Trust signals are what make a model willing to attach its own credibility to naming you. They are a different problem from entity signals. Entity signals answer who is this. Trust signals answer why should I believe them. A business can be perfectly resolvable and still never get recommended, and this is usually why.

If your entity is not yet clean, start there instead: entity signals for AI search visibility is the checklist and it comes first in the dependency chain.

What a model is actually doing when it “trusts” you

It is not making a judgement about your character. It is reducing risk.

A generated answer that names a business is a claim the system is making on its own account. If that claim turns out to be wrong, badly out of date, or about a company that does not exist as described, the cost lands on the platform. Everything that follows is downstream of that.

So the systems behave in ways that look like trust:

That reframing is useful because it turns a vague goal into a list of things to build. You are not trying to seem trustworthy. You are trying to be low-risk to cite.

The 12 signals that matter, ranked

Ranked by how much they move outcomes, not by how easy they are.

Tier 1: corroboration

1. Independent third-party mentions. Other credible sources saying the same thing about you that you say about yourself. This is the strongest signal available and the slowest to build. It is also the one that separates businesses that get named from businesses that merely rank.

2. Review text on platforms you do not own. Detailed, specific, on Google, Trustpilot or a sector platform. The full treatment is in review platforms and trust signals.

3. Public record agreement. Companies House, the FCA register, the SRA, the CQC, your trade body. Independent, authoritative, machine-readable and frequently out of date because nobody remembers to update them. See public records and governance surfaces.

Tier 2: verifiable specifics

4. Named people with real biographies. A named author with credentials that can be checked gives a model somebody to attach expertise to. Content published by “Admin” or by a company name has nowhere for expertise to live. This is the cheapest fix on the list and one of the most effective.

5. Dates on everything. Published date, updated date, and dates attached to any claim. A result quoted without a measurement period is unusable to a careful system. “Citation share rose from 4% to 41% between September 2025 and August 2026” is citable. “We deliver outstanding results” is not.

6. Methods stated alongside numbers. A figure with a method behind it survives scrutiny. A figure without one gets discounted. This is why our agency evaluation methodology is published rather than summarised.

7. Prices, where you have them. Publishing figures is a trust signal in itself, because it is falsifiable and most competitors will not do it. Five of the seven UK agencies we track publish nothing at all. Ours are published.

Tier 3: consistency

8. Identical facts across every surface. Name, address, founding date, description. Covered in social platform consistency and audited from your off-site ecosystem.

9. Schema that matches visible content. Markup asserting something the page does not show is worse than no markup. It is detectable, and it is a Google manual action as well as a credibility problem.

10. A stable web presence. Domain age, consistent branding, no recent rename. Every rebrand resets part of the corroboration clock, which is worth knowing before you commission one.

Tier 4: transparency

11. Stated limitations. What you do not do, who you are not for, where your approach does not apply. Counter-intuitive, and it is the single most reliable way to be cited on a comparison query, because a model resolving “is X right for me” needs a source that answers both directions and almost nobody provides one.

12. Visible correction and contact routes. An email address for corrections, a named contact, a published editorial standard. Small, cheap, and it signals a source that can be held to account.

The four that quietly cost you

Worth naming, because businesses invest in the first two believing they help.

Awards you paid to enter. A model that can see the entry fee page reads the award as a purchase rather than an assessment. Not damaging, but not the trust signal the badge implies.

Follower counts. Almost no weight. Reach is not corroboration. What your social profiles are actually for in this context is consistency of profile data, which is a different job.

Testimonials with no attribution. “Excellent service, J.S., Manchester” is not evidence. If a client will not be named, the honest options are to say the client is anonymised and why, or not to use it. MarGen’s own testimonial markup is built and deliberately emits nothing until real named quotes exist, which is the same principle applied to ourselves.

Superlatives without measurement. “Leading”, “award-winning”, “best in class”. These are noise. They cost you because a page dense with unfalsifiable claims reads as lower quality overall, which affects how the rest of the page is treated.

Trust signals in regulated sectors

The bar is higher and the opportunity is larger.

Models are demonstrably more conservative on health, legal and financial topics, because a wrong answer causes real harm. That means fewer sources get cited, and the ones that do are the ones that make verification easy: named practitioners with registration numbers, regulator references, publication dates, clear scope statements, and explicit acknowledgement of where advice is individual rather than general.

The compliance requirements you already meet are trust signals in disguise. SRA transparency rules require published pricing and complaints procedures. FCA Consumer Duty requires clear, fair communication. CQC registration is a public, checkable record. These map almost exactly onto what a model needs. Most regulated businesses treat them as an obligation and bury them. Publishing them properly is close to free.

Full detail in GEO for regulated industries, SRA transparency rules and AI visibility and FCA Consumer Duty as a content opportunity.

How this differs from E-E-A-T

They overlap heavily and they are not the same thing.

E-E-A-T is Google’s quality framework, assessed largely at page and site level, and its purpose is ranking. Trust signals for AI citation are assessed at entity level across the whole web, and their purpose is deciding whether to name you inside an answer.

The practical difference: E-E-A-T improvements that live entirely on your own site can move rankings. Citation trust is dominated by what exists off your site, which is why a business can have excellent E-E-A-T, rank well, and still never appear in a generated answer. That specific failure is common enough to have its own page: good SEO rankings but no AI visibility. The overlap itself is covered in E-E-A-T and GEO.

An order to build them in

  1. Week 1. Named authors with real biographies on everything. Dates visible on every page.
  2. Week 2. Schema audited against visible content. Remove anything asserting what is not shown.
  3. Weeks 2 to 4. Public records checked and corrected. Companies House, regulator, trade body.
  4. Weeks 3 to 6. Every superlative on the site replaced with a specific, dated, sourced claim or deleted. This will shorten your copy and improve it.
  5. Month 2. Review process changed to ask for specifics rather than stars.
  6. Month 2 onwards. A limitations section added to every money page. Uncomfortable, effective.
  7. Months 3 to 12. Independent corroboration, continuously. The slow one. It never finishes.

For what this looks like delivered, the method is published and the outcomes are documented rather than described: an IFA from 0 to 31% citation share in 90 days.

AI Brand Trust Signals: Common Questions

What is the difference between entity signals and trust signals?

Entity signals answer who you are, so a model can resolve your business to one organisation. Trust signals answer why it should believe what you say, so a model is willing to attach its own credibility to naming you. Entity work is a prerequisite. Trust work is what turns a resolvable business into a recommended one, and the two need different tactics.

What are the strongest AI brand trust signals?

Independent corroboration comes first, meaning credible third parties saying the same thing about you that you say about yourself. Then verifiable specifics such as named people, dates, methods and figures. Then consistency across every public surface. Then transparency about limitations. Design polish, follower counts and awards you paid to enter do very little.

Do AI models actually assess trust?

Not as a judgement, but the effect is the same. Retrieval and generation both favour sources that are corroborated, specific and internally consistent, because those reduce the chance of producing a wrong answer. What looks like trust assessment is really risk reduction, which is useful to know because it tells you what to optimise.

Do reviews count as trust signals for AI?

Yes, and the text matters more than the score. Assistants read review content, so reviews that name the specific service, context and outcome give a model matchable evidence. A five star average with 200 reviews that all say great service carries less usable trust than 20 that describe what actually happened.

Does an About page affect AI trust signals?

More than most pages on your site. A named team with real biographies, credentials that can be checked, a founding date, a registered address and a plain statement of what the business does gives a model verifiable anchors. An About page of stock photography and mission language gives it nothing to work with.

Can trust signals be faked?

Some can, briefly. Fabricated reviews, invented awards and schema that claims ratings the page does not show are all detectable, and schema that contradicts visible content is a direct route to a manual action as well as a credibility problem. The signals that matter most, meaning independent third-party corroboration over time, are the hardest to fake, which is exactly why they carry weight.

How long do trust signals take to build?

Consistency fixes land in weeks because they are corrections. Specificity improves as fast as you rewrite pages. Independent corroboration takes three to six months minimum because it depends on other people publishing, and no budget compresses that meaningfully. Plan on a year before the trust layer is genuinely strong.

What is the fastest trust signal to fix?

Named authorship with real biographies. Most business content is published by Admin or by the company name, which gives a model no person to attach expertise to. Adding a named author, a genuine biography, credentials and a link between the two is a day of work and it changes how every article on the site is read.

Where to Go Next

The layer underneath this: entity signals checklist · building your brand’s entity signals · entity authority versus entity manipulation

Where trust signals live off-site: ecosystem validation for AI search · review platforms and trust signals · media distribution and entity reinforcement

Related frameworks: E-E-A-T and GEO · what content AI prefers to cite · AI visibility strategy

Have it delivered: MarGen is a UK GEO agency with published pricing, a published method and documented results.