People decide whether to believe a website before they finish reading the first paragraph, and they rarely change their mind afterwards. The decision runs on a handful of cues: who is behind this, can I check it, is it current, does it look cared for. The same cues drive Google’s quality raters and, increasingly, the AI systems that decide which pages to cite. Credibility is not a design style. It is a set of verifiable signals, and most of them are cheap.
Key Takeaways - Credibility is verifiability plus care. Stanford's research on thousands of users and Google's rater guidelines converge on the same signals
- Twelve signals in four groups: who is behind the site, whether claims can be checked, whether the site is current, and whether it is maintained
- Small errors cost more than they seem. Stanford found that typos and broken links hurt credibility "more than most people imagine"
- Currency is a credibility signal in its own right. Sites that show they were recently updated or reviewed are trusted more
What makes a website credible?
Two things: readers can verify what it says, and the site shows that someone competent is looking after it. Every credibility signal is a version of one of those two.
The most cited research on this is still the Stanford Web Credibility Project. Its ten guidelines are based on studies with more than 4,500 people. The first guideline is the whole idea in one sentence:
Stanford University Stanford Web Credibility Project, Guidelines for Web Credibility Make it easy to verify the accuracy of the information on your site. You can build web site credibility by providing third-party support (citations, references, source material) for information you present, especially if you link to this evidence.
Google’s Search Quality Rater Guidelines arrive at the same place from the other direction. Raters are told to find out who is responsible for the website. They look for About, contact and service information and research the reputation of the site and its authors. And they lower their trust assessment when they find factual inaccuracies. We covered that framework in E-E-A-T explained and its most important element in trustworthiness.
The twelve signals below merge the two lists into one you can audit.
Group 1: Who is behind this?
1. A real organisation with a real address
Stanford’s second guideline: “Show that there’s a real organization behind your site”, and the easiest way is a physical address. For a business this means a complete About page, an imprint or legal notice where required, and contact details that work. The rater guidelines name About, contact and service information explicitly as things to look for. How much is needed depends on the type of site.
2. Named, verifiable people
An author with a name, a photo, a bio specific to the topic and an external profile that matches. Stanford’s third guideline asks sites to “highlight the expertise in your organization” and to give contributors’ credentials. We built the full checklist in author pages that build trust. The rater guidelines go further: they instruct raters to research the content creator’s reputation, so the bio has to survive a search.
3. Consistency across the site and the web
The same company name, address and author identity everywhere: on the site, in structured data, on LinkedIn, in directories. Inconsistency is a credibility leak because it forces the reader to guess which version is true. Organization and Person markup with sameAs links makes the consistency machine-readable, which matters for the AI side of this article.
Group 2: Can I check this?
4. Sources that are linked and primary
Every statistic, quote and legal claim links to its original source, not to another blog. This is Stanford’s first guideline and the core of Google’s trust definition. It is also the signal most sites get half right. They do cite, but the citation points to a second-hand summary, or to a page that no longer holds the number.
5. Claims that match today’s reality
A source from 2021 that was correct in 2021 is not a credibility signal in 2026 if the number has changed. Verifiability includes currency. How fast that breaks is in the real cost of outdated statistics. A Harvard study found 25% of deep links in news articles dead. Pew found 38% of pages from 2013 gone a decade later.
6. Honest scope and limits
A credible page says what it does not know. “This applies to sites under 500 pages” or “we tested three of the six tools” reads as honesty; universal claims read as marketing. Stanford’s ninth guideline asks for restraint with promotional content, because hype and credibility pull in opposite directions.
Pick your three most visited pages and count the claims a sceptical reader would want to check: numbers, prices, dates, “studies show”. Then count how many have a working link to a primary source. The ratio is your verifiability score. Below half, credibility is leaking regardless of how the site looks.
Group 3: Is this current?
7. Visible, honest dates
A publish date and, when content actually changed, a visible “last updated” date, mirrored in datePublished and dateModified. Google’s guidance on dates asks for a prominent, labelled date and for consistency between what users see and what the markup says. Stanford’s eighth guideline states the effect on readers directly:
Stanford University Stanford Web Credibility Project, Guideline 8 Update your site’s content often (at least show it’s been reviewed recently). People assign more credibility to sites that show they have been recently updated or reviewed.
The parenthesis is the important part: reviewed recently. A page that was checked and confirmed current, with a note saying so, earns the credibility even if nothing needed changing. A page with a fresh date and stale content loses it twice, once when the reader notices and once when the rater does.
8. No time-relative language that ages
“This year”, “recently”, “the latest version” all become wrong silently. Absolute references (“in 2026”, “version 4.2”, “as of October 2026”) stay true or age visibly. Visible ageing is honest; silent ageing is not.
9. Working outbound links
Stanford’s tenth guideline is blunt: “Avoid errors of all types, no matter how small they seem. Typographical errors and broken links hurt a site’s credibility more than most people imagine. A dead source link is both an error and a broken promise of verifiability. Checking outbound links is part of currency, and it is the part that decays fastest.
Group 4: Is someone looking after this?
10. A site that works
No broken pages, no forms that fail, no 404s from the navigation, fast enough to feel maintained. Stanford’s tenth guideline includes “keep your site up and running”. Readers infer the care behind the content from the care behind the site.
11. Editorial transparency
How articles are researched, how sources are chosen, how often content is reviewed, how corrections are handled, and whether AI tools are involved. A short editorial policy answers all of it and signals that a process exists. Google’s guidance on AI-generated content calls disclosures useful where readers might wonder how content was made.
12. Restraint in design and promotion
Stanford’s ninth guideline asks for restraint with promotional content, and the sixth for a design that looks professional or is appropriate to the site’s purpose. Pop-ups that block the text, ads that look like content, urgency banners and stock imagery all signal that the site’s priority is not the reader. The rater guidelines treat design that deceives or obstructs the main content as a reason for low ratings.
Richard Roth Founder CiteRanger, SEO Strategist Of the twelve, the ones I see fail most often are 4, 5 and 9: sources that are secondary, numbers that are stale and links that are dead. They are also the three that nobody notices from inside the company, because the page looks exactly as it did on launch day. Everything else on this list is visible in the CMS. Those three are only visible if you go and check, which is why we built a tool that does.
How do AI systems read credibility?
Through the same signals, with extra weight on the machine-readable ones. AI answer engines and Google’s AI features need to decide, at scale, which sources are safe to quote. They cannot read the About page the way a human does, but they can see:
- Entity clarity:
OrganizationandPersonmarkup withsameAslinks that resolve to consistent external profiles (signals 1 to 3) - Citations: outbound links to primary sources that exist and support the claim (signals 4 and 9)
- Currency:
dateModified, absolute dates in the text and numbers that agree with other current sources (signals 5, 7 and 8) - Consistency: the same facts stated the same way across the site (signal 3 again)
Google’s documentation on AI features in Search says the same foundational practices apply as for Search overall, with no special markup required. The selection mechanics are covered in how AI search engines choose which sources to cite. The short version: verifiable and current beats authoritative and stale.
How do you audit and improve credibility?
Score the twelve signals per page, one point each, on your most important pages. Then fix in this order, because it matches both impact and effort:
- Currency (7 to 9): fix stale numbers, dead links and vague dates. Fastest to fix, most visible to readers, and the only group that gets worse while you wait. How often to update content gives the cadence.
- Verifiability (4 to 6): replace secondary sources with primary ones, add sources where claims have none, state limits honestly.
- Identity (1 to 3): complete About and contact pages, build the author page, add
OrganizationandPersonmarkup, align external profiles. - Maintenance (10 to 12): fix broken functionality, write the editorial policy, remove the promotional clutter.
Groups 2 to 4 are largely one-time work with a yearly check. Group 3 is a routine. A review cadence per content type, and continuous monitoring for pages with prices, statistics or legal facts. Otherwise the signal expires quietly between reviews.
Credibility compounds. A reader who verifies one number and finds it correct extends trust to the rest of the page. A reader who finds one wrong number withdraws it from everything, including the parts that are right. The asymmetry is why the currency group comes first: one stale fact undoes the other eleven signals.
Conclusion
Website credibility comes down to two questions asked in the first seconds. Can I check this? And is someone looking after it? Twelve signals answer those questions: who is behind the site, whether claims are sourced and current, and whether the site is maintained with care. Stanford’s research and Google’s guidelines agree on all of them, and most cost less than a day to put in place.
Start with the currency group on your top pages. It is where credibility leaks fastest, where readers notice first, and where a routine, rather than a redesign, fixes the problem.
Credibility on the page is one half. The other is being referenced by others, which earning authority without buying links covers. What the metric does and does not tell you about it is in Domain Authority.
Dead and drifted citations are the credibility signal most sites never check; the outbound link checker finds them across the archive.