What is an AI note taker? How it works, the main types and the limits

An AI note taker is software that captures the audio of a meeting or call, turns the speech into a transcript, and uses a language model to write a summary, action items or structured notes from that transcript. The main difference between tools is how they get the audio: a bot that joins the call as an extra participant, an app that records the audio on your own device with no bot, or a feature built into Zoom, Microsoft Teams or Google Meet. As of October 2026, Otter, Fireflies and Granola are widely used general-purpose examples, and BrightHire and Metaview are built specifically for recruiting interviews.

This page explains how the technology works, the types, what they produce, the consent and data questions that come with them, and how to choose one. It is written for people who run interviews, but the definitions hold for any meeting.

How does an AI note taker work?

Almost every AI note taker runs the same five steps. Knowing them tells you where the mistakes come from.

1. Capture. The tool gets the audio, either from a bot connected to the meeting, from your device's microphone and system sound, or from the meeting platform itself.
2. Speech-to-text. A speech recognition model converts the audio to text, live or after the call. Accuracy drops with crosstalk, weak microphones, background noise, and unusual names or jargon.
3. Speaker labels. The transcript is split by speaker, a step called diarization. A bot or built-in feature can often use the platform's participant names; a device-based app may label speakers by audio channel, such as your microphone versus everyone else, or by voice.
4. Summarization. A large language model reads the transcript and writes the notes: a summary, decisions, action items, or fields in a template such as a sales call form or an interview scorecard.
5. Output. The notes are stored in the vendor's app or on your computer, and can be searched, shared, exported or sent to other systems such as a CRM or an applicant tracking system (ATS).

Two outputs come out of that process, and they deserve different levels of trust. The transcript is a record of what was said, with recognition errors. The summary is an interpretation of the transcript, written by a model that can leave things out or state them more confidently than the speaker did. Check the summary against the transcript before you rely on it.

What are the main types of AI note taker?

The capture method decides most of what matters in practice: whether other people see the tool, which calls it works on, and where the audio goes. The examples below are taken from each vendor's own site, as of October 2026.

The bot-or-no-bot choice has its own guide, AI note takers without a bot, which covers why some candidates react badly to a bot arriving in an interview.

What does an AI note taker produce?

Most tools produce some combination of these:

• A transcript, timestamped and searchable, usually with speaker labels.
• A summary of the conversation in a few paragraphs or bullet points.
• Action items: who agreed to do what, sometimes assigned to named people.
• Template notes: the same fields filled for every call of a given type, such as a discovery call, a one-to-one, or an interview with a fixed list of competencies.
• Search across calls: asking a question of everything recorded, such as which candidates mentioned a specific system.
• Exports and integrations: email, Slack, documents, or fields written into a CRM or ATS.

For interviews, the most valuable output is usually the least glamorous: the candidate's exact words on the answers that decide the hire. A summary that says "strong stakeholder management" is an opinion; a quote of what the candidate actually said is evidence a second interviewer or a client can weigh.

How is an AI note taker different from transcription or interview intelligence?

The labels overlap in marketing, so it helps to be strict about them.

Do you need consent to use an AI note taker?

Recording law does not care whether a bot is visible. What matters is whether the people on the call agreed. Under US federal law, recording is not unlawful where one party to the conversation consents (18 U.S.C. § 2511(2)(d)), but several states go further. California's Penal Code § 632, for example, requires the consent of all parties to a confidential communication, and the strictest rule among the participants' locations is the one to plan for.

A bot-free tool makes this more important, not less, because nobody on the call can see that notes are being taken unless you say so. The working rule is simple: say it at the start, ask, and stop if the person declines. State-by-state detail is in is it legal to record a job interview and two-party consent states, and wording you can read aloud is in the interview recording consent script.

Where do the recordings and notes go?

Ask every vendor the same questions, in writing, before anyone records a candidate or a client:

• Is the audio stored after the transcript is made, and if so, where and for how long?
• Are transcripts and notes kept in the vendor's cloud, on your device, or both?
• Does the vendor use your recordings or transcripts to train its models, and can you switch that off by contract?
• Who at the vendor, and which subprocessors, can access the content?
• Can you delete one person's data on request, and how quickly?

The answers differ more between products than the feature lists do. The data section of how to use AI in recruiting responsibly turns these questions into a short policy.

What can go wrong with AI notes?

• Misheard specifics. Names, employers, numbers and technical terms are where speech recognition fails most, and they are often the details that matter.
• Wrong speaker. A statement attributed to the candidate that the interviewer actually said will mislead anyone reading the notes later.
• Overstated summaries. A model can turn "I supported the migration" into "led the migration." Compare the verbs in the summary with the transcript.
• Notes you should not keep. A candidate mentions a health condition or childcare in passing, and the summary records it. See what not to write in interview notes.
• Less listening. A tool that takes notes is meant to free your attention for the next question, not to replace paying attention.

How do you choose an AI note taker?

Work through these questions in order; each one rules options out.

1. Which calls? List the platforms you use and whether phone calls are part of the job. A bot cannot attend a phone screen.
2. Bot or no bot? Decide what you want the other person to see, then plan how you will tell them either way.
3. What output? A generic meeting summary, or notes structured for your work, such as a scorecard or a client submittal.
4. Where does the data live? Use the questions above, and check them against your own retention rules.
5. What does it connect to? If a tool claims an integration with your CRM or ATS, ask to see it working with your own system.
6. What does it cost per call? Seat prices, minute caps and free tiers vary; AI interview note-taker pricing compares published prices.

Agencies have extra requirements around phone screens, submittals and ATS notes, covered in AI note takers for staffing agencies.