How recruiters use an AI note taker for screening calls and job interviews

The recruiting week this page is about

A recruiter's calendar is a stack of short conversations. Six to twelve screening calls. Two or three intake calls with hiring managers. A handful of debriefs. Maybe an in-person panel or a coffee meeting with a passive candidate you spent a month getting on the phone. Notes get typed live, while the candidate is still talking, and that costs you twice: you stop listening properly, and you still end up with three lines in the ATS that say good comms, 8 weeks notice, wants hybrid.

The gaps show up in predictable places. The hiring manager asks whether the candidate actually built the payment system or just maintained it, and you are not sure. You need one exact sentence for a submission summary and you paraphrase it into something the candidate never said. A candidate you rejected in March applies again in July and there is nothing on file to explain why they were rejected. The fix is not typing faster. It is capturing the conversation once, then pulling the same fields out of every single call.

Consent comes first, and it is not the same everywhere

Ask before you record, out loud, at the start, and get the yes inside the recording itself. Then move on. That is the whole rule and it takes eleven seconds.

Interviews deserve more care than an internal standup for one reason: the power in the room is uneven. The candidate wants something from you. Many of them are job hunting quietly and are nervous about a recording existing at all. So make refusing genuinely easy, and mean it when you offer.

The law also moves depending on where the two of you are sitting. Some places let one party in the conversation consent for everyone. Others require every person on the call to agree. Several countries treat a recording of an identifiable person as personal data with rules attached regardless of who consented. We cover the basics in do you have to tell people you're recording a meeting, and the country-by-country picture in meeting recording laws by country. If you recruit across borders, read the second one before you record your first call.

• A script that works: Before we start, I record my screening calls so my notes are accurate and I do not misquote you to the hiring manager. The recording stays with me and I delete it when the role closes. Are you comfortable with that?
• Wait for a real answer. Do not start the recording, then ask. Start it, ask, and let their yes be the first thing on the file.
• If they say no, say fine and take written notes. Do not negotiate. A candidate who felt pushed into a recording is a candidate who will not take your next call.
• Say what happens to it. Who hears it, how long you keep it, that it is not sent to the employer as raw audio.
• Panel interviews: ask the panel too, not just the candidate. Your colleagues are also on the recording.
• Never record a candidate's referee or back-channel call without telling them. Same rule, no exceptions.

Where interviews actually happen, and what captures each one

Recruiting is not one capture situation. In a normal week you will hit four or five. Here is what covers each.

What you should get out of a screening call

Not a transcript. A scorecard, with the same fields filled in for every candidate, so two people can be put side by side without you rereading two hours of conversation.

This is what the output styles are for. A recording can be reshaped into several formats, and for screening the useful ones are Custom (you describe your scorecard once in plain words and every call comes back in that shape), Email (the submission note to the hiring manager), and To-do list (your follow-ups, not the candidate's). Action items get pulled out on their own, which for a recruiter usually means send the JD, check notice period in writing, ask about the visa.

Write your custom style once, in normal sentences, and reuse it. Something like: Summarise this screening call as a scorecard with these headings: current role and level, why they are open to moving, evidence for each must-have skill with a quote, compensation expectation, notice period, location and work pattern, concerns, recommendation. That is it. No template to fill in by hand.

Six candidates a week without mixing them up, and quoting each one correctly

Put every call for a role in one folder named after the requisition, not the candidate. Then the shortlist is one folder and the comparison is one screen. Search runs across every note's title and AI notes, so a name you half remember from four months ago is one query away.

The part that changes how a debrief goes is asking questions across all your notes at once. Which of the five backend candidates has run Kubernetes in production, not just in a side project? Who asked about parental leave? Which candidates named a number above 90k? You get an answer pulled from the actual calls, with the notes it came from, instead of scrolling five summaries.

Quoting is where the recording earns its place. Every segment carries a timestamp and speaker labels, so you can see who said what, and tapping a line plays that exact moment back. When a hiring manager pushes back on a claim in your submission note, you play four seconds of audio and the argument is over. If the transcript spells a name or a product wrong, fix it in the transcript and re-run the summary so the corrected version flows through, which is how editing and re-running works.

One habit worth building: never paste a raw transcript into an ATS or send one to a hiring manager. Send the scorecard. The transcript is your source, not your deliverable.

Interviewing in more than one language

Agency recruiters and anyone hiring offshore run into this every week. A candidate in Dubai answers in English but switches to Arabic when they explain something they care about. An engineer in Bangalore moves between Hindi and English inside one sentence. A candidate in Madrid uses Spanish for everything except the technical nouns.

Most transcription tools make you choose a language before you press record, and that choice is the thing that ruins the file. Whatever you did not pick comes back written out phonetically, so a real answer arrives as nonsense, or the second language is dropped from the transcript entirely. You find out afterwards, when the recording is the only copy of what was said.

Noter AI tags language at the word level instead of locking a recording to one language, and detection is on for every recording. So you never choose, and a call that moves between two languages comes back with both rendered properly in one transcript. Transcription covers 60+ languages. That is the difference between a usable record of a bilingual interview and a wasted hour, and it is covered in more depth on the multilingual transcription roundup.

The second half matters if you are an agency. The interview happens in one language and the client hiring manager reads in another. A finished note can be translated into 18 languages: Arabic, Chinese (Simplified), Chinese (Traditional), Dutch, English, French, German, Indonesian, Italian, Japanese, Korean, Polish, Portuguese, Russian, Spanish, Thai, Turkish and Vietnamese. Screen in Arabic, send the scorecard in English, keep the original on file.

Fairness, recall bias, and how long you keep the file

Be clear about what a recording does and does not fix. It does not make your hiring fair. It does not remove bias from the questions you ask or the person you like more. What it removes is one specific problem: deciding between six people using memories of different quality.

Human recall in interviews is uneven in ways that are well documented. You remember the first candidate and the last one better than the middle three. You remember the person who reminded you of yourself. You remember your own conclusion far better than the evidence that produced it, so an hour later you defend a gut call with reasons you invented afterwards.

A consistent record narrows that gap. When every candidate has the same eight fields filled in from what they actually said, a debrief becomes a comparison of evidence. Disagreements get settled by playing the moment back rather than by whoever speaks with more confidence. If a rejected candidate asks why, or a claim about the process is made later, you have contemporaneous notes rather than a reconstruction.

That same file is also personal data about someone who does not work for you. Decide the retention window before you start recording, write it into your consent script, and hold to it. A rule as simple as delete recordings 30 days after the role closes, keep the scorecard is defensible and easy to follow. How long should you keep meeting recordings walks through how to pick a window. Notes are encrypted in transit and at rest, and you can delete any of them at any time, but nothing deletes them for you.

A realistic recruiting week, step by step

1. 1Monday morning, set up once. Connect the calendar you book screens in, Google or Outlook. Create a folder per open requisition. Write your custom scorecard style in plain sentences and save it. This is the only setup you do.
2. 2Intake call with the hiring manager. Record it. The must-haves they name in that call are the fields you will score every candidate against, and six weeks later it is the only proof of what senior was supposed to mean.
3. 3Screening calls, all week. Bot joins the video screens from the calendar. For in-person and cafe meetings, one tap on the Control Center widget and the phone goes on the table. Ask for consent first, every time.
4. 4Straight after each call, spend three minutes. Read the scorecard output, fix any name or product the transcript got wrong, re-run the summary so the fix carries through, and add your own recommendation line. Three minutes now saves twenty later.
5. 5File it. Drop the note in the requisition folder. Copy the scorecard into your ATS by hand or export it. Send the hiring manager the Email version, not the transcript.
6. 6Thursday shortlist. Open the requisition folder and ask across all of the notes: who has the must-have, who named a number, who is available inside the window. Build the shortlist from the answers.
7. 7Friday debrief. When a claim is challenged, open the note, tap the line, play four seconds. Decide on evidence.
8. 8When the role closes. Delete the recordings you said you would delete, and keep the scorecards if your retention policy allows it.

What it will not do for a recruiter today

This is the section that should decide whether you bother. All of these are true right now.

• No ATS integration. None. Nothing pushes into Greenhouse, Lever, Workday, Bullhorn, Ashby or anything else. You export the scorecard to PDF, Word or plain text, or copy it as rich text and paste it into the candidate record. If a one-click sync into your ATS is the reason you are shopping, stop here.
• No integrations at all, so no Slack, no Zapier, no HubSpot, no Salesforce. Sharing is export or copy and paste.
• Audio only. You cannot import a video file of an interview, and there is no screen recording. If your video platform gave you an MP4 and no audio track, this cannot read it.
• No live captions during the call. Transcription happens after the recording ends, not as the candidate speaks. If you want text scrolling on screen while you interview, this is the wrong tool.
• Processing happens in the cloud, not on the phone. For most recruiting that is fine. If your policy forbids candidate audio leaving your own infrastructure, it is a hard no.
• It does not score or rank candidates for you, and you should not want it to. It gives you the same fields from every conversation. The judgement stays yours, and so does the responsibility for it.
• It does not manage consent for you. No consent prompt, no per-candidate log, no automatic redaction of personal details from a transcript. You ask, and you keep the record of asking.
• Transcription is strong but not flawless. Heavy crosstalk in a panel, a bad phone line, or a very strong accent will produce errors, usually in names and product names. This is exactly why the transcript is editable and the summary can be re-run.
• No web or desktop app. Recording and reviewing happen on iPhone, iPad or Android.

Pricing when you are one recruiter, or five

Per-seat pricing punishes small recruiting teams harder than almost anyone, because a recruiting team is exactly the shape that gets charged the most: several people, each interviewing all day, each needing their own account.

Do the sum, and put every tool on the same basis. Fireflies runs $10 to $19 per seat per month. Fathom Team is $19 per user per month and Business is $34. tl;dv Pro is $18 per user per month. Granola's Business tier is $14 per user per month. Otter.ai Pro is $16.99 per user per month and Notta Pro is $13.99 per user. Five recruiters means five of whatever you picked: $70 a month on Granola, $90 on tl;dv, $95 on Fathom Team and $170 on Fathom Business, before anyone has recorded anything.

Noter AI is $9.99 a month or $49.99 a year per person, flat. No seat tiers, no admin console, and no jump in price when the sixth person joins. Five recruiters means five subscriptions, so the honest monthly comparison is $49.95 against $70, $90, $95 and $170 above. The annual price is where it stops being close: five recruiters on the yearly plan is $249.95 for the whole year, about $4.17 per person per month, which is less than a quarter of the cheapest per-seat monthly rate on that list. Nothing is negotiated and there is no team plan to get onto, because each person records from their own phone rather than from a shared bot licence. There is a free trial, so test it on a real screening call before you decide, ideally a hard one: an accented candidate, a noisy room, or a call that switches language.

Download it on iPhone and iPad or Android. Ask for consent on the first call, save one custom scorecard style, and see whether Thursday's shortlist takes less time than it did last week.