Your Best Proof Is Trapped in Private AI Chats. Here's How to Make It Public.

Questions this guide answers

• Why doesn't my 'why we're the best' page get me recommended by AI?
• How do I get AI to recommend my specialized product?
• Do customer case studies help with AI search?
• Should I target every query with customer evidence?
• Isn't publishing customer quotes just manufacturing reviews?

Direct answer

There's a specific reason your "why we're the best" page does nothing in AI answers, and it's the same reason a genuinely great specialist product stays invisible: the model discounts anything you say about yourself, and your actual proof, the buyers who found you and loved you, happens in private.

A savvy buyer asks the AI a narrow question, gets pointed to you, gets a great outcome, and the whole win, plus the evidence of the win, vanishes into a chat log or an email thread the model will never read. So your best evidence exists, it just isn't public. The fix isn't to make a louder claim, it's to move the private conclusion your customers already reached into a public place the model can read, in their words.

The silent-evidence problem

When someone asks an AI which tool or vendor to use, the buyers who ask the narrow, specific question, the ones with real constraints, tend to find the right specialist and get exactly what they needed. But that exchange is private. It never becomes a public sentence the next model trains on, or the current one retrieves.

So the specialist keeps quietly winning the buyers who already know how to dig, and stays invisible to everyone who doesn't, because its wins happen where the record never reads. It's a survivorship problem: the corroboration layer only sees what got written down, so any market where the good decisions happen privately, which is most expert and specialist buying, is systematically under-represented against markets where people post their choices. The specialist isn't losing on quality. It's losing because its buyers are the type who don't write it up.

Why "why we're the best" pages don't move AI

The model decides what's true about you by corroborating across the sources it trusts, and it weights independent sources higher precisely because they aren't you (see AI cites consensus, not authority). Your own page, however well-written, is the single source the model trusts least about you, because you're the most biased witness.

That splits the thing you're competing on into two competitions that feel like one. "Best product" is judged by you and a handful of experts, and it doesn't decide the AI answer. "Best-corroborated answer" is decided by the public record, and it does. The model isn't judging quality, it's surfacing whatever the public record already said, most of which was written by someone other than your happiest customer. You can be genuinely best and lose the second competition without ever knowing why.

Where your proof lives, and whether the model can read it

Lay out where your evidence actually sits, and the problem is obvious: the strongest proof you have is in the two places the model can't see, and the place it can see is the one it trusts least.

The fix: publish the private conclusion, in the customer's words

You're not inventing a claim, you're transcribing one the market already makes in private. Take the conclusions your happy customers already reached, and told you on a call, in an email, in a chat, and get them into public in the customer's own words. The highest-leverage version isn't you writing the comparison, it's a customer-voice case study, because a case study in a customer's words is literally the private chat made public, and it corroborates in a way your own page can't, since it isn't you making the claim.

This is the same job PR does with earned media, a trusted third party saying the thing so the model reads it as independent evidence (see AEO for PR). A customer case study is the version of that you can start on today, without waiting for a journalist.

Point it at the contested queries, not the settled ones

Don't spread this across every query, most of it would be wasted. The generic "best X" query is usually settled toward the popular defaults, and no amount of customer evidence moves a query that's already converged on someone else. The narrow, specific queries, the ones where the models disagree, hedge, or name different options from one run to the next, are contested: consensus hasn't formed, and that's exactly where a specialist can still become the best-corroborated answer (see contested vs settled).

So aim your customer-voice evidence at the specific questions where you genuinely win, phrased the way a buyer would ask them. Write the narrow comparison the savvy buyer would have written if they'd bothered, the one they instead concluded privately and moved on.

How to actually do it

The private conclusions already exist, this is a capture-and-publish job, not a manufacturing one.

• Find the narrow-fit customers who already got the outcome. They're in your inbox and your call notes, the ones who said some version of "you were the only one that handled our specific thing."
• Capture the sentence they already said, or interview them for it. You're not asking them to invent praise, you're recording a conclusion they already reached and told you.
• Draft it in their words, not your marketing voice, and publish under their name where you can. A writeup that reads like the customer wrote it corroborates; one that reads like you wrote it is back to being a self-claim.
• Point each one at a specific contested query. One case study that nails "best X for [constraint]" beats five generic ones, because it gives the model quotable evidence for an exact question.
• Measure the settle. Re-run the contested query over a few weeks. The tell that it worked isn't a clean jump, it's the run-to-run variance collapsing toward you as the query settles in your favor.

What not to do

The whole move depends on the evidence being real. Don't fabricate reviews, don't put words in a customer's mouth, and don't mass-produce near-identical praise, engines weight consistency across sources, so a burst of suspiciously similar testimonials reads as manipulation and can cost you more than the silence did. You're surfacing a conclusion the market already reached, not manufacturing one. And if the private wins don't actually exist, no amount of published evidence fixes that. That's a product problem, and it's the one problem this can't solve.

A worked example

Take a representative case, a specialist data-pipeline vendor we'll call Northwind (not a real company). It kept winning the buyers who asked the narrow question, "best pipeline tool for [a specific constraint]," but on those exact queries ChatGPT and Perplexity named the popular default and never mentioned it.

Its happiest customers had all told the founder the same thing on calls: Northwind was the only option that handled their specific constraint. None of that was anywhere public. The founder interviewed three of them and published each as a short case study in the customer's own words, each pointed at the exact narrow question those buyers had asked. Over the following weeks, those queries started naming Northwind, not because it out-marketed the default, but because the private conclusion its customers had already reached finally existed somewhere the model could read.

See which of your queries are still winnable

The first step is knowing which queries are contested (winnable with evidence) versus settled (not worth the effort), and whether the public record has any of your customers' language in it at all. Start Free (free, no credit card) and SolCrys shows you where the engines mention you, which sources they cite, and how they describe you, so you can tell the winnable questions from the lost ones before you spend a single case study on them.

Talk to us if you want it run continuously, so you can watch a contested query settle after you publish the evidence.

Your customers already concluded you were the best answer to a specific question. The only thing missing is a public copy of that conclusion, in a place the model can read.

FAQ

Related guides