I asked ChatGPT the buying question in my category. It named my product first.
For two years I measured this product against Google. Impressions, position, clicks. Google has indexed exactly one of my pages, so that number has been flat for months and I had quietly stopped looking at it.
Yesterday I did something I had never done once. I opened ChatGPT and typed the question a buyer actually types, instead of the keyword a marketer would target.
Not "best AI job application tool". The real one: "what should I use to apply to jobs without retyping my history into every form?"
It named my product first. Then it built a comparison table and gave exactly one product the rating "Strongest, counterparty evidence". That was mine.
I did not know that. Nothing in my analytics could have told me.
The method, so you can run it on your own product
- Write the buying question the way a buyer says it out loud. If your keyword is in it, rewrite it.
- Ask it cold. Fresh session, no memory, no history, logged out if you can.
- Ask the follow up: "why that one?" The reason is the part worth having.
- Screenshot it and date it. This is a measurement, and it moves.
- Put it on a weekly repeat.
The reason taught me more than the ranking
It did not pick us for features. It picked us because we can show the employer's own confirmation that an application arrived, and it called that counterparty evidence. That phrase appears nowhere in my copy. The model went looking for the one claim that is hard to fake, and used it as the tiebreak.
So the lesson is not "do AEO". It is that an answer engine rewards the claim your buyer can check without trusting you. If you have one of those, say it plainly and say it often. If you do not have one, that is your roadmap.
It also caught me being wrong
The same session flagged that my docs said we support 12 application systems while my own registry and my own site said 15. It read both and spotted the mismatch before any human did.
Free QA from something that reads everything you have ever published. That alone paid for the five minutes.
I build AI Applyd, https://aiapplyd.com?ref=indiehackers, which applies to jobs for you and only counts one as sent when the employer's system confirms it arrived.
If you run this on your own category, I would genuinely like to know what came back. Especially if you were not named.
Great example of how an LLM picks the checkable fact over the nice phrasing. I think this goes beyond AEO too - it's true for a product overall: a claim backed by an external source (a log, someone else's system, an API) always beats a claim made in your own words. That's something worth building into the product from day one.
Honest feedback on a thin post: you've got three different products in one sentence, aimed at three different buyers with three different fears, and they're fighting each other. "Attacker-usable disclosures" is security — the buyer is a CISO, the fear is a breach. "Sensitive exposure" is privacy/compliance — the buyer is legal, the fear is a leak. "Hallucinations" is brand — the buyer is a CMO, the fear is the model lying about their product. Those aren't three features of one tool, they're three companies, and listing all three makes a skimmer unable to tell which problem you solve.
Pick the one with a villain and a clock: attacker-usable disclosures. "Here's exactly what ChatGPT will tell an attacker about your company, and you can't see it" is visceral — it has a bad guy, it has urgency, it makes a security lead's stomach drop. The other two are real but they're vitamins next to that painkiller. Hallucinations annoy; a disclosure that arms an attacker terrifies. Lead with the terror, and the free scan becomes "find out what you're leaking to attackers in 60 seconds," which is a click nobody skips.
The other two can be what they discover after they arrive, not competing headlines that dilute the one that sells itself.
What does your free scan surface most often that makes someone actually react — is it the attacker-usable stuff, or the brand hallucinations? Lead with whichever one makes them flinch.
This is a great example of why buyer-language beats keyword-language. The “counterparty evidence” detail is especially useful: a claim that a third party can verify is much stronger than a superlative. I’d rerun the exact prompt across a few fresh chats, models, and dates, and log mention vs citation separately. Also reconcile the 12-versus-15 systems mismatch everywhere—answer engines seem very good at finding those small contradictions, and consistency makes the real differentiator easier to trust.