BlogGuidesVoice AI terms, explained
What is an AI meeting summary, and what does it leave out?
An AI meeting summary is a short account generated from a transcript to help people recall the discussion [1]. Microsoft says its Teams AI notes require a transcribed event that is five minutes or longer [1]. It helps people who need a quick reminder and can check details against the source.
Short answer
- A summary compresses meeting material into a shorter account; it is not a full transcript [1].
- The output depends on what the system captured and transcribed [1][2].
- Names, numbers, decisions, and follow-up owners deserve a check against the recording or transcript [2][3].
- The word “summary” does not tell you what audio, transcript, or other meeting data a service stores.
Disclosure. Candor is an app that will compete with some products named here. We describe every product, including Candor, from its own documentation, and we'll test them all with the same published method.
You can join the Candor waitlist at candor.inc.
What does an AI meeting summary contain?
The phrase describes an output, not one fixed format. It may be a paragraph of main points, a list of decisions, or suggested tasks. For example, Microsoft documents AI-generated notes and follow-up tasks in Teams recaps, while its wider recap can also show the recording, transcript, shared files, notes and agenda [1]. The summary is the compressed layer; those other materials give a reader more context.
That distinction matters when someone asks what was actually agreed. A transcript is an attempt to write down the spoken words. A summary selects and restates some of them. A task list turns parts of the discussion into proposed work. Those records answer different questions, and a service may offer them together or separately [1][3].
How is a summary made?
In a transcript-based system, the meeting first has to be recorded or transcribed. Speech recognition turns the audio into text; a language model then uses that text to produce a shorter account, sometimes with notes or tasks. Microsoft says the event must be at least transcribed to use its AI recap feature [1].
This is a little like a reporter writing a brief from a court transcript. The brief is easier to scan, but the transcript remains the record to consult when a sentence matters. A summary can only work with the material its system receives. If a microphone misses a quiet speaker, the transcript may miss the words; the summary can then leave out the point or attach it to the wrong context. That last step is an inference from the documented transcript dependency, not a claim about every product.
What can a summary leave out or get wrong?
Compression removes detail. A short recap may preserve the topic and drop a qualification, a rejected proposal, a reason for a decision, or the difference between “we will do this” and “we might do this.” That can make the prose read more firmly than the conversation did. If the output includes a follow-up task, check that the meeting actually assigned it and that the named owner accepted it; generated task text is not proof of agreement [1][2].
Transcription adds a second source of error. A name, date, amount, acronym, or negation can be heard or transcribed incorrectly, and later text can repeat the mistake. Microsoft says AI-generated content may be inaccurate or incomplete and advises checking it against the original recording or transcript [1][2]. A polished sentence is still a claim to verify.
The summary also may not carry everything that shaped the discussion. A slide, shared document, chat message or visual demonstration may be present in the meeting record but absent from a text summary. Teams documents shared files and presentation data as separate recap material, which is a useful reminder to open the source item when the summary refers to it [1].
How should you check one before sharing it?
- Read the summary once for the main topic and stated decision.
- Check names, dates, figures, deadlines, and any sentence containing “will,” “must,” or “approved” against the transcript or recording.
- For each action, confirm the task, owner, and due date in the meeting record or with the person named. Edit or remove a task that was only suggested.
- Open linked files or slides if the recap depends on them. The transcript captures speech, not every visual detail [1–3].
Keep the full record nearby when the notes will drive a consequential decision. A summary is a useful index into a meeting, but it is a poor substitute for the evidence when a disputed word or commitment matters. Microsoft's FAQ for Teams meeting audio recap likewise says users can listen to the original recording or read the transcript to check AI-generated content for accuracy [2].
Can you trace a summary line back to the transcript?
Step 2 above is much faster when each summary point links to the passage it came from. Without a link, you search the transcript for a name or a number and hope the summary used the same words. With one, you open the cited passage and read the sentences around it, including any "might" or "if" the summary dropped.
Candor is one example: its summaries include citations, and clicking one opens a source preview, from which View in Transcript opens the cited passage (Open a citation). Candor also lets you edit the generated summary, which is where you would remove a task that was only suggested, and your own Notes are not an input to the summary [4]. You can join the Candor waitlist at candor.inc.
A citation shows where a point came from. It does not show that the transcript heard the words correctly, so names and figures still need the checks above.
What this page can't tell you
There is no single format or guarantee attached to “AI meeting summary.” The term alone cannot tell you whether a tool records audio, requires a transcript, identifies speakers, includes shared material, or how long it retains those records. Those details depend on the specific product and its settings; read its current documentation before recording or sharing a recap. This page does not compare summary quality or claim that one tool is more accurate than another.
Questions
Which AI note taker are you using for meetings?
Any good AI note takers for in person meetings?
For an in-person meeting, the first question is whether the recording captures everyone clearly; a summary cannot restore words the recording or transcript missed. Check the transcript and recording before sharing decisions or names [2].
Any AI tool that can handle both online and offline meeting notes?
“Online” and “in person” describe where the sound comes from, not what the summary includes. Check the tool’s documentation for its audio sources, transcript requirement, and whether it can attach shared files or notes to the recap [1].
About this page
Candor is being built as personal voice intelligence for everyone. You can join the Candor waitlist at candor.inc. Candor's own documentation is at candor.inc/docs.
Candor's docs on this page's topic: Summaries and citations.
Last checked 4 November 2026 for Microsoft's pages; Candor's docs were checked 1 November 2026. Report an error to support@candor.inc.
Sources
- Microsoft Support, “Recap in Microsoft Teams” — https://support.microsoft.com/en-us/teams/meetings/recap-in-microsoft-teams — checked 2026-11-04
- Microsoft Support, “Frequently asked questions about meeting audio recap” — https://support.microsoft.com/en-us/teams/meetings-events/frequently-asked-questions-about-meeting-audio-recap — checked 2026-11-04
- Microsoft Support, “Take meeting notes in Microsoft Teams” — https://support.microsoft.com/en-us/teams/meetings/take-meeting-notes-in-microsoft-teams — checked 2026-11-04
- Candor docs: Summaries and citations — https://candor.inc/docs/meetings/summaries — checked 2026-11-01