You can turn an interview transcript into an article with AI in an afternoon. The risk is a quote your source never said.
This guide shows you how to lock quotes before AI drafts, pick Q&A or narrative format, and check every line against the transcript.
Get the speed without the misquote.
Key Takeaways
- Draft From Clean Verbatim: Draft from a clean transcript, and keep the full verbatim file and timestamps for checks.
- Pick the Format First: Use Q&A when the person is the draw and narrative when the idea is.
- Lock Every Quote: Mark quotes with timestamps before AI drafts, and never let it write inside them.
- Verbatim or Paraphrase: A quote stays word for word, or it loses its quotation marks.
- Verify Row by Row: Every quote must read verbatim or paraphrased in your table before publishing.
- Send Quotes, Not Drafts: Ask the interviewee to approve their quotes and claims, not your whole article.
Struggling with Clarity in your writing?
You're not alone. Many writers face this exact challenge.
Orwellix provides you with advanced writing tools specifically designed to overcome common writing hurdles. Our AI-powered platform helps you craft clearer, more engaging content with less effort.
Start for FreeThe Short Answer: Five Moves From Transcript to Article
To turn an interview transcript into an article with AI, let the AI handle the structure and the first draft, and keep every quotation out of its reach. The order matters more than the tool:
- Read the whole transcript yourself. Don’t skim the AI summary first. You need to hear the interview’s shape before anyone shapes it for you.
- Mark the five to ten strongest passages. These are the lines a reader would repeat. Note the timestamp next to each one.
- Choose the format. Q&A, narrative, or hybrid, based on whether the person or the idea is the draw.
- Let AI build the draft around locked quotes. It can reorder, summarize, and write transitions. It can’t touch the words inside quotation marks.
- Verify every quote against the transcript. Do this before the interviewee, your editor, or your readers see the draft.
Five Moves, One Rule: AI Never Touches the Quotes
This guide starts after the transcript exists. If you’re still planning the conversation, first run the interview with a question-to-claim matrix, or draft questions with the free interview question generator. If you want software that turns audio into a post for you, see podcast-to-blog tools compared. This article covers the editorial method that keeps any of those drafts accurate.
Pick the Transcript Style Before You Draft
Draft from a clean-verbatim transcript, but keep the full verbatim file and its timestamps for checking quotes. That one rule prevents most quote problems before they start.
Otter’s guide to interview transcripts separates three transcript styles. Each one keeps a different amount of what was actually said:
| Style | What it keeps | Use it for |
|---|---|---|
| Verbatim | Every word, filler, and pause | The record you check quotes against |
| Intelligent (clean) verbatim | Every meaningful word, no fillers | The copy you and the AI draft from |
| Edited | Smoothed sentences, fixed grammar | Internal summaries, never quoting |
Why the Edited Style Is a Trap
An edited transcript has already changed the speaker’s words. If you quote from it, you’re quoting the transcriber’s phrasing, not your source’s. That’s how an article ends up with a crisp, confident line that nobody actually said.
A clean-verbatim transcript avoids that. It removes “um,” “you know,” and false starts, but it doesn’t reword anything. Speech without the noise is easier for you and for an AI to read, and the words are still the speaker’s own.
Make a Drafting Copy, Keep the Record
Save two files before you write a single sentence:
- The record: The full verbatim transcript with timestamps. Never edit this file.
- The drafting copy: A clean-verbatim version with the same timestamps.
If your transcription service only gives you verbatim text, you can strip the filler words for the drafting copy with a free filler words remover. Then read through it once. Some “like” and “so” words carry meaning, and a tool can’t always tell which ones.
Choose Q&A, Narrative, or Hybrid Format
Use Q&A when the person is the draw, narrative when the idea is the draw, and hybrid when both are. Decide this before you prompt anything, because the format changes which passages you lock and how much the AI writes.
Proofed’s guide to interview-based articles describes the same three options. What it doesn’t give you is a way to choose, so here’s a decision rule:
| Format | Choose it when | Watch out for |
|---|---|---|
| Q&A | A known name or a personality piece | Long answers that need cutting |
| Narrative | An expert explains a method or trend | Paraphrase drifting from meaning |
| Hybrid | A strong voice and a useful idea | Intro claims the Q&A never backs |
A short interview with a founder people already follow works as Q&A. Readers came for the person. A 45-minute call where a support lead explains a new metric works as narrative. Readers came for the method, and the speaker’s words are the evidence.
One Excerpt, Three Formats
Here is a clean-verbatim excerpt from a fictional interview. Dana Ruiz and Lumen Desk are invented for this example, and so are the numbers.
[00:21:04] Interviewer: Why did you stop measuring first-response time?
[00:21:09] Dana Ruiz: We were hitting the target every week, and customers were still angry. I’d look at the dashboard and it was green, and then I’d read the tickets and it was red. What we found was that a fast first reply that doesn’t solve anything just starts the clock on a second, slower reply. So we switched to time-to-resolution, and the first month it looked terrible. It went from four hours to almost two days on paper. But repeat contacts dropped by about a third, and that’s the number the CFO actually cared about.
Q&A version. The question stays, and the answer is trimmed but not reworded:
Why did you stop measuring first-response time?
We were hitting the target every week, and customers were still angry. I’d look at the dashboard and it was green, and then I’d read the tickets and it was red. … So we switched to time-to-resolution. … Repeat contacts dropped by about a third, and that’s the number the CFO actually cared about.
Narrative version. You explain the idea, and her words prove it:
Lumen Desk hit its first-response target every week, and customers were still angry. “I’d look at the dashboard and it was green, and then I’d read the tickets and it was red,” says Dana Ruiz, the company’s head of support. Her team realized a fast reply that solves nothing only delays a slower second one. After the switch to time-to-resolution, the average looked far worse on paper, rising from four hours to almost two days. But repeat contacts “dropped by about a third,” Ruiz says, “and that’s the number the CFO actually cared about.”
Hybrid version. A short narrative frame, then the exchange:
Most support teams chase first-response time. Dana Ruiz, head of support at Lumen Desk, dropped it, and her dashboard looked worse for a month. She explains why.
Why did you stop measuring first-response time? We were hitting the target every week, and customers were still angry. …
Notice what changes and what doesn’t. The structure, the order, and the sentences around the quotes are different every time. The words inside the quotes are identical to the transcript.
Case Studies and Webinars
Customer stories almost always run as narrative, because the result is the draw. If you’re still preparing that call, use these case study interview questions. If your recording is a product demo rather than a conversation, it’s a different job: turn a webinar transcript into a resource page instead.
Lock the Quotes Before AI Touches the Transcript
A quote lock means every quotation you plan to use is marked as a protected string, with its timestamp, before any AI step. The AI may write around a locked quote. It never writes inside one.
Most guides tell you to “edit for clarity, but don’t change the meaning.” That instruction is where misquotes come from, because clarity edits feel harmless one at a time. Some guides go further and say you may rephrase direct quotes. Newsroom practice is simpler: a quotation stays verbatim, or it becomes a paraphrase without quotation marks. There is no middle setting.
AI Writes Around the Quote, Never Inside It
How to Lock a Quote
Mark each quote in your drafting copy with a numbered tag and its timestamp. Any marker works as long as it’s consistent. For example:
[[Q3 00:21:09]] I'd look at the dashboard and it was green, and then I'd read the tickets and it was red. [[/Q3]]
Then list every locked quote at the top of the drafting copy, in order. That list becomes the first column of your verification table later. Five to ten quotes is plenty for a 1,200-word article. If you lock twenty, you haven’t chosen yet.
The Five Rules Inside the Lock
- Cut filler and false starts only. Removing “um” or a restarted phrase is fine. Use an ellipsis (…) where you drop words mid-quote.
- Don’t change words. No swapped synonyms, no fixed grammar, no tidier verb. If the grammar distracts, paraphrase instead.
- Don’t merge two answers into one quote. Two sentences from minute 8 and minute 31 are two quotes, even if they sound like one thought.
- Don’t move a quote into a different context. A line answering “What went wrong?” can’t introduce a section about what went right.
- Reword it, and it’s a paraphrase. Drop the quotation marks and write it in your own words. Keep the attribution: “Ruiz says her team…”
Rule 5 is where AI drafts slip most. Swapping a few words in a quote isn’t a paraphrase. It’s a misquote. If you need a refresher on the difference, here is how to paraphrase in your own words instead of patching someone else’s sentence.
Publisher style rules point the same way. Proofed’s guide notes that AP style suggests only very minor changes to quotations, such as cutting “umms” and “aahs.”
The Interpretation Test
Accurate words can still mislead. NPR’s accuracy guidelines tell reporters to “guard against subjective errors.” When quoting or paraphrasing anyone, NPR says to consider whether the source would agree with the interpretation. It also asks that a quote reflect what the person was actually asked or responding to.
Apply that test to every locked quote and every paraphrase:
Quick test: Read the sentence before and after the quote. Would the speaker agree that this is what they meant? If you hesitate, go back to the timestamp and listen.
Three AI Prompts for an Interview Article
Three prompts cover the whole job: one to find the story angle, one to draft, and one to review. They work in any AI assistant. The wording around them can change. The quote rule inside them can’t.
Prompt 1: Angle and Selection
Here is a transcript. List the ten passages a reader would quote, with timestamps, and propose three article angles. Don't rewrite any passage.
Why it’s phrased this way: Asking for timestamps makes every suggestion checkable. “Don’t rewrite any passage” stops the AI from handing back a polished version you might mistake for the original. Compare its ten picks with your own five to ten. Where they overlap, you have your quotes. Where they don’t, read both passages again.
Prompt 2: The Draft
Write a 1,200-word narrative article on angle 2. Use only the quotes between the markers, word for word. Everything outside quotation marks is your paraphrase of what the speaker said and must stay true to it.
Why it’s phrased this way: “Only the quotes between the markers” limits the AI to your locked list, so it can’t lift a new line you never checked. “Word for word” states the lock plainly. The last sentence covers the other half of the risk: paraphrase still has to be accurate, even without quotation marks. Swap “narrative” for “Q&A” or “hybrid” to match your format decision.
Prompt 3: The Review
List every sentence that states a fact the transcript doesn't support.
Why it’s phrased this way: It asks for a list, not a fix. If you let the AI repair its own draft, it may smooth over the problem instead of showing it to you. This prompt catches invented numbers, stretched claims, and confident transitions such as “the change paid off within weeks” that no one said.
A list from the AI is a starting point, not a clean bill of health. The same habit applies to any AI editing: prompt for edits, not rewrites, so each change stays small enough to check.
Verify Every Quote With a Four-Column Table
The verification table is how you prove the quote lock held. Give every quote in the draft one row with four columns: timestamp, quote as drafted, quote in transcript, and status.
Status takes one of four values:
- Verbatim: Matches the transcript word for word, filler aside.
- Trimmed: Words were cut, and the cut isn’t marked or changes the sense.
- Paraphrased: Reworded, with the quotation marks removed.
- Cut: Dropped from the article.
Before publication, every row must read verbatim or paraphrased. A “trimmed” row gets fixed: restore the words, add an ellipsis, or turn it into a paraphrase. A “cut” row comes off the list.
A Worked Example
Say an AI pass on the Lumen Desk narrative tidied two quotes. The table catches both:
| Time | As drafted | In transcript | Status |
|---|---|---|---|
| 21:09 | “green … red” line | Same words | Verbatim |
| 21:09 | “dropped by a third” | “by about a third” | Trimmed |
| 21:09 | “the CFO cared about” | “actually cared about” | Trimmed |
| 21:09 | Fast replies delay fixes | Quote marks removed | Paraphrased |
Both trimmed rows look harmless. But “about a third” is an estimate, while “a third” sounds like a measured result. Dropping “actually” is smaller, yet it’s still an unmarked change inside quotation marks. Restore the missing words and both rows move to verbatim.
Check the Transcript Against the Audio
Your transcript can be wrong before the AI ever sees it. A 2024 study by Koenecke and colleagues examined OpenAI’s Whisper speech recognition as of 2023. It found that roughly 1% of audio transcriptions contained entire phrases or sentences that were never spoken. That finding covers one system at one point in time, not every transcription tool. Still, it’s a reason to replay the audio for any quote that carries your headline or a strong claim.
Check the Facts Around the Quotes
A correct quote can still contain a wrong claim. If the speaker says their churn rate halved, the quote is accurate and the number still needs a source. Google’s guidance on AI-generated content says it’s critical to manually fact-check and review all AI-generated content. For anything beyond the quotes, run a claim-by-claim fact-check on the draft.
Send the Interviewee Their Quotes, Not the Whole Draft
Send the interviewee their quotes and any paraphrase of a claim about their company or a third party. Don’t send the whole draft.
A full draft invites edits to your angle, your headline, and your structure, none of which are theirs to approve. A short list keeps the review focused on accuracy. It also gets answered faster, because checking eight lines takes a few minutes.
Send Their Words, Keep Your Draft
What to include:
- Every quote: Exactly as it will run, with a line of context for each.
- Paraphrased claims: Anything you say about their numbers, product, customers, or competitors.
- Name and title: Spelled the way they want it to appear.
This reuses the focused validation email from the subject-matter expert guide linked above. The difference is timing: you send it once the draft exists, so the quotes are final.
Subject: Your quotes for the [topic] article
Hi [Name],
Thanks again for the interview on [date]. Below are the quotes and claims we plan to use, with the context for each. Could you confirm they're accurate and that your name and title are correct?
1. "[Quote]" (context: [one line])
2. Paraphrase: [claim about your company]
We'd appreciate a reply by [date].
A source can correct a fact or flag a misheard word. They don’t get to rewrite a quote into something they wish they’d said. If they want to add a new point, treat it as a new quote with its own date. The editor, not the AI and not the source, decides what runs.
A 10-Point Pre-Publish Checklist
Run this list after the interviewee replies and before you hit publish. Each item takes seconds, and each one catches a different kind of mistake.
- Attribution: Every quote names who said it.
- Name and title: Spelled exactly the way the person wrote them, not the way the transcript heard them.
- No composite quotes: No quotation stitches together words from two different answers.
- Context intact: No quote sits next to a question or topic it wasn’t answering.
- Paraphrases checked: Each one passes the “would the speaker agree” test.
- Claims about the world sourced: Facts the speaker stated about markets, competitors, or research are confirmed by a second source.
- Headline honest: The headline doesn’t make a claim the speaker never made.
- Recording consent noted: Your notes show the person agreed to be recorded.
- Transcript archived: The verbatim file and timestamps are saved where your team can find them.
- AI use disclosed: If your publication expects it, say how AI helped.
Item 10 follows Google’s advice to give readers context on how content was created. Item 9 matters later. If someone disputes a quote six months from now, the timestamp settles it in a minute.
How to Cite an Interview Transcript in APA and MLA
For a journalism-style article, attribution in the text (“says Dana Ruiz, head of support at Lumen Desk”) is all you need. If you’re writing for a school or a journal, here are the two common rules.
APA: An interview you conducted that readers can’t access counts as a personal communication. APA Style’s page on personal communications says to cite it in the text only, with no reference-list entry, using initials, surname, and an exact date. Participants in your own research are quoted directly instead.
MLA: The MLA Style Center’s guidance on recorded interviews offers two options. You can add a works-cited entry that lists the interviewee, the word “Interview,” who conducted it, and the date. Or you can add one note saying all quotations come from an interview you conducted on a given date.
If the interview was published or posted online, cite the published version instead. Check the latest edition of either style guide before you submit.
Draft Around Locked Quotes in Orwellix
The method works in any editor. In Orwellix, the AI does the restructuring and you own every quote, because each change arrives as a tracked edit you accept or reject. Here’s the workflow:
- Import the transcript. Upload your clean-verbatim drafting copy as a TXT, DOCX, or Markdown file, up to 15MB.
- Mark the quotes you’ll keep. Add your numbered markers and timestamps around each locked passage.
- Find the angle in Ask Mode. Ask: “List the ten strongest passages with timestamps and propose three angles.” Ask Mode answers in the chat and leaves your document alone.
- Draft in Agent Mode. Ask: “Draft a 1,200-word narrative article on angle 2. Keep every marked quote word for word.” Agent Mode reads the full transcript and proposes the draft as tracked changes.
- Accept or reject each change. Pay extra attention to any change that lands inside a marker. The habit is the same one you’d use to review each AI edit before accepting it.
- Fill in the verification table yourself. Compare the draft against your verbatim record, row by row.
- Send the quotes to the interviewee.
- Export. Download the finished article as DOCX or Markdown for your CMS.
On Pro, Premium, or the trial, Agent Mode can also search the web. That helps you check a claim the interviewee made about the world, such as a market figure. It doesn’t check the quote itself. Only the transcript and the recording can do that.
Write smarter with Orwellix
Orwellix's AI capabilities help you craft clearer, more effective content.
Conclusion
Turning an interview into an article comes down to a few decisions made in the right order. Draft from a clean-verbatim copy and keep the verbatim record. Choose Q&A, narrative, or hybrid based on whether the person or the idea is the draw. Lock your quotes with timestamps before any AI step, prompt for drafts that work around them, and prove the lock held with a verification table. Then send the interviewee their quotes, not your draft.
Each step guards the same thing: the speaker’s actual words. AI is good at the slow work of an interview article, such as finding the angle, ordering the material, and writing transitions. It is also good at small, confident changes that turn an estimate into a fact or one answer into another. The quote lock lets you keep the first and catch the second.
As more interview content is drafted with AI, accurate quotes will matter more, not less. Sources will agree to talk to writers who quote them correctly. Orwellix supports this method by proposing each draft change as a tracked edit you can accept or reject, so the final call on every quote stays with you.
Frequently Asked Questions (FAQs)
1. Can AI turn an interview transcript into an article?
Yes. AI can find strong passages, suggest angles, and write a full first draft from a transcript. It shouldn’t rewrite the quotes, so lock them before you prompt and check every one against the transcript before you publish.
2. Should I use Q&A or narrative format for an interview article?
Use Q&A when readers come for the person, such as a well-known founder or a personality piece. Use narrative when they come for the idea, such as an expert explaining a method. A hybrid opens with a short narrative frame and then runs the questions and answers.
3. Can I clean up a quote from an interview?
You can remove filler words and false starts, and you can use an ellipsis where you cut words. You can’t change words, fix grammar, or combine two answers into one quote. If a quote needs rewording, turn it into a paraphrase without quotation marks.
4. How do I cite an interview transcript?
In a published article, name the speaker and their title in the text. In APA, an interview readers can’t access is a personal communication, cited in the text only with the person’s initials, surname, and the exact date. In MLA, you can add a works-cited entry for the interview or a note saying all quotations come from it.
5. How long should an interview article be?
There’s no fixed length, so let the material and your publication decide. As a working target, 1,200 words with five to ten quotes suits a 30- to 45-minute interview. A Q&A can run shorter, because the questions carry the structure.
6. Do I need the interviewee’s approval before publishing?
It isn’t always required, but sending their quotes for a check is good practice and catches misheard words. Send only the quotes and any paraphrased claims about their company, not the full draft. The interviewee confirms accuracy, and the editor decides what runs.
Try Orwellix for Free
Experience Orwellix AI Agent's capabilities on the free forever plan, with 25 AI credits a month and no card needed. Need more? Try Pro or Premium free for 7 days.
Start for Free





