reflecting with an AI coach on real upcoming meetings, people said it sharpened their focus and preparation, and saw important or uncertain meetings as the best fit. One participant said attendees cannot be told about a new agenda item 10 minutes before a meeting.
  • Field experiment: nudging attention to meeting goals. Over two weeks of rating their meetings, people with and without a pre-meeting checklist reported clearer goals on average and said they communicated them more often. The checklist added no detectable gain on top, and those who got it were less keen on future prompts.
  • Probe study: AI goal reflection during meetings. Replaying their own meetings, people said a passive goal display helped them keep focus, though 6 of 15 found it easy to overlook. Pop-up questions led people to say they would act, but 4 of 15 found them intrusive. People wanted support that steps up only when it is needed.

Also points this way

Complicates it

  • Diary study: video meetings in the pandemic. In diaries, some people said that as meetings became more intentional, with less room for casual talk, they felt more productive but less connected.
  • Research synthesis: collaboration and meetings. In our pandemic study of remote meetings, which the chapter reports, a small majority said their meeting preparation changed, and most of them felt better prepared. Opinions split on whether this helped others, and respondents agreed it had not reduced meetings.

A meeting's purpose is everyone's job, but asking what it is can feel risky, especially for junior staff

Organisers and attendees each wait on the other to raise a meeting's purpose. Asking can feel risky in new or unequal relationships, so people ration attention and voice by status and relevance instead. Some people saw AI as a neutral party, but individual contributors feared being flagged in public.

Strongest evidence

  • Field experiment: nudging attention to meeting goals. Some attendees felt unable to change meetings they did not organise. Answering as attendees, 21% of participants said thinking about goals was not their responsibility, and answering as organisers, 27% said attendees would not read or understand them.
  • Interviews: mental models of meeting goals. All 21 interviewees said they had asked an organiser what a meeting was for, yet five described avoiding it, more so in ambiguous or unequal relationships. Two who served customers kept goals out of invitations to prevent resistance.
  • Probe study: AI goal reflection during meetings. Junior staff and people not running a meeting noticed it drifting but held back. Of 15 participants, 9 saw AI goal reflection as sharing the job of keeping a meeting on track, and 5 valued it as a neutral mediator. Of 11 individual contributors, 6 feared the AI would call them out in front of the team for straying off topic.

Also points this way

  • Participatory prompting: AI-assisted prospective reflection. Sharing the coach's notes on what could go wrong in a meeting could seem condescending or distrustful, and attendees often felt they could not change meetings they did not run.
  • Conversation analysis: meeting-centred design. Our analysis of three moments of technical trouble in one hybrid meeting showed the manager's authority shaping when remote colleagues were included or excluded to keep the meeting going. We proposed configuring the meeting itself, with its rules and expectations visible to everyone beforehand.
  • Survey and telemetry: chat and voice in meetings. Linking a survey of 291 employees to their Teams telemetry, we found women chatted more and men spoke more, with men at principal level or above speaking most. Junior staff who felt psychologically safe sent more chats than those who did not.
  • Interviews and narration: what attention in video is for. Interviewees said joining remotely made it hard to read others' reactions, judge when to speak, and work out who was important.
  • Survey and diaries: chat alongside video meetings. In our poll and diaries, participants said chat let them contribute without interrupting and gave a voice to those who could not get the floor. They also said chat that drifted from the spoken conversation could distract listeners and the speaker.
  • Interviews: low engagement as a choice. Interviewees described joining remotely to signal low interest in a meeting, and reading colleagues' remote attendance as a sign that it was unimportant.
  • Telemetry and diaries: multitasking in remote meetings. In 100 sampled diary responses, 17 linked multitasking to meetings that seemed irrelevant, and 32% linked it to muting or turning off video, which some said kept it unnoticed.

Complicates it

A meeting's purpose is easily lost between meetings unless something carries it forward

Three people who used our AI coach said that asking attendees to add to the agenda improved attendance and engagement, which suggests one person's reflection can spread. Across eight weekly meetings, two research teams looked back and ahead throughout, and we argue those links are soon lost. Survey respondents and interviewees described follow-up scattered across tools. Recaps that feed the next meeting, and goals that people and AI agents share, could carry purpose forward, though none has been tested yet.

Strongest evidence

Also points this way

AI can generate interfaces around what people intend, as long as people stay in control

Several of our systems explore whether generative AI can build interface elements on the fly around what people are trying to do. In a guided walkthrough of a simulated meeting, CoExplorer drafted a meeting's phases from its invitation and opened files and apps as talk moved on, yet 21 of 26 people wanted it to ask first. In a lab study, 16 knowledge workers rated controls generated for each prompt more effective for steering than fixed ones, though predicting what a control would do stayed hard. In an experiment with 297 workers, AI-generated controls for setting goals and breaking down tasks improved their briefs, though people found the AI less helpful and the work more demanding. We argue that generated interfaces are most useful as a way to give people control over AI and what it makes. Our designs for a workspace that turns a meeting's activity into tools, and for generated recaps and handoffs, take the idea further but have not been tested with users.

Strongest evidence

  • Guided walkthrough: AI that plans and runs meetings. In a guided walkthrough of a simulated meeting, 26 employees saw AI generate a meeting's phases, window layouts, files, and apps from its invitation, and 21 of 26 wanted it to ask before showing any file or app.
  • Lab study: prompt refinement controls. We built dynamic prompt middleware, which generates refinement controls for each prompt. In our lab study, 16 knowledge workers rated them more effective for steering than fixed controls, though how the AI would apply an option was often unclear.
  • Randomised experiment: scaffolds for AI-assisted writing. In our experiment with 297 workers, AI-generated controls for goal specification and task decomposition improved their briefs, with higher felt load and lower rated helpfulness, a trade-off we call productive friction.

Also points this way

Questions by audience

For researchWhat we still need to study

What did our field experiment on meeting goals show, if a pre-meeting prompt added nothing measurable?

Over two weeks, people with and without the prompt reported clearer goals and better communication. Of those who commented, 57% with the prompt and 60% without described changes in awareness or behaviour. Adding a pre-meeting prompt brought no detectable gain in rated effectiveness, p = 0.36, though a small effect is not ruled out. With no group that went without surveys, we cannot tell whether rating meetings, the study itself, or time explains the change.

From Nudging Attention to Workplace Meeting Goals

When and how often should goal reflection be prompted?

Our studies have not settled the timing, though five minutes before a meeting was too late for many participants, and some said prompts for every meeting were tiring. People using the AI coach saw important, uncertain meetings as the best candidates. Replaying their own meetings with AI, people found a passive display easy to overlook and pop-up questions sometimes intrusive. Studies should compare timing, frequency, and depth directly.

From Nudging Attention to Workplace Meeting Goals, What Does Success Look Like?, Are We On Track?

How can meeting intentionality be measured beyond self-report?

Our outcomes were mostly self-reported, and the field experiment suggests measuring meetings may itself prompt reflection. Measures should track change for each person and meeting, not one bar for all, because meetings vary so much. Behavioural signs could include goals appearing in invitations or talk returning to the goal. Our coding of recorded meetings, for stating and revisiting goals and for looking back and ahead, offers starting points.

From Nudging Attention to Workplace Meeting Goals, Are We On Track?, Designing Interfaces that Support Temporal Work Across Meetings with Generative AI

For industryResearch-informed practice and what to design next

Should a meeting tool ask organisers for a goal when scheduling?

Our studies suggest it should, with some care, because goal reflection is worth encouraging. In our field experiment, people who rated their meetings for two weeks reported clearer goals, and over two-thirds said they were interested in a future pre-meeting goal prompt. Our survey in chat five minutes before each meeting came too late for many participants to act on, and many pointed to scheduling as the right moment. Those who had that survey were less keen on more, which may reflect fatigue from answering twice as many surveys. Interviewees asked to imagine a goals field in invitations thought it would help, though three worried it would add pressure, be ignored, or be hard to update. Ask at scheduling, target meetings that need it, and let organisers skip it.

From Mental Models of Meeting Goals, Nudging Attention to Workplace Meeting Goals, What Does Success Look Like?

How much should an AI meeting assistant act on its own?

Our studies suggest it should mostly act with people's approval. In our CoExplorer study, 21 of 26 people said the system should ask before showing files or apps, yet they also wanted fewer prompts during meetings. Move approval checks to quieter moments, such as before the meeting. Let organisers keep the final say, and explain why the AI suggests a change.

From The CoExplorer Technology Probe, Are We On Track?

What should AI do with meeting goals once they are stated?

We suggest that AI carry stated goals forward into invitations, pre-reads, facilitation during the meeting, and recaps. Three people who used our AI coach said that asking attendees to add to the agenda improved attendance and engagement. Our design work proposes recaps when a meeting ends and handoffs to the next one. These designs have not been tested with users, so pilot them before any rollout.

From Nudging Attention to Workplace Meeting Goals, What Does Success Look Like?, Designing Interfaces that Support Temporal Work Across Meetings with Generative AI

What should AI agents know about the goals behind the work?

We propose that goals become persistent objects shared by people and agents, carried across applications and shown during the work. Agents could then report progress against the goals they were given, as well as the tasks they completed. In our review of current goal tools, a handful offered explicit goal objects and status tracking. None supported goal articulation, the reflective process of working out what you are actually trying to achieve. These are proposals from a position paper, not tested designs.

From Goals as First-Class Abstractions in Human-AI Collaboration

For governmentHow this research relates to policy

What happens to meeting goals and AI summaries as government records?

In many governments, AI-generated meeting content counts as an official record, and may be open to freedom of information requests. So choices about what a meeting tool keeps matter. Stated goals and recaps of decisions can make meetings more accountable. Personal reflection is different, because staff in our studies feared AI flagging them as off-topic, and asking why a meeting is needed can feel risky. Our work also warns that reflections exposed to employers could deepen power gaps. We suggest keeping private reflection separate from shared meeting records. None of our studies were done in government.

From Are We On Track?, Mental Models of Meeting Goals, What Does Success Look Like?

Agencies often use AI to summarise meetings. What should a summary record?

Summarising meetings is among the most common uses in government trials of AI assistants. A summary records what was said, not whether the meeting did what it was for. We argue that recaps should report against the meeting's goals, and carry unresolved goals into the next meeting. Our recap and handoff designs drew on an analysis of eight recorded team meetings, and have not yet been tested with users.

From Designing Interfaces that Support Temporal Work Across Meetings with Generative AI, Goals as First-Class Abstractions in Human-AI Collaboration

How should agencies evaluate AI meeting tools?

Agencies should be careful about relying on surveys. In our field experiment, rating meetings afterwards seems to have changed how people approached them, so the comparison group was nudged too. Self-reported benefits can be blurred this way, and self-reported time savings may overstate them. Evaluations should include a group that is not surveyed, or measures that do not rely on self-report, with privacy safeguards.

From Nudging Attention to Workplace Meeting Goals

What we don't know yet

Open questionWhy it matters
Does rating meetings afterwards, on its own, lead to clearer goals?In our field experiment with 361 employees, both groups reported clearer goals and communication by the end, with or without the pre-meeting prompt. The post-meeting surveys, meant only to measure effectiveness, seem to have worked as regular reminders to reflect on goals.
Does AI coaching before a meeting change what happens in it?In follow-up surveys from our participatory prompting study with 18 employees, four said they felt more prepared and their meetings were more effective, and four said their meetings stayed focused. Three said asking attendees to add to the agenda improved attendance and engagement, but the study had no control group.
Does in-depth support for important meetings beat light prompts for all?The two ends were tested in separate studies, never against each other.
Do AI recaps and handoffs help real teams over time?Our designs for instant recaps and meeting handoffs have not been tested with users.
Do shared goal objects improve how people and AI agents work together?Our position paper proposes shared goal objects and six requirements for them, but no system built on them has been tested.
How can organisers and attendees share the work of reflecting on goals?In our field experiment, some attendees felt unable to change meetings, and answering as organisers, 27% of participants doubted attendees would read or understand the goals.
Do these findings hold outside one global technology company?Nearly everyone in our studies came from one global technology company.

The studies

Where this started. The intentionality work grew out of nine earlier papers on remote and hybrid meetings, several from research run during the pandemic. They found purpose problems before we set out to study purpose itself.