Many teams, workshops and collaborative processes are starting to use AI to summarise notes, draft strategies, cluster themes, prepare options and review emerging ideas. These uses can be valuable, especially when they help people move from scattered material to something they can discuss. But an AI-assisted draft can also shape what a group notices, questions and treats as ready to work from. This post looks at how teams can use AI-generated syntheses as working material, while keeping the framing open to collective judgement.
In collaborative work, the first organised account of an issue often matters more than we realise.
A facilitator groups workshop comments into themes. An evaluator synthesises interview findings. A programme team drafts a Theory of Change. A policy group turns scattered evidence into an options paper. These are working drafts, but they still carry influence.
A first synthesis gives people something to respond to. It also brings choices forward: the language used, the boundaries drawn, the way ideas are grouped, and the account of what may already be known. A set of themes suggests what belongs together. A diagram suggests what counts as a pathway, a boundary, a condition or an outcome.
This has always been part of facilitation, evaluation, planning and systems work. Generative AI does not create the issue, but it changes the conditions under which early syntheses are produced.
AI can now summarise notes, cluster comments, draft a Theory of Change, compare options, produce a risk scan or prepare a discussion paper much more quickly than before. This can help people move from scattered material to something they can question, revise and discuss.
But it can also make early framing feel more complete than it is.
The issue, then, is not simply whether AI supports or replaces human judgement. It is how collective judgement is exercised when AI has helped prepare an early version of what people are asked to judge.
What AI changes
The practical change is speed and fluency.
AI can make early syntheses faster, smoother and easier to produce. That can be a real gain. It can create a starting point where people are struggling to get started, help a team compare different ways of grouping the same material, or reconnect people with ideas scattered across documents, notes and memories.
The risk is that fluent first drafts can feel more settled than they are.
The categories can arrive already named. A synthesis can be broadly accurate and still frame the issue in a way that narrows later discussion. It can include the main points and still smooth over disagreement. It can summarise the available material and still miss what people know but have not written down.
When a working draft becomes the frame
This becomes sharper when an AI-assisted synthesis becomes a shared object for a group.
The concern is not only that one person may be influenced by an AI output. It is that a group may begin to organise its shared attention around an artefact that no one in the group has fully authored, and that no one has yet tested carefully enough.
A working draft can then become more than a draft. It shapes the questions people ask, the disagreements they notice, the pathways they consider, and the decisions that seem ready to be made.
I saw this in one collaborative process, where I was working with a small team developing a strategy for a complex area of practice. The group knew the work well, but the discussion stayed close to how the work was already organised. It was harder for them to step back and articulate the wider strategy that might be needed.
To broaden the discussion, I asked an AI tool to prepare a first draft strategy, using material that was suitable for that purpose, rather than confidential notes or unreviewed stakeholder input. The draft was not context-rich, and it was not ready to use. But it pulled together a wider set of functions, dependencies and practical requirements than the group had been able to name from within the existing conversation.
I brought the draft back to the team and made clear that it had been generated with AI. That disclosure mattered. Without a visible author, it was less clear whose lens the draft carried. Once its provenance was clear, they seemed better able to treat it as something to be tested rather than as an authoritative account.
The draft was more useful than expected, not because it was right, but because it opened up the discussion. The team then changed it substantially: correcting the language, adding context, removing what did not fit, and reshaping it around their own understanding of the work.
For me, the example holds both the value and the risk. The AI draft helped unlock the conversation. But it only became useful because people treated it as provisional and reworked it carefully. In another setting, with less time or support for review, the same kind of fluent first draft could easily have been accepted too quickly.
This does not mean AI should not be used to help prepare material for shared work. It does mean that the first AI-assisted synthesis needs to be treated as something to examine, not simply something to improve.
What needs to stay open
The issue is not only what AI can or cannot do. It is what people allow AI-shaped material to frame before they have had a chance to examine it together.
Some things AI cannot hold in practice. It can process text about relationships, place, history, culture, trust and lived experience, but that is not the same as carrying those things in the work itself. It does not hold the relationships, or share the accountability. It does not know what it means for a group to work together over time in a particular setting.
Other things AI may be able to influence, but should not be allowed to define. It should not determine who has standing in a conversation, whose knowledge counts, what trade-offs are acceptable, or when enough agreement has been reached. These are questions of legitimacy, responsibility and judgement.
There is also a middle space that needs care. AI can work with codified local knowledge: reports, plans, transcripts, submissions and datasets. It can help reconnect people with material that is too easily lost across long processes.
But codified knowledge is not the same as knowledge held in practice. A report may contain a community’s words, but not the relationships, histories or tensions that shaped those words. A transcript may preserve what was said, but not always what was difficult to say. In a long-running place-based programme, for example, a report may record what people said, but not the trust, caution or history that shaped how they said it.
So the task is not simply to ask whether the synthesis is correct. We also need to ask what kind of account has been produced: what has been grouped together, what has been left unnamed, and what someone who was present might recognise as missing.
In shared work, a first synthesis should help people enter the conversation. It should not quietly decide the terms of that conversation.
Collective judgement as a practice
Collective judgement is not just the final decision a group reaches. It is the shared work of noticing what matters, questioning how things have been framed, and deciding what needs attention.
AI can support this work when it helps people see more, prepare better, or test their thinking. But it can weaken the work when it makes a particular framing feel more settled than it should be, encouraging premature closure around a version that still needs to be questioned.
This is why the practice around AI matters as much as the output.
If AI has helped prepare a summary, diagram, briefing or draft framing, its role should be visible enough to discuss. People need to know what kind of artefact they are working with. That helps them decide how much confidence to place in the material, and how to question it.
There is also a facilitation task here. A group may need clear permission to challenge the synthesis rather than simply improve it. It may be useful to ask people what feels missing, what feels too tidy, what wording does not sit right, or what another group might say differently. The point is to reopen the frame before it becomes too comfortable.
This is not about slowing every process down. But where AI has helped create an early synthesis, part of the work is to make sure the group can still judge the synthesis itself.
Questions to keep the frame open
The following questions may help teams, facilitators, evaluators and programme leads work with AI-assisted syntheses more carefully.
• How has AI been used in preparing this material, and what material did it work from?
• What has the synthesis grouped, renamed, smoothed or made more prominent?
• What assumptions does this version make easier to accept?
• What forms of knowledge, context or disagreement may sit outside the version now in front of us?
• Who has reviewed the framing, and what still needs to be questioned before this becomes the version people work from?
The point of these questions is not to create a new compliance process. It is to keep the work of judgement visible. A first synthesis is useful when it helps people think together. It becomes risky when it quietly closes down too much before people have had a chance to examine it.
Closing
The ease of producing an early synthesis changes the practice around it.
When a draft appears quickly, clearly and confidently, people may move too soon to improving it rather than questioning how it has framed the issue. They may treat it as a neutral starting point, especially when it arrives without a visible author whose angle they can read.
The challenge is not to keep AI away from collaborative work, or to pretend that human syntheses are neutral. The challenge is to make the work of synthesis visible enough that people can question it together.
AI may help prepare the ground, but the work of judgement still needs to be held by people who can ask what this draft has made easier to see, what it has pushed to the side, and what the group still needs to work through.
This post is part of a wider set of Learning for Sustainability pages on using AI in research, practice and collaborative work. Start with the generative AI hub for the full section. Related reflections include Working with AI in the room, AI in place-based practice, and AI prompts for shared thinking.
[* Image by Jintana / Adobe Stock]
