AI Note Taker for Recruiters: Candidate Consent and Tool Choice

Otter, Fireflies and Fathom will all capture a candidate call acceptably, so choose on price, ATS routing and how visible the bot is. The harder question is consent: tell every candidate before recording starts, in writing and aloud, and get a clear yes. Some places require everyone's agreement, so the safest rule is the strictest.

The tool decision takes an afternoon. The consent step, the retention rule and the route into your ATS take longer, and they are where recruiters get into trouble: a bot that joins a call uninvited, a recording kept for years, a transcript pasted into a candidate record with details nobody should store. Some applicant tracking systems now build the consent step in: Greenhouse's own Notetaker asks candidates for consent when they book their interview slot. All of this is practical guidance; have an adviser check the final wording for your situation.

Follow me on Instagram@sagnikteaches

When a candidate must be told, case by case

Rules on recording calls differ between places. Some require every participant to agree, others only one party, and data-protection law, such as the GDPR if you speak to candidates in the EU, adds a duty to tell people what personal data you collect and why. You may be speaking to candidates in several places in one week. The workable policy is therefore not to work out which rule applies each time, but to follow the strictest one everywhere.

Connect on LinkedInSagnik Bhattacharya

One more point deserves its own line: the note taker should never be the last thing a candidate learns about. The first mention belongs in the booking message, so nobody is surprised on the call.

Subscribe on YouTube@sagnikteaches

Consent wording you can copy

Three pieces of wording cover most of the process. Edit them for your agency, and have an adviser read them once.

In the call invitation

Hi [first name],

Our call is booked for [day, time]. I use an AI note-taking tool so I can listen properly instead of typing. It records and transcribes the call, and I keep the notes only for this recruitment process. If you would prefer I did not record, just reply and I will take notes by hand instead. Our privacy notice is here: [link].

[your name]

At the start of the call

Before we start: I use an AI tool to record and transcribe this call so my notes are accurate. The recording is kept for [period] and seen only by our recruitment team and, if you are put forward, the hiring client. Are you happy for me to record? If not, that is completely fine and it will not affect anything.

Wait for an explicit yes, and let the transcript capture it. The recording of the yes is useful evidence that the step happened. An illustrative opening that shows the pattern working: the recruiter says the sentence, the candidate says "that's fine", the recorder is switched on after the yes, and the transcript begins with the consent exchange.

If the candidate says no

No problem at all. I'll switch it off and take notes by hand. Shall we carry on?

Keep that reply in muscle memory. The recruiters most likely to push on after a refusal are those on a tight day, which is why a written rule beats a private judgement.

Otter, Fireflies and Fathom on price, storage and consent features

Prices below come from the vendors' pages, read on 1 November 2026, except Fireflies, whose page showed no figures that day; its prices are from a late-October check. Check each before buying, because note taker pricing moves often.

Read the consent row carefully. A feature that lets you disable the in-meeting banner, as Fathom lists, is a convenience that creates a duty: if the banner is off, your spoken notice and written invitation must carry the load. Conversely, features such as Otter's recording disclaimer email push the notice to the candidate for you, which suits a high-volume desk.

Seat prices against 2,500 minutes of calls a month

An illustrative agency has five recruiters, each holding about five candidate calls a day of roughly 25 minutes, over 20 working days. That is 125 minutes a day, or about 2,500 minutes a month per recruiter.

That volume matters for Otter: its Pro plan allows 1,200 minutes a month per user, so a busy recruiter would hit the cap in the first half of the month and need Business. At the annual rates above, five Business seats cost 5 x $19.99 = $99.95 a month. Five Fireflies Pro seats at $10 annual would be $50 a month, five Fathom Team seats at $15 would be $75 (the plan has a two-user minimum, which five seats clears), and five Fathom Business seats would be $125. Five Fireflies Business seats at $19 come to $95.

Storage changes the Fireflies sum. Its Pro plan does not cap the meetings its bot records each month, but it stores 8,000 minutes per user in total, so a recruiter recording 2,500 minutes a month fills it in a little over three months. A 30-day deletion rule, which you want for candidates anyway, keeps each library near 2,500 minutes and Pro is enough; a keep-everything habit pushes you to Business, where storage is unlimited. For a solo recruiter, note that Fathom's Team and Business plans need two users, while its Premium plan, at $16 a month billed annually, is for one person.

Now the time side. If each recruiter saves five minutes of typing per call, that is 25 minutes a day, roughly eight hours a month. At a loaded cost of $30 an hour, that is $240 of recruiter time per person per month against a seat price of $10 to $25. The sum is comfortable on any of the three, so price should not decide it. What decides it is whether the consent features, the integrations and the retention controls fit how your desk works. The related tutorial on time saved per role gives a method for measuring your own figure rather than trusting mine.

Where the notes land in the ATS

A transcript that lives only in the note taker's own library is a second, unmanaged candidate database. Decide how the useful part reaches your applicant tracking system, and what stays out of it.

There are three routes. The first is a native one: some applicant tracking systems now have their own notetakers, so the transcript sits on the candidate record. Greenhouse's Notetaker is now included in its Core, Plus and Pro tiers: it records Zoom, Google Meet and Teams interviews, maps the notes to scorecard questions, collects candidate consent through self-scheduling and lets admins set retention of up to 60 days. Ashby sells its AI Notetaker as a paid add-on on every plan, with consent controls. If you already use one of those, test it before paying for a separate tool. Be aware that vendor comparison pages are written by vendors; ask for a live demonstration on your own data.

The second route is an automation tool. None of the three pricing pages I read names an ATS, though Otter and Fathom list HubSpot and Salesforce sync, and Fireflies offers Zapier on its paid plans. An automation can post a summary and a link to the candidate record when the call ends. Test that it attaches to the right candidate, since a mismatch is a data error that takes minutes to create and weeks to find.

The third route is manual and fine for a small desk: the recruiter pastes a short structured summary and a link into the record. Here is a filled-in example of what to paste, and what to leave out. All details are invented for illustration.

CALL NOTE - [date] - [recruiter initials]
Role discussed: Senior accounts payable, hybrid
Current status: In post, 3 months' notice
Salary expectation: $62,000 to $66,000 (stated by candidate)
Reason for looking: limited progression
Skills confirmed: month-end close, supplier reconciliations, one ERP system
Availability for interview: weekday evenings
Consent to record: given at start of call; recording kept 30 days
Next step: send to client Friday if reference check agreed

Notice what is not there: health, family plans, age, religion and anything else the candidate mentioned that does not bear on the job. AI summaries often include those details because they were said. Edit them out before filing. A recruiter's first prompt can enforce this, as the next section shows.

Prompt the summary to keep to job-relevant facts

Most of these tools let you choose or edit a summary template. If yours allows a custom prompt, use something like this.

Summarise this recruiter call using only job-relevant facts: current role, skills, salary expectation, notice period, availability, location preference and reasons for looking. Do not include health, family, age, religion, politics or any personal circumstance unrelated to the role. Quote salary and notice period exactly as the candidate said them. If something is unclear, write "unclear" rather than guessing.

An illustrative before and after shows why. The raw AI summary says: "Candidate mentioned she is expecting a baby in the spring, is looking for something closer to home, and wants about 65k." The edited note says: "Wants a role closer to home. Salary expectation about $65,000 (candidate said '65k', confirm currency and basis)." The first version is an accurate record of what was said and a liability to store; the second keeps what the job needs. For a related method, see how recruiters use AI to write candidate summaries clients read.

Retention and storage: decide before the first call

The question nobody asks at purchase is how long the recording lives. Set a rule in writing and apply it in the tool's settings, not in anyone's memory. An illustrative schedule, to be set with your adviser and checked against the law that applies to you:

• Audio and video: delete after the call has been summarised and checked, or after a short fixed period such as 30 days.
• Transcript: keep for the length of the live recruitment process, then delete or anonymise.
• Short structured note in the ATS: keep according to your normal candidate-data retention policy.
• Candidate who asks for deletion: delete the recording, transcript and note taker library copy, then confirm in writing.

Look at what each tool lets you control. Fireflies lists custom data retention and an option it calls "Transcript and Summary Only Mode", which suggests keeping text without the audio, without saying which plan includes either. Fathom lists custom data retention on Enterprise, and Greenhouse's built-in Notetaker lets admins set retention of up to 60 days. Whether a setting exists on your plan is a question for the vendor, in writing. The point of the rule is that the default setting in most tools is to keep everything, and a default is not a policy.

Access matters as much as time. Make sure recordings are private by default, that shared links need a sign-in, and that a recruiter leaving the agency does not take a library of candidate calls on a personal account. The tutorial on what to check in an AI tool's privacy policy and terms lists the questions to ask about training, sub-processors and deletion.

Where AI note takers misreport candidate calls

Transcription errors cluster where recruiters care most: numbers, names and short answers. Four failures to look for, with how they show up.

1. Salary figures. "Sixty-five" becomes "fifty-five" on a poor line, or a monthly figure is recorded as annual. Check every figure against the audio when it goes to a client.
2. Notice periods and dates. "Four weeks" and "forty weeks" sound alike on a noisy call. Read them back to the candidate at the end of the call.
3. Names and employers. Uncommon names and company names are mangled. Correct them before the note reaches the ATS, or search will not find the candidate again.
4. Confident additions. A summary states that the candidate "has experience managing a team of ten" when she said she had supervised two interns. The AI smoothed a vague statement into a firm claim. Compare each claim with the transcript line before it goes into a client submission.

For checks on how AI treats different candidates, run through the bias checks every recruiter should run before you let any AI output influence who goes forward.

A one-week trial that tests consent as well as accuracy

Run two tools side by side for a week, with colleagues playing candidates. Score each against these items, and keep the results in a small table.

1. Does the consent message reach the candidate before the call, and can you prove it?
2. When a "candidate" says no, how quickly can the recruiter stop the recording, and does anything still get stored?
3. Does the transcript start with the consent exchange?
4. With a strong accent, a poor line or two people talking at once, how many of ten key facts (salary, notice, location, skills, dates) are captured correctly?
5. How do you delete one recording, and is it really gone from the library and any shared link?
6. Can you push a summary to your ATS in a way the vendor demonstrates live?
7. Who else in your organisation can see the recording by default?

An illustrative result after a trial might read: Tool A captured nine of ten facts but kept recordings by default and had no retention control on the plan tested; Tool B captured eight of ten but allowed a 30-day auto-delete and a disclaimer email. For a regulated, high-volume desk, Tool B can be the better choice despite the lower accuracy score, because the weaker spot is fixable by a human check and the retention gap is not.

Five recording slip-ups on a recruitment desk

• The calendar bot. Many tools can join every meeting on your calendar automatically. A recruiter's calendar includes candidate calls, client calls and personal appointments. Switch auto-join to opt-in per meeting, and check which meetings the tool can see.
• The client's call. A recruiter's bot joins a client's panel interview, and the client's candidate finds a third-party recorder in the room. Ask first, and prefer the client's tool.
• The personal account. A recruiter signs up for a free plan with a personal email. When she leaves, a library of candidate calls leaves with her. Use team accounts and keep an admin who can delete.
• The shared link. A transcript link posted to a client chat is open to anyone with the URL on some settings. Check sharing defaults before the first share.
• The silent refusal. A candidate says "I'd rather not" in a tone the recruiter treats as joking. Ask again plainly, and log the answer.

Other professions with confidential calls have worked through similar rules; AI note takers in lawyers' client meetings is a useful comparison for tone and structure.

Matching the set-up to the size of the desk

• Solo recruiter or very small desk. A free or entry paid plan from any of the three, a strict consent script, 30-day deletion, and manual paste of a structured note into the ATS.
• Five to fifteen recruiters. Pick one tool for everyone, with team admin and retention controls, and test an automation into your ATS. Compare it with your ATS vendor's own notetaker if it has one.
• Agency that submits to enterprise clients. Expect client security questionnaires. Choose on retention, access control and deletion first, price second, and keep the consent evidence for every call.

If your ATS has AI features of its own, evaluating the AI features in your recruitment software will help you decide whether a separate note taker is worth having. The AI hub covers the wider picture.

If you would rather map this with someone, my AI implementation consultation covers that kind of tool, consent and filing decision; the details are on the contact page.

Questions recruiters ask about recording candidates

Do I have to tell a candidate an AI note taker is on the call?

Yes, as a rule of practice. Recording rules differ from place to place, and some require every participant to agree. If you speak to candidates in the EU, data-protection law adds a duty to tell them what you collect and why. Telling everyone, every time, before recording starts is the one rule that works everywhere.

What if a candidate says no to recording?

Switch the tool off, say so, and take handwritten or typed notes instead. Record the refusal in the candidate's file without penalty. Never let a refusal affect how you treat the candidate; that is the point of asking, and it protects you if the process is ever questioned.

Can the AI score or rank candidates from the call?

Avoid it. Scoring from a transcript can reproduce bias, and some places regulate automated tools used in hiring decisions. Use the note taker for accurate records and summaries, and keep the judgement with a person using a written scorecard. Ask an adviser before switching on any rating feature.

Which tool puts notes straight into my ATS?

None of the three pricing pages I read names an ATS. Otter and Fathom list CRM sync for HubSpot and Salesforce, and Fireflies offers Zapier on paid plans. Some ATS vendors, such as Ashby and Greenhouse, have their own notetakers. Ask any vendor to demonstrate a push to your own system before you buy.

Further reads

• AI Screening: Bolt It Onto Your ATS or Switch Systems? — Add AI to your current ATS or move to one that has it built in.
• AI CV Screening for Recruitment Agencies: Setup and Safeguards — Set up CV screening with safeguards for a small agency.
• How to Build Interview Questions and Scorecards With AI — Build the written scorecard that keeps judgement with people.
• Questions to Ask a Recruitment Agency About Its AI Screening — What clients will ask you about your own AI use.
• AI Note Takers for Financial Advisers: Compliant Meeting Records — Compliant meeting records in another regulated trade.
• Telling Patients You Use AI Note-Taking: Consent Wording That Works — Consent wording for recorded consultations in another trade.
• How Staffing Agencies Automate Candidate Follow-Up With AI — What to do with call notes once they reach the ATS.
• AI for Recruitment Business Development: Winning New Clients — Use the time saved on winning new clients.
• AI Interview Software for Video Interviews: A Hiring Checklist for Trades — A hiring checklist for trades firms: which AI interview features to use, which to avoid, and the notice, consent and bias checks to run first.
• AI Sourcing Tools for Recruiters: Seat Costs and Consent Rules Compared — Five sourcing tools priced for three recruiters, with data sources, consent handling and a one-live-search trial you can run in a week.
• Free AI Tool for Recruitment: What a Small Agency Gets for Nothing — What free AI plans and notetakers cover for a small recruitment agency, where free limits and candidate-data risks bite, and when paying makes sense.
• Resume Formatting AI Tools for Recruiters: Check Nothing Was Altered — Which AI CV formatters reflow and which rewrite, what each costs per CV, and a step-by-step check that nothing in the candidate's CV was altered.
• AI Agents for Recruitment Agencies: Sourcing, Screening and Review Points — What an AI agent can take off a recruiter's desk, the four points where a person must check its work, and what agent credits cost per role.
• ChatGPT Prompts for Recruiters: Intake, Outreach and Shortlists — Eight ChatGPT prompts for job intake, outreach messages and shortlist write-ups, with redaction rules that keep candidate data out of the chat.
• Job Intake Form for Recruiters: Fill It From the Call With AI — A call-to-form method with a filled example, extraction prompt and confirmation email for ambiguous role requirements.
• AI for Recruitment Agencies: Re-Engaging Your Candidate Database — A practical sequence for cleaning a candidate database, drafting relevant outreach and recording replies without letting AI make hiring decisions.
• ChatGPT for Recruiters: Can It Prepare a Counter-Offer Call? — A prep workflow for agency recruiters: a de-identified case brief, candidate and client call scripts, objection handling and an ATS note.
• How to Recruit Tutors: Job Ads, Screening and Scorecards With AI — Recruit tutors fairly: an AI-drafted job ad, screening questions, a scored interview, reference and background-check reminders, and a bias check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Otter, Fathom and Fireflies pricing pages and Fireflies' storage help article; Greenhouse Notetaker page and Ashby pricing page (read 1 November 2026; Fireflies prices from a late-October check; plans change, so confirm before buying). This is practical guidance, not legal advice.