The Proof AI Is Looking For

The pattern I see in every review

I've run more than thirty AI visibility reviews now, across roofing, home services, professional services, healthcare, and nonprofits. Different industries, different budgets, different levels of marketing sophistication. And the same thing happens in almost every one.

I get on a call with the owner and ask what they do for a client. And out comes twenty minutes of genuinely impressive material — the inspection process nobody else runs, the certification that took two years, the job they fixed for free because it wasn't right, the number they hit for a customer last spring. Real proof. Specific, checkable, hard-won.

Then I go back to their website and almost none of it is there.

This is Signal 1 of the 8 Signals, and it's worth 20 of the 100 points on the AI Readiness Scorecard — the largest single category. It's also the one most businesses score worst on, and not for the reason you'd guess. It isn't that they lack proof. It's that the proof never made it out of their heads.

What an AI engine can actually weigh

When someone asks ChatGPT or Perplexity who's best at what you do, the engine isn't forming an opinion. It's assembling one from whatever it can read — your site, your Google Business Profile, your reviews, your citations, third-party mentions — and comparing what it finds about you against what it finds about your competitors.

That comparison happens in seconds, and it can only include evidence that exists in a readable form. Your twenty-year track record, if it's never been written down, weighs exactly nothing. A competitor with half your experience and a page describing it in plain language will outscore you every time.

The framework behind this signal is E-E-A-T — Experience, Expertise, Authoritativeness, and Trust. It's four separate things, and that's the part people miss. Most sites are strong on one of the four and silent on the other three, and each one is demonstrated differently.

Experience: the jobs you've actually done

Experience is first-hand. Projects you completed, problems you hit, what you learned, photographs of the actual work. It's the difference between "we have decades of experience" and a page describing a specific job, what went wrong on it, and how you handled that.

The generic claim is worth nothing to an AI engine because every competitor makes it. The specific account is worth a great deal, because it's evidence rather than assertion, and because no competitor can copy it.

The most useful thing you can publish here is a job that didn't go perfectly. Something went wrong, you found it, you fixed it, and here's exactly what happened. Businesses resist this instinctively. But a flawless record reads as marketing, and a recovered failure reads as true — to a skeptical buyer and to an engine weighing which source to trust.

Expertise: why you do it the way you do

Expertise is different from experience, and conflating the two is why sites end up thin. Experience is what you've done. Expertise is what you know — credentials, professional background, technical explanation, and original insight nobody handed you.

The credential part is easy and most businesses do half of it: a certification listed on an About page. What's usually missing is the explanation. Why does your process have that step in it? What does a homeowner or a buyer misunderstand about your category? What would you tell someone who's about to make an expensive mistake?

That's the material AI quotes, because it answers a question rather than describing a company. And it's the material you give away for free on every sales call without thinking of it as content.

Authoritativeness: who vouches for you

Experience and expertise are things you say about yourself. Authoritativeness is what other people say — industry credentials that can be independently checked, awards, association memberships, references to credible sources, and reputation that exists somewhere other than your own domain.

The checkable part matters more than it sounds. A credential nobody can verify is a sentence on a page. A credential with a link to the issuing body's own directory is evidence. When I publish that I've been a Certified StoryBrand Guide since 2020, I link to the StoryBrand Guide directory so it can be confirmed without taking my word for it.

This overlaps with how AI fact-checks you off-site, which is its own signal. The difference is that this one is about whether your credentials and affiliations appear on your pages at all — the off-site signal is about whether the rest of the web corroborates them.

Trust: the boring signals that decide everything

Trust is the least glamorous of the four and the one that quietly caps the others. It's accuracy, citations, transparent authorship, visible dates, real contact information, corrections when you get something wrong, and honest limitations when you don't know.

Every one of those is checkable by a machine in a fraction of a second. Is there an author on this page? Is there a date? Does the phone number match the one on the Google Business Profile? Are the claims sourced? Does the site ever admit uncertainty, or does every page read like a brochure?

The accuracy piece deserves particular attention, because it's where businesses lose points without ever noticing. If one page says you have a 4.9 rating and another says 5.0, that's not a rounding difference to an engine cross-checking you — it's an inconsistency, and inconsistency lowers confidence in everything else it found about you. Getting your facts to agree with each other is covered in entity clarity, and it's worth doing before you publish anything new.

The honest-limitations part is the one nobody expects to be an asset. Saying plainly what you don't know, or what your service doesn't cover, reads as credibility rather than weakness. It's also the thing almost no competitor is willing to do.

What this looked like on a real site

Advantage Roofing in Cincinnati is a good example, because they weren't a business with a proof problem. They were a business with a publishing problem.

Here's what they actually do on a roof replacement. Every inspection is recorded on video, and if there's rotted decking underneath, it gets filmed and shown to the homeowner so nothing is discovered later. After the crew finishes, a separate independent inspector runs a nineteen-point check — which changes how the crew works, because they know someone else is coming. Anything the inspector finds gets fixed on the spot. Then the property gets a fourth magnet sweep for nails. And a year later the owner personally comes back for what they call the birthday inspection, and fixes anything he finds himself.

That's why they can offer the warranty they offer. They know what's under the shingles because they checked four times.

Almost none of that was on their website. So when I asked the AI engines who had the best roofing warranty in Cincinnati, two competitors came up first, across every engine I tried.

We ran the review, did a recorded interview to pull the details out, and published the material as a structured set of pages. Within about an hour of pushing it live, the answers changed — Advantage Roofing came up first, across the board, for the warranty question.

Nothing about the business changed that day. The nineteen-point inspection had existed for years. The only thing that changed is that it became readable. The full case study is here.

Getting it out of your head

Here's the part that surprises people: the writing isn't the hard part. Extraction is.

If you sit down at a blank page and try to list everything that makes your business worth choosing, you'll produce about six bullet points and most of them will be generic. The good material doesn't surface that way. It surfaces when someone asks you a specific question and you answer it out loud, the way you would on a sales call.

That's the entire reason our review includes a recorded interview rather than a questionnaire. One hour of someone asking the right questions produces more usable proof than weeks of trying to write it yourself — and it produces it in your own words, which is what makes it sound like a person instead of a brochure.

You don't need us to do this. You need someone to ask you questions and write down what you say. A few that reliably produce something worth publishing:

• Walk me through what happens after a customer signs. What are the steps a competitor probably skips?
• Tell me about a job that didn't go right the first time. What did you do?
• What do people misunderstand about your category?
• What's the question you wish every prospect would ask you?
• What does your warranty or guarantee actually cover — and what doesn't it?

Answer those into a voice memo and you'll have more proof than most websites in your industry carry. The next step is structuring it so a machine can lift it cleanly, which is its own signal.

What I don't know yet

This is a new field and I'd rather name the open questions than pretend they're settled.

I don't know how long a published proof point keeps its weight, or whether recency matters as much in AI retrieval as it does in traditional search. I don't know how much of the E-E-A-T assessment happens live versus from what a model already absorbed in training. And I don't know how the balance shifts as more businesses catch on and everyone's pages get denser.

What has been consistent across thirty-some reviews: the proof already exists, it's almost never published, and publishing it changes what the engines say. I'll update this page as the rest gets clearer.

Frequently asked questions

Related reading

• The 8 Signals — the full framework, all eight scored categories, and what your total means.
• Entity Clarity: How AI Decides Who You Are — Signal 2, and why conflicting facts cost you trust.
• How to Write Content AI Can Quote — Signal 4. Once you've extracted the proof, this is how you structure it.
• How AI Fact-Checks You Off-Site — Signal 8, and where corroboration for your claims comes from.
• llms.txt, Schema & the Files AI Looks For — Signal 6. None of your proof counts if an engine can't reach the page.
• Advantage Roofing case study — the full version of the example above.

This is Signal 1 of 8. See the full framework →