How journalists verify AI transcripts before a quote ships. A listen-back routine for names, numbers, and off-record lines.
On this page
Transcribe faster with File Transcribe
Upload audio or video, get speaker labels, timestamps, and editable text free to try.
How journalists verify AI transcripts is a listening routine, not a confidence score in an app. The model will hand you a clean paragraph that sounds like the source. Your job is to prove the sentences you plan to print.
This is the quote check. Tool choice for the draft is best transcription tools for journalists. Press conferences add a noisy room on top: best AI tools for press conference transcripts.
The rule
If a reader could challenge the sentence, you play the audio. If they could not, you can skim.
That usually means you listen to:
- Every direct quote
- Every number, date, and name
- Any denial ("we did not", "never", "no comment")
- The line before and after a quote, so you do not clip the condition off
You do not listen to the whole hour again unless the story depends on the whole hour.
Set the transcript up so checking is fast
Upload the original file so each line has a timestamp. On File Transcribe, use the homepage, keep speaker labels on, and rename the source before you hunt quotes. Clicking a segment should play that moment. If your tool will not play the source audio next to the words, switch tools before you edit.
Interview files: interview recordings.
A five-pass check
Do these in order. Mixing them is how a name stays wrong.
- Speakers. Is this the source, the spokesperson, or you? Fix labels first. Method: how to add speaker labels to transcripts.
- Proper nouns. Search for the names in your notes. The model will invent a near-spelling.
- Numbers. Search for digits and number words. Play each one. "Fifteen" and "fifty" are a classic swap.
- Negatives. Search for not, never, didn't, no. Dropping a "not" is a libel problem, not a typo.
- Quotes you will actually use. Play from a few seconds before to a few seconds after. Write what was said, including the hedge.
Then stop. The rest of the transcript can stay messy.
What AI gets wrong on news tape
| Failure | Why it matters |
|---|---|
| Dropped "not" | The meaning flips |
| Merged speakers | You attribute a quote to the wrong person |
| Smoothed grammar | The source sounds more certain than they were |
| Wrong title or agency | The story looks sloppy, or wrong |
| Filled-in word under noise | A fact appears that nobody said |
General accuracy factors, if you want the audio side: what impacts AI transcription accuracy. Faster cleanup habits: how to fix AI transcripts faster.
Off the record, background, and the cut-off sentence
The transcript does not know your ground rules. Before you quote:
- Mark off-record and background stretches in the transcript so a later editor does not lift them
- If the source trailed off or was interrupted, do not let the model complete the thought
- If you are paraphrasing, the transcript is your notes, and the audio is still the source
A clean paragraph is not permission.
When to send it to a human
Pay for a human transcript when:
- The quote will be disputed and you need a third party to have heard it
- The audio is evidence, not just color
- You cannot understand the line yourself, so you cannot "verify" it
Rev and similar services are the usual path. An AI tool that says "verified" because a second model agreed is still AI. Otter vs Rev vs File Transcribe.
Do not verify from a summary
Meeting bots and "AI notes" drop caveats. If the only text you have is a summary, you do not have a transcript. Get the recording and run it through a tool that shows words against time.
Summaries can draft a nut graf after the quotes are checked. They are not the check.
FAQ
How do journalists verify an AI transcript?
Listen to every quote, number, name, and negative you plan to publish, using timestamps on the original audio. Skim the rest.
Is a high accuracy percentage enough?
No. The errors that matter are rare and serious. A score will not flag a missing "not" in the sentence you print.
Should I send sources the AI transcript to approve?
Only if your outlet already does quote checks that way. If you do, send the lines you will use, after you have listened, so you are not asking them to edit the model's mistakes into a new meaning.
What if two tools disagree?
Play the audio. The recording wins. If you still cannot tell, do not quote that line.
Does File Transcribe certify quotes?
No. It gives you a draft and playback. You certify the quote by listening.
Check one quote the slow way
Upload the interview on File Transcribe. Pick the sentence you want in the story. Play it twice. If the words match, use it. If they do not, the rest of the transcript needs the same treatment on every line you print.
More guides
- Transcription guides
- Try File Transcribe free
- Transcript format guide
- Best transcription software
- Transcribe Zoom meetings
Further reading
Written by

Rasif Ali Khan
Founder, File Transcribe
I made File Transcribe to turn recordings into editable text without extra steps. I write these guides from the workflows I use myself, like meetings, podcasts, lectures, and the rest.
Related articles
All posts →How to Tag Themes in Interview Transcripts (2026)
How to tag themes in interview transcripts in 2026. A practical coding workflow after AI transcription, from speaker cleanup to quote verification.
Best Otter.ai Alternatives for Journalists (2026)
Otter.ai alternatives for journalists who need interview transcripts without a meeting bot in the source call.
Best Transcription Tools for Journalists (2026)
Best transcription tools for journalists in 2026. Interview WAVs, phone M4As, and when human Rev still beats AI self-edit.
Best User Interview Transcription Software (2026)
Best user interview transcription software for 2026. Ranked for UX research sessions and prototype tests as files, with speaker labels and export for analysis.