Customer Proof for AI Search: How to Write Results ChatGPT Can Repeat
Short answer: AI assistants such as ChatGPT and Perplexity recommend brands they can justify in a sentence. The proof that does this best is a named customer result: one plain sentence saying who, where they started, where they ended up and how long it took. Put it on the service page that produced it, and link to evidence anyone can check.
“Client X reduced support wait times by 40%.”
You’ll see sentences like that on more and more service pages, because businesses have worked out that AI assistants favour brands that show results. They’re right about results. They’re wrong about Client X.
Nobody can check Client X: not a buyer, not a journalist, and not an AI model deciding who to name. A result with no name, no date and no link is a claim. When an assistant narrows a long list of candidates down to three, claims are the easiest thing to leave out.
This guide covers how assistants decide who to name, which kinds of proof they can actually use, a template for writing your own, and two named results you can check yourself.
How does ChatGPT decide which brands to recommend?
An AI assistant recommends a brand when it can find a reason to name it. Asked “what’s the best X for Y”, it gathers candidate pages, narrows them to a shortlist, and then explains its choice. No AI company publishes the full process, so treat the diagram as a working model rather than a spec.
Most sites lose at the last step. Being retrieved isn’t enough: the model needs a sentence on your page that justifies putting you on the list. “We help teams improve support” gives it nothing to work with. “FileSeal went from zero AI citations to #1 in ChatGPT, ahead of DocuSign” does.
What kind of customer proof can AI repeat?
AI can repeat proof that works as one sentence with no context: a named customer, a measurable change and a timeframe. The test is simple. If a model copied that one sentence into its answer, would it make the case for you on its own?
| Format | Can AI repeat it? | Why |
|---|---|---|
| Named result on the service page | Yes | Name, before, after and timeframe in one sentence that anyone can check. |
| Anonymous result (“Client X cut wait times by 40%”) | Weakly | Specific, but nobody can verify it, so it reads as a claim. |
| Generic promise (“We help teams improve support”) | No | Every competitor says the same thing. |
| Logo wall | No | A logo image carries no result and no sentence. |
| Result only in a downloadable PDF | Unreliable | It lives in a separate file from the service page that gets retrieved for the question customers ask before choosing. |
| Testimonials loaded only by JavaScript | Often invisible | Vercel’s crawler study found none of the major AI crawlers render JavaScript, so script-loaded carousels may never be read. |
How to write a customer result AI can quote
Write each result as one sentence that follows a fixed shape, then put it in the page’s HTML next to the service that produced it.
The template:
[Customer name] went from [starting point] to [result] in [timeframe] after [what you did].
Before and after (illustrative; the brackets are where your real figures go):
| Before | After |
|---|---|
| “We help clinics grow with expert marketing.” | “[Clinic name] went from [X] to [Y] new patient enquiries a week in [timeframe] after we rebuilt their booking page.” |
| “Trusted by leading firms.” | “[Firm name] cut onboarding from [X] days to [Y] in [timeframe] using our [service].” |
Where it goes:
- On the service page, beside what you sell. A result on a separate case-study page helps that page. The same sentence on the service page helps the page that gets retrieved when someone asks who to hire.
- In plain HTML text. Not in an image, and not only in a script-loaded slider.
- With a link to the evidence: the full case study, a public result or anything else people can check.
- With the customer’s permission to name them. If you can’t name them, give as much as you can: industry, location, size. Know that it’ll carry less weight.
On schema: mark up the case-study page as an Article with a named author. Don’t count on self-written Review or AggregateRating markup about your own business; Google doesn’t show review snippets for self-serving reviews of that kind.
Does adding statistics and quotes help AI visibility?
Yes, according to the best-known study on the subject, with caveats. In the GEO paper (Aggarwal et al., KDD 2024), researchers rewrote web sources in different ways and measured how much of each source appeared in AI-generated answers.
- What worked best: adding statistics, adding quotations and citing sources. These raised visibility by up to 40% across a benchmark of about 10,000 queries.
- What didn’t: keyword stuffing, the classic SEO tactic, performed poorly.
- On a live engine: on Perplexity, adding quotations gave about a 22% lift on one visibility measure, and adding statistics up to 37% on another.
- Who gained most: lower-ranked sources gained more than top-ranked ones.
Two caveats. “Up to” is the best case, not the average. And the study measured how much of a source appeared in answers, not whether a buyer then chose the brand.
What the research supports is narrower but useful: concrete, checkable content gets used more than vague content. A named customer result is about as concrete and checkable as content gets.
Two named results you can check
Here are two of our own results, each written to the template above. One is a software company and one is a local practitioner, because the rule is the same at any size.
FileSeal: software, national competition
FileSeal went from zero AI citations to #1 in ChatGPT, ahead of DocuSign, in under 45 days. They ran the free SearchScore audit and did the work themselves.
You can open the ChatGPT answer and read the full FileSeal case study.
The Heart Shaman: one practitioner, one city
Across 160 AI searches, Ryan Jones of The Heart Shaman is cited in 83% of assistant answers and comes first in four out of five that rank anyone.
We built and optimised his site, so we’re not a neutral party here. The measurement method is in the full Heart Shaman case study.
What these examples do and don’t show
Neither result came from adding testimonials alone. FileSeal worked through a full fix list, including crawler access, structured data, answer-first pages and entity signals; The Heart Shaman’s came from a full site build. They’re here as examples of proof done well, not as evidence that a results sentence caused them. What makes them useful is that each is named and linked, so a model, a journalist or a buyer can check it in one click. “Client X” can’t offer that.
Where else should your customer proof appear?
Your proof should also appear on pages you don’t own. When someone asks for “the best” of anything, assistants often draw on pages that compare options: listicles, comparison guides, review sites and forum threads. A result that only exists on your own site asks the model to take your word for it.
Take the same sentence to the places that compare you:
- The people who maintain “best X” lists in your category. Send the result with its evidence link. A checkable result is easier for an editor to add than a feature list.
- Review platforms. Ask the customer behind the result to describe it in their own words.
- Industry publications and podcasts, where the customer tells the story.
- LinkedIn and community posts, written as the story of what changed, not as an ad.
Keep the wording consistent. When a model finds the same named result on your site and on independent pages, the claim is backed up by someone else.
Checklist: make your best result quotable
- Pick your three strongest customer results.
- Ask each customer if you can name them.
- Write each as one sentence: name, before, after, timeframe.
- Link each to evidence anyone can check.
- Put each sentence in plain HTML on the service page that produced it.
- Use the same sentence when you pitch list owners, ask for reviews and post on LinkedIn.
If your service pages describe what you do but never show what it did for someone named, start there. Then run a free SearchScore audit to see what else is stopping AI assistants from naming you.
Sources
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024
- Vercel, The rise of the AI crawler, 17 December 2024
- SearchScore, How FileSeal beat DocuSign in ChatGPT and the live ChatGPT answer
- SearchScore, The Heart Shaman case study
Frequently asked questions
Do testimonials help you get recommended by ChatGPT?
They can, if they contain a specific result. A testimonial that says "great service" gives an assistant nothing to repeat. One that names the customer and states a measurable change gives it a reason to recommend you. Put it in plain text on the relevant service page, not in an image or a script-loaded slider.
Should I name the customer or keep them anonymous?
Name them whenever they agree. A named result can be checked by a buyer, a journalist or an AI model; an anonymous one can't. If naming isn't possible, add what you can, such as industry, location and company size, and accept that the result will carry less weight.
Does review schema help AI search?
Structured data helps machines understand a page, so mark up case studies as Articles with a named author. Don't count on self-written Review or AggregateRating markup about your own business: Google doesn't show review snippets for self-serving reviews of that kind. Independent reviews on third-party platforms are stronger evidence.
Is a case-study page enough on its own?
Usually not. A case-study page helps itself, but buyers' questions usually retrieve your service pages and third-party comparison pages. Put a one-sentence version of the result on the matching service page and on independent sites, and link back to the full case study.