Your best AEO content ideas are hiding in your client calls (not in a keyword tool)

For months, I built my query lists from keyword tools. You drop in your topic, you get back a spreadsheet with 300 rows, you sort through it, and you tell yourself you've got a content plan. Except it's all guesswork. You're guessing what your audience searches for, guessing the words they use, and in the end you're writing for a tool, not for a human.

One day I realised the raw material was already there, right under my nose, and I was doing nothing with it. My call transcripts. Every week I spend hours with clients and prospects, and these people ask me exactly the questions I'm trying to guess with my tools. So I stopped guessing and started from what people actually say.

Why Your Calls Beat a Keyword Tool

When someone asks ChatGPT or Google in AI mode a question, the engine doesn't run a single search. It breaks the question into a bunch of sub-queries. Comparisons, use cases, pricing questions, the classic "is it worth it," the "X or Y for this situation."

SEOs call this fan-out. A seemingly simple question triggers a dozen searches in the background. And those sub-questions? Your clients ask them out loud during your calls. In the same order, with the same words. All you have to do is jot them down as the call happens.

Here's a concrete example. I'm on a call with a prospect in fashion who's hesitating about outsourcing their acquisition. In the same conversation he tells me, "how do I know if an agency is doing good work," "what does this usually cost," "are you better than a standard Meta agency," "and what if I handle this in-house." Those four sentences are exactly the fan-out ChatGPT will generate when a founder asks whether to hire an agency. Except I heard it for real, I didn't guess it.

The Workflow, Step by Step

Here's how I set it up. Nothing complicated, but you have to be disciplined about capture.

I record all my client and prospect calls in Granola. It runs in the background, takes notes, and most importantly it gives me a clean transcript at the end. Without that, everything else falls apart. No transcript, nothing to work with.

Then I go back through and pull out every question, every objection, every comparison, keeping the exact wording. This is the most important part. I don't rewrite it into "expert language." If the guy says "how do I keep from getting ripped off by an agency," I note it exactly like that. Not "how to assess the reliability of a provider." People don't talk like that, and neither do LLMs when they cite content that sounds human.

I group all of it around the central prompt a buyer would type into an AI engine. Something like "should I outsource my eCommerce acquisition." Around that prompt, I organise all the sub-questions I've collected.

I build out each prompt as a fan-out, meaning I list the queries that prompt is likely to trigger. Some come straight from my calls, others I fill in by putting myself in the engine's shoes.

Then I check which of these queries cite us today and which ones cite competitors. You ask ChatGPT, Perplexity, Google in AI mode, and you see who shows up. That's the moment you spot the gaps in your coverage.

And finally I create or update the content to fill those gaps, written in the client's words. Not my own jargon.

The Detail That Changes Everything, the Exact Words

Anybody can reproduce the fan-out with a tool. What makes the difference is where your list comes from, real buying conversations.

A keyword tool gives you an averaged-out, smoothed-over vocabulary. On a call you hear something else, the fear behind the question. The prospect doesn't say "what's the average ROAS on Meta." He says "how do I know if I'm burning my budget for nothing." Same topic, but the second one shows you right away where the guy hurts. And that's the exact phrase he's going to type into ChatGPT at midnight when he's stressing over his numbers.

I draw the parallel with retail. The best in-store salesperson doesn't know the spec sheet by heart. They've heard customers hesitate outside the fitting room a thousand times, and they know what to say. Your transcripts play the same role as that fitting room. You just have to re-read what you noted down.

It's the same logic I apply when I work on email subject lines that actually perform. The best subject lines come from what people say themselves, never from a brainstorm.

How I Stay Organised Without Drowning

One call is easy. Thirty calls a month gets unmanageable fast if you don't have a system. So I built myself a little flow.

All my Granola transcripts go into one place. Behind that, I have an agent that reads each transcript and extracts the questions and objections along with the answer I gave in the moment. That generates an idea bank that grows on its own, with every call. So I never start from a blank page, everything comes from what real people have already said.

On the citation-checking side, I don't do it for every query. I prioritise the prompts that come up most often in my calls, because if five prospects ask me the same question in a month, that means dozens more are typing it into an engine without ever talking to me.

One practical tip, keep a record of who cites you and who cites your competitors at a given moment. Re-run the test two months after you publish your content. That way you know whether your work actually moved the ranking, instead of guessing.

Where to Start This Week

If you want to test this without burning entire days on it, here's the smallest possible step.

Take your last three prospect calls. Not your happy clients, your prospects, the ones still on the fence, because they're the ones who verbalise the real objections. Re-read the transcripts and note the ten questions they asked, word for word. Group them around one or two central prompts. Go put those prompts into ChatGPT and see who gets cited. If it's a competitor and not you, there's your next article.

Do it once and you'll see how much richer it is than a tool export. Every call gives you material for three articles. All that's left is to sit down and re-read the transcripts.