The Interview Agent: Turn One Phone Call Into a Month of Content

Your customers can smell AI slop

Somewhere around 2024, everyone's blog started sounding the same. "In today's fast-paced world..." You've read it. Your customers have read it. They skip it, and worse, they quietly downgrade what they think of the company that published it.

Meanwhile, the best content your company could possibly publish is already being spoken out loud, every single day, on the phone. Your service managers and top salespeople answer the same customer questions hundreds of times a year — with detail, with opinions, with the kind of specifics that only come from doing the job. And in most businesses, none of it is written down anywhere.

The fix for AI slop isn't writing harder. It's capturing the knowledge your people already give away for free on every phone call — and using AI as the transcriptionist, interviewer, and editor instead of the author.

Here's what you'll build: one 30-minute call turned into a blog post, four LinkedIn posts, and an FAQ page — all in your expert's actual voice. One sitting to set it up.

Step 1: Pick the expert and the topic

The expert is whoever customers trust most. Your best salesperson. Your service manager. The tech everyone asks for by name. Not the best writer — the best knower.

The topic is the 10 questions customers ask them most. Here's the 5-minute way to get that list: walk over and ask. "What are the ten questions you answer over and over on the phone?" They'll rattle them off without pausing, because they're sick of answering them. Write them down verbatim.

At a dealership, that list might look like: "What actually needs to be winterized?" "Why is my slide-out making that noise?" "Extended warranty — worth it or not?" Yours will be just as specific to your business. That specificity is the whole point. Nobody can prompt their way to it.

Step 2: The interview

Book 20–30 minutes with your expert. Two ways to run it.

Option A — record a call. Sit down together (or call them) and interview them through the 10 questions. Ask like a curious customer: "Wait, why does that happen?" "What do people get wrong about this?" Record it:

• iPhone: the built-in Voice Memos app. Android: your phone's Recorder app.
• On a video call: Zoom, Teams, and Google Meet all have a built-in Record button (Meet and Teams need a paid workspace plan for recording).

Get consent. Tell them you're recording and what it's for. Say it on the recording. If you're recording a phone call, check your state's rules — some require both parties to agree. Recording your own employee in a scheduled interview with their knowledge is the easy case, but say it out loud anyway.

Option B — the zero-recording version. Send the expert the 10 questions and ask them to answer each one in a voice memo on their phone, like they're explaining it to a customer. Then they text or email you the files. Some people give better answers alone in their truck than across a table. Either way works.

Step 3: Transcribe

You need the audio as text. Honest rundown of what actually works right now:

• Recorded on Zoom/Teams/Meet: they generate a transcript for you. Download it. Done.
• iPhone Voice Memos: on recent iOS versions, open the memo and tap the transcript icon — Apple transcribes it on-device. Copy the text out.
• ChatGPT: on a paid plan you can attach common audio files (like .mp3 or .m4a) to a chat and ask for a word-for-word transcript. It handles this well.
• Claude: as of this writing, claude.ai does not accept audio file uploads, so transcribe first, then bring the text to Claude. (Claude is a strong choice for the next steps — it's very good with long transcripts.)
• Dedicated tools: otter.ai or a meeting recorder like Fathom, if you already use one.

Before you paste the transcript anywhere: strip customer names and any PII the expert mentioned ("so I told Mrs. Henderson..."), and use a business-tier AI account with training turned off (Claude for Work, or ChatGPT Team — roughly $25–30 per user per month). Your expert's knowledge is a business asset. Treat it like one.

Step 4: The interviewer prompt — this is the agent part

Don't jump straight to "write me a blog post." The first pass of raw interview always has gaps — the expert assumed context, skipped a step, said "you know how it is." A good human editor would go back and ask. So make the AI do that first.

Start a new chat in Claude (or ChatGPT), paste the transcript, then paste this:

You are a sharp trade-magazine interviewer preparing this transcript for
publication. Your job right now is NOT to write anything.

Read the full transcript below. Then give me exactly 5 follow-up questions
to take back to the expert. Choose questions that:
- Fill in gaps where they assumed knowledge a customer wouldn't have
- Push for a specific number, example, or story where they spoke in
  generalities ("it depends" — okay, on what?)
- Chase the most interesting thing they said in passing but didn't explain
- Would make the final content noticeably better, not just longer

For each question, add one line on why you're asking it.

TRANSCRIPT:
[paste transcript here]

Take those 5 questions back to your expert. This takes 10 minutes — voice memo replies are fine. Transcribe the answers and paste them into the same chat with: "Here are the expert's answers to your follow-ups. Add them to your working transcript."

That loop — read, ask, fill the gaps, then write — is what makes this an interview agent instead of a summarizer. It's also where most of the quality comes from.

Step 5: The content prompts

Now the writing. Three prompts, same chat, one at a time.

Prompt 1 — the blog post:

Using ONLY the transcript and follow-up answers in this chat, write a blog
post answering the question: [pick one of the 10 questions, e.g. "Is an
extended RV warranty worth it?"].

Rules:
- Write in the expert's voice. Keep their actual phrasing, examples, and
  opinions wherever possible. If they said it a memorable way, use their
  exact words.
- 800-1,100 words. Short sentences. Plain language. Subheadings.
- Do not add facts, statistics, or claims that aren't in the transcript.
- At the end, under a divider, list anything you wrote that the expert did
  NOT actually say, labeled "ADDED BY AI - EXPERT MUST VERIFY." If you
  added nothing, say so.

That last rule is the anti-slop insurance. The AI tells on itself, and your expert checks exactly those lines.

Prompt 2 — four LinkedIn posts:

From the same transcript, write 4 standalone LinkedIn posts, each built
around ONE specific insight, story, or contrarian opinion the expert
actually voiced.

Rules:
- 100-180 words each. First person, as the expert.
- Open with the expert's actual point, not a hook formula. No "Unpopular
  opinion:" openers, no emojis, no hashtag pile - 2 hashtags max.
- Keep their phrasing. If a post needs anything they didn't say, flag it
  the same way: "ADDED BY AI - EXPERT MUST VERIFY."

Prompt 3 — the FAQ page:

From the same transcript, write a customer FAQ page covering every question
the expert answered.

Rules:
- Each answer 50-120 words, in the expert's voice, using their explanations.
- Order the questions the way a customer would think of them.
- Where the expert said "bring it in and we'll look at it," keep that -
  don't replace their honest answer with generic advice.
- Flag anything not from the transcript: "ADDED BY AI - EXPERT MUST VERIFY."

Step 6: The human gate

Nothing publishes until the expert reads it and signs off. Not skims — reads. This is a hard rule, and it takes them 15 minutes.

The edit pass rules:

1. Every "ADDED BY AI" flag gets verified or deleted. No exceptions.
2. Cut anything that sounds like nobody you know. If your service manager wouldn't say "it's important to note," it doesn't ship.
3. Let the expert add. Reading their own words back usually shakes loose one more story. That addition is often the best line in the piece.

This is why the result beats generic AI content and always will: the knowledge is yours. A competitor can copy your prompt in thirty seconds. They can't copy twenty years of your service manager answering the phone. That's not something anyone can type into a chat box, and it's the only durable edge in content right now.

Step 7: The cadence

One call a month. That's the whole system.

• First week of the month: 30-minute interview on the next batch of questions (the expert's list of ten regenerates itself — customers keep calling).
• Same prompts, same human gate.
• Output per call: one blog post, four LinkedIn posts (one a week), and an FAQ page that keeps growing.

Twelve calls a year is twelve blog posts, roughly fifty LinkedIn posts, and an FAQ page your competitors' AI can't write. Total expert time: about ten hours a year, all of it talking — the thing they were already doing on the phone for free. If you want the rest of the team using AI this deliberately, training that actually sticks covers the rollout.

Tool names, plans and prices in this guide were checked in August 2026. AI products change fast — verify anything before you buy it.

Where Hoven fits

I build a heavier-duty version of this for clients — and audit where else AI actually pays off in their business versus where it's just noise. Most clients start with a $999/day audit; most take 1 to 5 days. You get a prioritized plan plus a "don't build this" list, and if it isn't worth more than you paid, you don't pay. Details on the pricing page, or get in touch.