How journalists use an AI note taker for interviews and field reporting

The reporting week this page is about

You do somewhere between four and a dozen recorded conversations a week. A sit-down interview in an office. A twenty-minute phone call with a press officer who talks fast. A doorstep that lasts ninety seconds and has to be captured before it ends. A conference corridor. A source who sends a voice note at eleven at night instead of writing anything down.

What you carry is a phone, and sometimes a notebook. What you do afterwards is scrub back and forth through the audio hunting the one sentence you remember. An hour of tape costs roughly three hours to transcribe properly, and most reporters do not transcribe properly. They do the ten minutes they think they need, then the story changes and the useful line is in the part they skipped.

The things that go wrong are always the same ones.

• The exact wording is gone. You have the sense of the quote in your notes, not the words. Paraphrase is fine for context and useless for a direct quote.
• You cannot find the moment. You know he said it about forty minutes in. Forty minutes in is a long place to look.
• Two voices blur together. A two-hander or a panel, and later you are not certain which of them said the sharp thing.
• Names and figures are unreliable. You wrote the surname down phonetically because you did not want to interrupt, and now you are guessing at the spelling.
• A month later somebody disputes it. You have a story, a notebook and a memory. What you needed was the audio at 41:17.

The quote has to survive a challenge

This is the whole argument for using a transcription app rather than a plain voice recorder. A recording proves the words exist somewhere in a ninety-minute file. A timestamped transcript proves where.

Every segment in a Noter AI note carries a timestamp, and tapping a line plays the audio from that exact moment. Checking a quote stops being a search. You find the sentence in text, tap it, hear it said, and either publish it or fix it. On a long interview that is the difference between verifying six quotes in ten minutes and verifying two of them and hoping about the rest.

Speaker labels do the second half of the job. The app tags who said what across the recording, so a two-hander or a panel comes back separated instead of as one wall of text. It marks the voices as distinct speakers rather than knowing their names, so you attach the names yourself once, at the top. More in can AI tell who said what in a meeting.

Then the part nobody selling transcription likes to say out loud. AI transcription gets things wrong. Surnames, place names, technical terms, figures with decimals, crosstalk, a strong regional accent. There is no published accuracy number for Noter AI, and you should distrust any tool that gives you one, because the number changes completely between a quiet office and a windy street. See how accurate is AI meeting transcription.

So treat the transcript as a fast, searchable index of the audio, not as the record itself. The audio is the record, and the rule that follows is short: never publish a direct quote you have not heard back. When you spot an error, fix it. The transcript is editable, and after correcting a mangled surname you can re-run the summary so your notes carry the right version instead of the wrong one.

Where interviews actually happen, and what captures each one

Reporting is not a calendar of scheduled video calls. Here is each situation you hit in a week and what handles it.

Under the table are the field habits that matter more than any setting inside an app. None of them are about software.

• Phone flat on the table, mic end toward the source. Not in your hand, not under a notebook, not in a jacket pocket. Two feet of distance costs you more than a noisy room does.
• Recording keeps running with the screen locked. You can put the phone down and forget it is there, which matters in an interview where a screen lighting up changes the mood.
• Constant noise is worse than voices. Air conditioning, traffic and cutlery on a hard table degrade a whole recording. Sit away from the vent and move the plate.
• If you can get their audio, take it. A press office recording or a fixer's file will beat your phone from the fourth row every time.
• Start early. Record through the small talk. The line you use is often in the first minute, before anyone settles into press-officer mode.

What you should get out of an interview, and how to shape it

A raw transcript is not a story asset. What you want out of the file is a short list of things you can act on the same day, and the app can return the same recording in several shapes, so pick the one that matches the job.

For interviews the shape worth setting up is a custom one. You describe the format once and every interview comes back in it: every quotable statement with its timestamp, every factual claim that needs checking, every name and title mentioned, and everything the source refused to answer.

The stock styles have their moments too. Minutes of meeting suits a council session or an AGM you are covering. To-do list turns a long interview into follow-ups. Custom instruction lets you describe the shape you want in plain words, which is useful for a first pass at the file note and useless as a substitute for writing the piece. The summary is a working document. The story is still yours.

The other thing worth setting up once is chat across your notes. You can question one interview, or every note you hold at once. "Who else mentioned the 2029 funding date?" across six months of interviews is something a filing system cannot answer and a search box only half answers. See asking questions across all your meeting notes.

A worked example. You interview a city councillor for forty-five minutes about a housing budget. What lands afterwards:

• Key bullet points covering what was actually said, which goes at the top of the story file so future-you knows what is in this interview without opening it.
• Quotable lines with timestamps, so the three sentences worth using are already located and each one plays back in seconds.
• Claims to check, the figures you cannot print without a second source. "The scheme is fully funded to 2029" goes on this list, not into the copy.
• Action items, pulled automatically. Call the developer. Request the committee minutes. Email the press office about the figure he would not confirm.
• Names and spellings in one place, so you confirm all of them in one email instead of three.
• A recap email, if you promised the councillor's office a note of what was discussed. The email output style writes it and you set the tone.

Reporting in a language you do not write in

This is where a general transcription app stops being enough, and it is the situation foreign correspondents, fixers and diaspora reporters are in constantly.

Transcription covers 60+ languages and detects the language on its own. You never pick one before pressing record, which matters because you often do not know what a doorstep will be conducted in until it starts.

The harder case is the one most tools fail. Real interviews switch language mid-sentence. A source in Cairo answers in Arabic and drops the English name of an agency into the middle of it. A Delhi interview runs in Hindi with English for every technical term. Tools that lock to one language when you press record transliterate the second language phonetically, so the agency name comes back as nonsense syllables, or they drop the words entirely. Noter AI tags language at the word level, so both halves of a mixed sentence come back written correctly. That is the sharpest reason to use it for this work. More in the best app for transcribing multilingual meetings and, for Arabic, the Arabic meeting transcription page. If you are weighing it against HiNoter for interviews that mix two languages, that comparison is here.

Once a note exists you can translate it into 18 languages: Arabic, Chinese (Simplified), Chinese (Traditional), Dutch, English, French, German, Indonesian, Italian, Japanese, Korean, Polish, Portuguese, Russian, Spanish, Thai, Turkish and Vietnamese. An interview conducted in Turkish can be read as an English summary by a desk editor who speaks no Turkish, with the original intact underneath.

Two honest cautions. Transcription reaches 60+ languages but translation output stops at those 18, so a rare source language will transcribe and may not translate. And a machine translation is not a verified quote. If you are publishing a direct quote from a language you do not read, someone who does read it still has to check the wording against the audio.

Source protection, said plainly

Transcription happens in the cloud. Your audio leaves the phone and is processed on a server. There is no on-device or offline mode. If you are handling a source whose safety or job depends on nobody knowing they spoke to you, that fact has to be part of your decision, and no marketing page should make that decision for you.

What the app does offer: data is encrypted in transit and at rest, and you can delete your data at any time. What that does not do is remove a copy from existence the moment you hit stop, or make a file immune to a legal demand served on a service provider. Any cloud tool has the same shape, including the ones that do not mention it.

So run two tiers, the way most reporters already run two notebooks. Run the same security questions against any tool you put a source into, including this one.

• Routine interviews. Officials, press officers, industry sources, anyone you would name in print anyway. Record and transcribe normally. There is nothing here to protect.
• Sensitive sources. Whistleblowers, people at risk, anyone whose identity is the story's biggest liability. Keep those off every cloud service, including this one. An offline recorder and manual transcription still exists for a reason.
• In between, reduce what the file reveals. Do not say the source's name on the recording. Use an initial or a codename during the interview and keep the mapping somewhere else entirely.
• Delete on a schedule, not when you remember. See how to delete a recording permanently, and set the retention window before the first interview rather than after publication.
• Check the policy, not the pitch. Where the recordings are stored and whether the service trains AI on them are the two questions to settle for any tool you put a source into.
• Lock the phone properly. The weakest point in most reporters' security is not the transcription service. It is a handset with no passcode, six months of interviews on it, left on a table.

Consent, recording law and the doorstep

Recording law is not journalism ethics, and both apply. Get the legal part right first, because that is the part with penalties attached.

Rules vary by country and, in the United States, by state. Some places need only one party to the conversation to know it is being recorded, which is you. Others require everyone to agree, and a recording made without that consent can be a criminal matter as well as unusable. Phone calls are stricter than in-person conversations in many places, and a call that crosses a border can be governed by both ends. Start with meeting recording laws by country, which is orientation rather than legal advice.

In the field, say it out loud and record yourself saying it. "I'm recording this, is that alright?" at the top of the file is ten seconds of work, and it sits inside the same timestamped audio as the interview, which beats a note in your notebook. On video calls the bot appears in the participant list, so it discloses itself by being visible. That is disclosure by default, not a legal opinion. If your jurisdiction wants explicit agreement, still ask.

Two things get conflated. Consent to record is separate from agreement about what is on the record, so settle attribution in words at the time. And permission to record is not permission to publish the audio. That second one is a conversation with your editor or your lawyer, not with an app.

A reporting week, step by step

1. 1Set it up once on Sunday. Add the Control Center widget so a recording starts in one tap from the lock screen. Sync your Google or Outlook calendar so scheduled interviews are already there. Write your custom output style: quotable lines, claims to check, names, refusals.
2. 2Monday, sit-down interview. Ask for consent on the record. Phone flat on the table, mic toward the source, screen locked. Keep taking handwritten notes, because they are your map of what matters, not your record of the words.
3. 3Monday afternoon, file it properly. Put the note in a folder for the story, not a folder for the month, and attach the real names to the speaker labels while the voices are fresh.
4. 4Tuesday, phone interview. If the source will move to Zoom, Teams, Meet or Webex, send the bot. If not, speakerphone plus a second device, and accept that the transcript needs more checking.
5. 5Wednesday, doorstep. One tap from the lock screen before you get out of the car. Ninety seconds later you have it. Then record a thirty-second debrief while walking back, covering what the audio could not: who was there, what the place looked like.
6. 6Wednesday night, the voice note. A source sends a two-minute WhatsApp voice message. Share it into the app from the iOS share sheet and read it as text in the morning.
7. 7Thursday, write. Open the summary for shape, then work from the transcript. Every quote you plan to use, tap the line and hear it before it goes into the copy. Fix errors in the transcript, then re-run the summary so your notes match the audio.
8. 8Thursday, chase the gaps. The action items are your follow-up list and the claims-to-check list is what you send to the press office. Both were written while you were still in the interview.
9. 9Friday, ask across everything. Before filing, put the question to all your notes, not just this one. "Which sources have mentioned this contractor?" often surfaces a line from three months ago that changes the framing.
10. 10Friday, tidy up. Export what the desk needs as PDF or Word, keep plain text for your own file, and delete anything past the retention date you set for it.

What it will not do for a journalist today

The honest list. Read it before you subscribe, not after.

• No video files, at all. MP4 and MOV will not import. Not footage a colleague shot, not a clip of a council livestream. Audio only.
• No links. You cannot paste a YouTube URL, a council webcast address or a podcast page. Whatever you want transcribed has to be an audio file on your phone.
• No PDFs or documents. It transcribes speech. It will not read a leaked report for you.
• No live captions. The words do not appear while the person is talking. The transcript arrives after processing, usually a few minutes. For text in real time during a court session, this is the wrong tool.
• No offline transcription. Recording works with no signal, so a basement or a rural road is fine for capture. Turning that audio into text needs a connection.
• It cannot record a phone call on an iPhone. No app can. See can an AI note taker record a phone call for the workarounds that do.
• Speaker labels are not identification. It separates the voices. It does not know speaker two is the deputy mayor. You attach the names.
• No published accuracy figure, and the errors are real. Names, technical terms, numbers and crosstalk are where it slips. Anything you quote, you hear back first.
• No integrations. Nothing pushes into your CMS, Slack, Notion, Airtable or a Zapier chain. You export PDF, Word, plain text or a rich-text copy and paste it where it goes.
• Cloud processing, no exceptions. The most important line on this list for anyone handling protected sources.
• It is not a legal or ethical adviser. Consent, attribution and the right to publish stay yours to get right.

What it costs when you are one freelancer

Noter AI is $9.99 a month or $49.99 a year, per person, not per seat. There is a free trial, so test it on one real interview before deciding, ideally a difficult one rather than a quiet office.

Per-seat pricing is built for sales teams and it lands badly on the way journalism is staffed. Otter.ai Pro is $16.99 a month. Notta Pro is $13.99. Fathom Premium is $20 a month, with team tiers at $19 and $34 per user. tl;dv Pro is $18 per user per month. Fireflies runs $10 to $19 per seat. Granola's Business tier is $14 per user per month.

Two things about that list matter more to a reporter than the headline rates. The first is that seats are the wrong unit for how a desk staffs a story: you bring in a stringer for one week, a fixer for two days, a freelance photographer for an afternoon, and on a per-user plan each of them is a licence somebody has to approve and then remember to remove. The second is metering. Several of those tools ration AI summaries by the month, which is fine until a heavy news week runs you out on a Tuesday and you are back to scrubbing audio by hand on deadline. There are no summary credits here, and no seat tier to buy: each reporter records from their own phone. Be clear that this means one subscription each, not a desk licence — five reporters is five subscriptions, about $50 a month, or about $4.17 each per month on the annual plan, with no shared workspace and no single invoice. For a freelancer the sum is simpler still: a year at $49.99 costs less than three months of most of the alternatives. The wider argument is on meeting notes apps without per-seat pricing.

It runs on iPhone and iPad, and on Android through Google Play, so it fits whatever handset you already carry: iOS or Android.