The AI Notetaker joins virtual interviews, records and transcribes them, and can draft scorecard content from the transcript. This article covers how its settings actually interact, what it can and can't reliably do, and known accuracy limitations — for step-by-step setup and troubleshooting, see "Setup the Kula AI Notetaker" and "AI Notetaker Not Joining Interview."

How the on/off settings actually interact

• This trips people up, so it's worth being precise. There are two separate things at the org level: whether the Notetaker is enabled, and whether it's mandatory.

• If mandatory is on, every interview gets the Notetaker and the per-interview toggle is locked on — nobody can turn it off while scheduling.

• If mandatory is off ("not mandatory"), the Notetaker can be turned on or off for each interview individually, during or after scheduling.

• A real, reported gap: even with the org-level setting not mandatory, ad-hoc interview scheduling has been reported to still default the per-interview toggle to on — meaning "not mandatory" doesn't reliably mean "off by default."

• A privacy-conscious organization rolling out interviewer training before allowing any Notetaker use found interviews being recorded anyway, because the ad-hoc scheduling flow didn't respect the org-level intent. If your organization needs a guarantee that the Notetaker never joins until you say so, verify the per-interview toggle's actual default in ad-hoc scheduling rather than trusting the org setting alone.

• There's no hard "off" switch. Even with mandatory disabled, there's currently no feature-flag-level way to make the Notetaker completely unavailable to every user — it can still be manually turned on per interview, or default on for ad-hoc scheduling. If some people on your team want it and others don't, the only real mechanism today is training people to manually remove the bot from a call they don't want recorded, not a setting that keeps it off by default for some and on for others.

• Configuration is locked in at scheduling time. If an interview was scheduled while the Notetaker was enabled, it will still try to join even if you change the org-level or department-level setting afterward — Kula applies setting changes only to interviews created after the change, not retroactively. To stop it joining an already-scheduled interview, remove it from that specific interview, or have someone remove it from the call once it joins.

What it does

• Once it joins a Google Meet, Zoom, or Microsoft Teams interview, the Notetaker records and transcribes the conversation, then compiles a summary of key discussion points, accessible from the Interviews page. Interviewers can still add their own notes (public or private) alongside the AI-generated content.

• A related but separate feature, the Interview Assistant, lets you ask follow-up questions about a completed interview in a chat interface — it retrieves answers from the interview transcript and summary, similar in spirit to the AI Application Assistant but scoped to a single interview rather than a candidate's whole file.

Scorecards: two separate AI controls

• "Fill scorecard with AI" and the Notetaker recording/transcribing are related but independently controllable. As of a February 2026 release, AI scorecard assistance has its own two-level toggle, separate from whether the Notetaker records at all: an account-level control (Admins can enable or disable AI scorecard assistance org-wide) and an interview-level control (recruiters can toggle AI scorecard assistance per interview).

• This means you can keep the Notetaker recording and transcribing while turning off AI-assisted scorecard filling specifically — useful if you want interviewers to write their own evaluation in their own words rather than have it drafted for them.

• Finer-grained control — choosing exactly which scorecard questions AI is allowed to fill, rather than an all-or-nothing toggle per interview — has been requested by customers but isn't built yet.

• More structured, competency-based scorecards produce more reliable AI-filled feedback than a single general feedback field — if you're seeing shallow or generic AI-filled scorecards, a more specific rubric is the first thing to try.

What it doesn't reliably do

• Phone-call interviews aren't supported. The Notetaker only works with virtual meeting platforms (Google Meet, Zoom, Teams) — a real, requested capability, still unbuilt, that matters most for hiring processes where phone screens are the norm (e.g., healthcare recruiting).

• In-person interviews are unreliable at best. See the Needs Verification note above — even where the Notetaker can technically be brought into an in-person interview via a live virtual meeting link left running in the room, it can't reliably distinguish speakers or produce an accurate transcript; it's effectively just a recording in that scenario, not a real transcription/summary source.

• Interviews starting late may go uncaptured. The Notetaker joins at the scheduled time and can go inactive on silence or delay — a delayed interview start can mean it simply isn't recording when the conversation actually begins, with no clear indication anything was missed. This is a real, currently open concern.

• No custom vocabulary or spelling correction. The Notetaker will transcribe company names, product names, or technical terms exactly as it hears them, with no way to teach it correct spellings — a real, reported frustration for at least one customer whose company name was consistently mis-transcribed.

• No way to train or tune it to a team's preferences over time, and no option to see verbatim candidate answers separated from the AI summary — some customers have asked for both and neither exists today.

Good to know

• Summaries can read more positively than the actual interview warranted. This is a real, recurring pattern raised by more than one customer over time — described internally as the AI sometimes adding "spin" or missing concerns that were actually raised verbally. Treat AI-generated summaries as a starting point to verify against the transcript for anything that will materially affect a hiring decision, not as a substitute for actually reading or listening to the interview.

• The Notetaker can occasionally state something not actually said in the interview (misnaming, stale assumptions) — a second, related trust issue distinct from summary positivity.

• Non-English interviews have variable transcript quality. The Notetaker supports multiple languages, but accuracy depends on the specific language and audio quality.

• Google Meet requires a human to manually admit the Notetaker — this is a Google platform limitation (Google Meet doesn't support bots joining automatically), not a Kula one. Tell your interview panel in advance to expect and admit it, or it will time out.

• Microsoft Teams tenant policies can silently block recording even after the bot joins the call — some organizations' Teams policies block third-party bots from capturing audio, so the Notetaker can appear to join successfully but produce nothing.

• Removing the Notetaker from a call it already joined is the standard workaround for anyone who doesn't want a specific interview recorded, given there's no reliable default-off-for-some-users mechanism today.

• Recording-consent language is not inserted by Kula by default. If you want candidates or interviewers notified that the interview may be recorded and transcribed by AI, add that language to your interview invite yourself — Kula doesn't add it automatically, because the right wording varies by customer and jurisdiction. See "AI data privacy and candidate consent."

FAQ

• If I turn off "mandatory" at the org level, is the Notetaker off for everyone by default? Not reliably — ad-hoc interview scheduling has been reported to still default the toggle to on even when the org setting isn't mandatory. Check the actual per-interview default in your account rather than assuming.

• Can I stop the Notetaker from joining an interview that's already scheduled? Changing the org or department setting won't affect it — that interview's configuration was locked in when it was scheduled. Remove the Notetaker from that specific interview, or have someone remove it from the call once it joins.

• Does the Notetaker work for phone interviews or in-person panels? No reliable support for either. Phone calls aren't supported at all. In-person interviews may technically capture audio if a virtual meeting link is left running in the room, but transcription and speaker separation aren't reliable in that setup.

• Can I turn off AI-filled scorecards but keep the Notetaker recording? Yes — AI scorecard assistance has its own account-level and interview-level toggle, separate from whether the Notetaker records and transcribes at all.

• Should I trust the AI-generated interview summary as accurate? Treat it as a helpful starting point, not verbatim truth. Summaries have a known, recurring tendency to read more positively than the actual interview, and can occasionally state things not actually said. Verify against the transcript for anything that matters to your decision.

• Does Kula tell candidates their interview is being recorded by AI? Not automatically — there's no default recording-consent language inserted into interview invites. Add your own if you want candidates or interviewers notified.

Need help?

If the Notetaker isn't joining or recording correctly, see "AI Notetaker Not Joining Interview" for step-by-step troubleshooting. For other questions, reach out to us at support@kula.ai or use the in-app chat.