1. Clean the notes before AI ever sees them

This is the step everyone skips, and it’s the one that decides whether the final article is accurate. AI can’t tell the difference between a direct quote and something you paraphrased in a hurry. If you feed it messy notes, it will treat your guesses as facts.

Spend fifteen minutes on the notes first. Mark every direct quote clearly — I put them in quotation marks and add the speaker’s initials. Flag anything you’re unsure about with a question mark in brackets, like [did she say 2019 or 2021?]. Cross out anything you know is wrong rather than leaving it in to confuse the model later.

You’re not editing for style here. You’re building a fence around the facts. Everything inside the fence is safe for AI to use. Everything outside it isn’t.

2. Brief the AI like a junior writer, not a search engine

Don’t paste your notes and type “turn this into an article.” That prompt gets you a generic summary with your interviewee’s name pasted on top.

Instead, give it the same brief you’d give a junior writer: who the article is for, what the reader should take away, the tone you want, and the structure you have in mind. Mine usually looks like this: “Write a 900-word article for freelance writers based on these interview notes. The reader should finish knowing how a working journalist actually prepares for an interview. Keep the interviewee’s own words in quotes — don’t paraphrase quotes. Use subheadings. Don’t add facts that aren’t in the notes.”

The last line matters most. AI fills gaps by default; it’s helpful by design. Telling it explicitly not to invent anything is the single most important sentence in the brief. I also ask it to mark anything it had to guess, so I can check those spots myself.

3. Let AI build the structure, not the sentences

The first draft AI produces is scaffolding, not the article. I read it for one thing only: does the structure work? Does the opening earn the reader’s attention, do the sections flow in a sensible order, does the ending land?

Usually the answer is mostly yes, and that’s the point. Structuring is the part of this job AI is genuinely good at. It sees the shape of the material faster than I do, because it isn’t attached to any particular line.

But I almost never keep AI’s sentences. They’re too smooth, too balanced, too much like every other article on the internet. Once the structure is right, I rewrite the piece in my own voice, keeping the structure and the quotes and replacing nearly everything else. The AI draft becomes an outline with training wheels.

4. Protect the quotes — they’re the whole point

Here’s the rule I never break: a direct quote from the interview can only be shortened, never rewritten. I can cut words out of a quote with an ellipsis. I can choose a better starting point. But I cannot change the words, smooth the grammar, or “improve” the phrasing. The moment I do, it stops being a quote and becomes something I made up and put in someone’s mouth.

AI doesn’t understand this distinction, and it will happily polish a quote until it sounds like the model wrote it. So I check every quote in the final draft against my cleaned notes, word for word. It takes five minutes and it’s the most important five minutes of the whole process. If a quote reads awkwardly in the article, that’s the interviewee’s voice — and it’s usually the most interesting sentence on the page.

One practical trick: when I ask AI to draft around the quotes, I tell it to leave the quotes exactly as written and build the transitions between them. That division of labour works well — AI writes the connective tissue, the human being supplies the substance.

5. Add the one thing AI can’t: the room

An interview isn’t just words. It’s the pause before an answer, the laugh at a question, the thing someone said after you switched the recorder off. AI has none of that, and if you leave it out, your article reads like a press release.

For every interview article, I add at least one detail AI couldn’t know: how the person reacted to a question, what their workspace looked like, the story they told that didn’t fit the official narrative. It doesn’t have to be dramatic. “She laughed and said she’d tried that once and it went badly” does more for an article than three polished paragraphs.

This is also what makes the article unmistakably yours. Ten writers could interview the same person with the same notes. Only you were in the room.

6. Fact-check against the notes, not the draft

When the article is finished, I do one final pass with the notes open beside it. Not against the AI draft — against the original notes. I’m looking for three things: anything attributed to the interviewee that they didn’t actually say, any number or date that drifted, and any claim the article makes that the notes don’t support.

AI is remarkably good at inventing plausible specifics. A “five-person team” becomes “a team of six.” A project from “last spring” becomes “in March 2024.” These errors are small, plausible, and exactly the kind of thing that makes an interviewee email you the next day. The notes are the only source of truth, so check against them directly.

The honest truth

Interview notes are awkward raw material, and that’s precisely why they produce good articles. A blank page invites generic writing. A page full of someone’s real words, half-finished sentences and all, forces you to write about something that actually happened to someone who actually said it.

AI makes the middle of that job faster — the structuring, the first pass, the connective tissue. It doesn’t make the job easier in the ways that matter. You still have to protect the quotes, check the facts, and put yourself in the room. Skip those steps and you don’t have an interview article. You have a well-written guess.

Do it properly, though, and the result is something neither you nor the AI could have produced alone: a polished article that still sounds like the person you sat down with.

Frequently Asked Questions