7 AI prompts for meeting notes, with checks for each answer

Useful AI prompts for meeting notes specify the question, the meetings to use, the evidence required and what to do when the record is incomplete. Start with one concrete task, such as separating decisions from suggestions or comparing how a plan changed. Ask for sources, then check the answer before using it.

“Summarise this meeting” can be a reasonable first request. It is less useful when you already have a summary and need to decide what happens next.

The seven prompts below address different jobs after a meeting. They are selected for practical usefulness, a clear evidence requirement and an output a person can review. They are not ranked by model performance, and they are not tested guarantees of accuracy. Adapt the wording to the material and tools you are authorised to use.

Before you paste a prompt

Decide which records belong in the answer. A single meeting can establish what someone said that day. A question about changes, recurring issues or earlier commitments needs several relevant records and a defined period.

Name the source set explicitly. If you paste notes into a tool, label each set with a meeting title and date. If your tool works over a recording library, confirm the selected people, project or dates. Do not assume the model can see a meeting merely because you can see it elsewhere in the app.

Give it a rule for uncertainty: report what the records support, separate suggestions from facts and say when information is missing. If source links or timestamps are unavailable, request the meeting title and supporting passage instead. Do not invite it to invent citations that look convincing.

Finally, check the processing route before supplying confidential content. Weeve can work with saved recordings through local chat; optional ChatGPT processing sends relevant text to the external provider. A prompt does not change that boundary. See how Weeve stays private by default.

1. Separate decisions from suggestions

Use this when: an energetic discussion left several possible approaches in the notes, and you need to know which were actually agreed.

Using the selected meeting, separate confirmed decisions, proposals and unresolved questions. For each confirmed decision, show the supporting passage and source. Preserve any conditions attached to it. If agreement is unclear, put it under unresolved rather than treating it as a decision.

The useful output is a short set of clearly labelled items. It should preserve “if the test passes”, “subject to approval” and similar conditions rather than flattening them into a definite commitment.

Check the answer: open the cited passage and read the surrounding discussion. Did someone explicitly agree, or did a participant simply propose an idea? Was there a later objection or qualification in the same meeting?

For a fictional example, “We could invite the whole team next week” is a proposal. “Let’s invite the whole team once the pilot has passed the access check” is an agreement with a condition. Removing the condition changes what the team should do.

This prompt helps identify candidate entries for a decision log. A person still needs to review which entries belong in the maintained record.

2. Extract commitments without inventing owners or deadlines

Use this when: you need a reliable handoff from conversation to the place where work is tracked.

List the commitments in this meeting. For each one, include the task, the stated owner, the stated deadline and a source passage. Use “not stated” where an owner or deadline is absent. Put your suggested next steps in a separate section. Do not treat a commitment as evidence that the work is complete.

A useful answer distinguishes a request from an accepted commitment. “Could someone check the contract?” and “I will check it tomorrow” do not carry the same ownership information.

Check the answer: confirm who was speaking and what “tomorrow” refers to. If the record lacks the meeting date or timezone needed to resolve a relative deadline, leave the uncertainty visible rather than converting it into a confident calendar date.

Do not infer an owner because someone has the most relevant job title. A proposed assignment can be useful, but it needs an explicit label and human agreement. Likewise, a task repeated in several meetings may still be open; repetition is not completion evidence.

After review, transfer the approved work into your task system. The guide to tracking meeting action items explains why an extracted list needs that next step.

3. Compare what changed between meetings

Use this when: you remember an earlier plan but need to establish how the latest discussion differs.

Compare the initial planning meeting with the latest review for this project. Show changes to scope, timing, responsibilities and unresolved questions. Cite evidence from both meetings for each claimed change. Distinguish an explicit change from a topic that simply was not mentioned again. Tell me if either source is incomplete.

This is a comparison task, so the answer needs evidence from both sides. A detailed account of the latest meeting alone cannot establish what changed.

Check the answer: confirm the intended meetings were selected and that the comparison uses the same subject. A delivery date for the pilot is not necessarily a delivery date for the full rollout. An apparent contradiction can disappear once those scopes are separated.

If the earlier meeting proposes a September launch and the later one discusses an October launch, ask whether October was explicitly agreed or merely considered. If the latest meeting says nothing about the launch, the careful answer is that its status remains unclear in that record.

For a long project, compare meaningful checkpoints before asking for its entire history. This makes it easier to inspect the evidence and identify where an intermediate conversation may be missing.

4. Find recurring issues and contradictory evidence

Use this when: several conversations appear to raise the same concern, and you want to examine the pattern without overstating it.

Review the selected meetings from this project and period. Identify recurring issues, show which meetings support each issue and include evidence that contradicts or narrows it. Separate repeated mentions by one person from reports by different people. State what material you reviewed and what you could not establish.

The output should show the evidence behind a possible pattern, not just a confident theme label. Several mentions in one meeting are different from similar concerns raised independently across several interviews.

Check the answer: inspect whether the examples are about the same underlying issue. “I could not sign in” might refer to a forgotten password, an expired invitation or an unavailable service. Combining all three under one product defect could lead to the wrong response.

Treat the result as an aid to review. Selected conversations are not automatically representative of every customer or employee. The number of mentions in your collection is not a market-size estimate, a satisfaction score or proof of causation.

If you are doing formal research synthesis, use a method appropriate to that work. The research-synthesis guide goes deeper into evidence, contradictions and the limits of conclusions drawn from interviews.

5. Turn unresolved questions into a focused next-meeting agenda

Use this when: another meeting is approaching and you want it to resolve something specific.

From the selected earlier meetings, identify unresolved questions that still need a decision or clarification. Propose an agenda organised around those questions. For each item, include the earlier source, the outcome needed and any preparation explicitly requested. Keep your proposed priorities separate from previously agreed commitments.

This should produce a proposed agenda, not imply that a meeting has been scheduled or that everyone has accepted its priorities.

Check the answer: look for a later source that already resolves a question. Otherwise an old issue can return to the agenda simply because it appeared prominently in an earlier conversation.

For a fictional pilot review, “Discuss onboarding” is broad. “Decide whether invitation expiry must be fixed before the next cohort” gives the discussion a concrete outcome. The source should establish why that question exists; your team decides whether it belongs in the next meeting.

Remove items that can be resolved through a short written clarification. For the remaining items, identify the information people need beforehand. The client-meeting preparation guide explains how to turn prior context into useful preparation.

6. Draft a follow-up for a specific reader

Use this when: the same meeting needs different follow-ups for a client, an internal colleague or another stakeholder.

Draft a follow-up for the named reader using this meeting. Include confirmed decisions, agreed next steps and questions that need their response. Exclude internal commentary and material that is not appropriate for that audience. Do not add a promise, deadline or recipient detail that is absent from the record. Mark anything I need to confirm before sending.

The audience is part of the task. A full internal summary is rarely the right client email, even if every sentence is factually accurate.

Check the answer: verify names, commitments, dates and any language that could sound like approval. Review excluded material as well: a concise message still needs the conditions that make an agreement accurate.

If you want to propose a new deadline, add it deliberately and label it as a proposal. Do not let a fluent draft convert “we need to discuss timing” into “we will deliver on Friday”.

Weeve supports drafting follow-up text for review; opening or copying a draft is not the same as sending it. For the writing structure, see how to write a meeting follow-up email.

7. Audit an answer against its sources

Use this when: you have a useful draft but need to check whether its main claims are actually supported.

Review the answer above against the selected source records. For each important claim, identify the supporting passage, any relevant qualification and whether the evidence is direct, inferred or missing. Highlight contradictions and unsupported names, numbers, dates or commitments. Do not rewrite the source to make it agree with the answer.

This prompt creates a review map. It does not turn the model into an independent verifier, particularly when the same model produced the original answer. Use the map to guide your own checks.

Native Weeve preview with synthetic demonstration content.

Check the answer: open the source rather than trusting the second response’s description of it. Where wording or speaker attribution matters, inspect the surrounding transcript and retained audio where available. The audio-transcription error guide explains what to do when the underlying record itself is unreliable.

Weeve’s sourced chat can help you open supporting material from saved recordings. That is useful for review, but it is not a guarantee of exhaustive retrieval or exact transcription. See chat with recordings for the product workflow.

Choose the prompt by the decision you need to make

Use the decision prompt when agreement is unclear, the commitment prompt for a work handoff and the comparison prompt when the plan appears to have changed. Use recurring-issue analysis to investigate a possible pattern, agenda preparation to plan the next discussion and audience-specific drafting to prepare a follow-up. Use the audit prompt before relying on an important output.

You usually do not need all seven after every meeting. Running every prompt mechanically can create more material to review than the original conversation. Choose the smallest task that answers your current question, then ask a focused follow-up if a gap remains.

If a prompt produces a vague answer, narrow the source set or request the specific evidence you need. If it produces an unsupported answer, preserve the uncertainty and return to the source. Adding stronger wording such as “be completely accurate” cannot supply evidence that the records do not contain.

A short review before you use the result

Confirm that the answer uses the intended meetings and dates, separates facts from suggestions, retains conditions and leaves missing information visible. Check the passages behind important decisions, commitments and quotations. Review the audience and processing route before sharing anything further.

The useful outcome is not a longer AI response. It is a clearer question, a traceable answer and a next step that you can justify from the meeting record.