Does Your Site Show the Proof AI Systems Look For?
Reviewed August 2026.
Credential tokens, numbered reviews, featured-on links, and schema sameAs.
The Answer Patch Fix Report's Trust Signals pillar counts credential tokens such as Licensed, Certified, or DDS, one bucket if the page mentions numbered stars or reviews, and visible featured-on listing links. Schema sameAs with two profile URLs is a separate rule. Start with a free homepage score; proof-signal edits ship in the report.
When two pages both answer the question, assistants need a reason to treat one source as reliable. Google's people-first guidance talks about experience, expertise, authoritativeness, and trustworthiness as qualities of helpful content. The Fix Report's Trust Signals pillar scores three observable inputs: credential tokens in the HTML (Licensed, Certified, Accredited, DDS, and the rest of that token list), one reviews bucket if the main text matches a numbered stars/reviews/rated pattern, and visible featured-on listing links in the body. Schema sameAs, at least two profile URLs, is a separate rule. It does not score author bylines, case-study copy, years-in-business claims, or generic testimonials with no number, and it does not scrape ChatGPT for who already gets cited.
What does the Fix Report check for Trust Signals?
- Credential tokens in page text: Licensed, Certified, Accredited, DDS, MD, and the other tokens the extractor looks for, each match is one proof point
- Numeric reviews mention: one bucket if main text matches a number plus stars, reviews, or rated, a quote with no number does not count
- Featured-on listing links: visible directory or press links in the body, not a footer social-icon row
- Schema sameAs (separate rule): does Organization or LocalBusiness JSON-LD include at least two profile URLs?
- Not scored here: case studies, customer stories, years-in-business, generic testimonials without a number (Competitive Visibility); authorship; physical address (LocalBusiness in Structured Data)
Why do visible proof and sameAs links beat an implied reputation?
A clean opener and valid JSON-LD still leave a model asking why it should believe you. A credential token and a numbered reviews mention are strings a fetch can copy. Google's creating-helpful-content guidance is the public E-E-A-T frame. schema.org sameAs is a separate, machine-readable pointer at official profiles. This pillar scores those counters. It does not award points for a photo of a diploma the alt text never named, a "since 2009" line, or a case-study heading.
Two proof points is the pass band for the visible-proof rule, credentials plus the numeric-reviews bucket plus featured-on links, added together. One is a thin-proof finding. Zero is a medium finding: no visible proof signals detected. SameAs is a separate, lighter rule: two or more links pass; one is thin; zero is thinner. Case studies, customer stories, and years-in-business claims are Competitive Visibility proof signals, not this counter. The pillar score is not a Wikipedia ranking.
What does a site with copy but no receipts look like next to one with proof?
Imagine two roofers. Site A has strong service copy and a LocalBusiness block. The homepage never says Licensed or Certified. There is no "200 reviews" or "5 stars" sentence. There is no featured-on directory link. Schema sameAs is empty. Site B puts "Licensed" in the HTML and a "4.9 stars" mention in the same fetchable text, then links sameAs to a Google Business Profile and a trade-association directory. Neither site is a named customer. Neither example is a live result.
A buyer asks ChatGPT which roofer to trust after a storm. The assistant can quote a credential token and a numbered reviews line only if they exist as text. Site B supplied both, two counted proof points, which is the pass band, and a separate sameAs row with two URLs. Site A supplied a service list. The Fix Report would flag Site A for no visible proof signals and thin sameAs. It would leave Site B's proof-signals rule in the pass band. "Since 2009" and a case-study heading would not have helped Site A on this pillar. The recommended edit is to put existing credentials and a numbered reviews mention in HTML, not to invent five-star copy.
The pattern is the one the product is built to catch. Proof that lives only on a Google Business Profile, never on the site, is invisible to a page fetch. A testimonial with no number does not fill the reviews bucket. Illustrative example, not a real business or report result.
What counts as a visible proof point?
The proof-signals rule adds three kinds of input. Credential tokens in the page text, Licensed, Certified, Accredited, DDS, MD, DO, DMD, RN, PhD, Esq, CPA, each add one point when the token appears as a word. A numbered reviews mention adds one more bucket if the main text matches a number followed by stars, reviews, or rated ("4.9 stars," "200 reviews"). Visible featured-on listing links in the body each add a point. Two or more combined is a pass. One is thin. Zero is a medium finding titled "No visible proof signals detected."
That counter is not case-study copy, not a years-in-business line, and not a generic testimonial without a number. Those signals belong to Competitive Visibility's proof comparison, which looks at a wider set including customer stories and years-in-business. Do not hide the only Licensed mention in a PDF. Do not rely on a footer social icon row; the featured-on extractor ignores ordinary social-nav clusters so a Twitter glyph does not count as a listing.
The report will recommend surfacing two real, counted receipts. It will not draft fake star ratings, a fabricated license, or a featured-on link you do not have. If you have no numbered reviews yet, put a credential token you actually hold in the HTML. Empty proof is honest. Invented proof is a Structured Data and Trust problem waiting to happen.
Why does schema sameAs need two links?
sameAs is how Organization and LocalBusiness JSON-LD point at official profiles, Wikipedia, Wikidata, Crunchbase, a Google Business Profile, a licensing directory. The sameAs rule passes at two or more URLs. One link is a low finding. Zero is scored lower. Evidence is the count, not a claim that those profiles already rank you.
Relative sameAs paths fail a different pillar: Structured Data's absolute-URL rule. A sameAs of "/about" is not a profile. A sameAs that points at a parked LinkedIn company page you do not control is a fact you should not mark up. Answer Patch leaves Organization sameAs empty on its own site until real profiles exist. Follow that rule on yours.
The Fix Report can include sameAs in a schema snippet when the visible page already links those profiles. It will not invent a Wikipedia URL. The paid report's brand-authority section is extra context around entity presence. It is not a substitute for putting two real sameAs values in JSON-LD, and it is not a live citation tracker.
How does this relate to E-E-A-T without scoring authorship?
E-E-A-T, experience, expertise, authoritativeness, trustworthiness, is Google's language for why a source deserves to be believed. Author names, dates, and bios are part of that public guidance, especially on YMYL topics. This product has no authorship rule. Do not expect a finding titled "author bio missing." Bylines are still good practice. They are not this score, and they are not scored elsewhere as a dedicated check. Physical address belongs in LocalBusiness markup under Structured Data.
What this pillar can observe without guessing: credential tokens, a numbered reviews/rated mention, featured-on listing links, and sameAs as a separate rule. That is a slice of E-E-A-T, not the whole framework. Case studies, customer stories, and years-in-business claims are Competitive Visibility proof signals. Competitive Visibility may also show that a rival publishes more of those on a comparable page. That comparison is a different pillar.
We do not claim a proprietary E-E-A-T percentage. We do not invent a customer win-rate. The citations on this page are Google's helpful-content guidance and schema.org sameAs, the public documents, not a lab study of ChatGPT.
What should you do with a Trust Signals finding?
Put counted receipts in HTML. If you are licensed, use the word Licensed in a sentence a fetch can read. If you have reviews, put a numbered mention such as "4.9 stars" or "200 reviews" in the page text, a quote with no number does not fill that bucket. If you are listed in a directory, add a featured-on link in the body, not only in a badge image. Then add two sameAs URLs you control or that officially represent you. SameAs does not replace the proof-point counter.
The Fix Report drafts proof-signal copy and schema sameAs only from facts already on the page or in intake. Drafted sections are labeled. Human review before publish. If the page has no reviews, the draft will not invent them. If sameAs would be a guess, it stays omitted.
Start with a free homepage score if you want a teaser first. Copy-ready proof edits and schema snippets are what you buy. We improve the signals assistants are known to weigh. We do not control citations, rankings, or traffic.
What might a Trust Signals finding look like?
- Medium
No visible proof signals detected
- Low
Only one visible proof point
- Low
Schema sameAs has fewer than 2 links
Illustrative examples, not from a real scan.
Where do these claims come from?
Claims on this page cite the sources below. Numbers that are not in those sources are omitted.
Frequently asked questions
What are trust signals in AI search?
Trust signals are visible proof an assistant can fetch. The Fix Report's Trust Signals pillar counts credential tokens, one numbered stars/reviews/rated mention, and featured-on listing links, plus schema sameAs as a separate rule. It does not score author bios, case studies, or years-in-business, and it does not read ChatGPT citation lists.
What is E-E-A-T and how does it apply to AI search?
E-E-A-T stands for Experience, Expertise, Authoritativeness, and Trustworthiness, Google's language for why a source deserves to be believed. Answer Patch scores a slice of that: credential tokens, a numbered reviews mention, featured-on links, and schema sameAs. There is no authorship rule in this product. Physical address is LocalBusiness schema in Structured Data, not a Trust Signals rule.
Which trust signals does Answer Patch check for?
The scoring engine counts credential tokens in the HTML, one bucket if main text matches a number plus stars, reviews, or rated, and visible featured-on listing links. Two combined proof points is the pass band. Fewer is a finding. Schema sameAs, at least two URLs, is scored separately and does not fill the proof counter. Case studies, customer stories, and years-in-business are Competitive Visibility signals, not this pillar.
Other audit pillars
Read more on the blog
Also see: E-E-A-T
Get proof-signal edits in a Fix Report
Start with a free homepage score. No account. The $19 Fix Report includes copy-ready proof-signal edits and schema snippets when the visible facts allow them.