Trust signals

Trust signals, and what their absence costs you

Every other part of the audit asks whether your website is technically correct. This part asks whether a stranger would actually buy from it.

What a stranger checks before getting in touch

The trust analyser looks for four things: evidence that other people have already bought, a way to reach a human, reassurance about what happens if something goes wrong, and one obvious next step. None of it is broken the way a 404 page is broken. That is what makes it expensive.

Each absence is recorded at a different severity, and that ordering is the useful part. No reviews or testimonials anywhere on the page is critical. No phone number is a warning. No address, no About page, no payment methods and no visible price are informational. In the score a critical carries a weight of 20, a warning 8, an informational finding 3.

Missing is not the same as at fault

A software product or a remote agency has no shop to walk into, so a missing street address there is not a fault. The analyser first decides whether the page reads as a digital business — subscription wording, pricing plans, a demo request — and skips the address and phone findings only when nothing suggests visitors are meant to turn up: no opening hours, no store locator, no directions. A local business missing them is still flagged.

The same restraint applies to the next step. Judged on retail vocabulary alone, ordinary service sites came back with a false “no call to action”, so the wordlist covers “book a call” and “get in touch” as well as “add to cart”, and a form or a tel: link counts as an action even when the wording is unusual.

The phone check is the clearest case. The pattern this project used to run matched 9,726 phone numbers on one real shop — dates, map coordinates, product IDs. Sites with no phone number scored as though they had one, and the genuine finding never fired. It now counts digits, accepts between 7 and 15 of them, and rejects dates, decimals and identifiers. A check that cannot run at all is reported as unmeasured with the reason, never as a failure.

Not every finding is worth the same

A flat list puts “no H2 heading” beside “the pricing page is set to noindex”. Both may be tagged warning; only one is costing money. So each finding is ranked on two measured signals multiplied together: what it blocks, and where it was found.

What it blocks. Stopping the page being found at all carries the heaviest weight, 3.0. Exposing data, or letting someone send email as your domain, is 2.6. Blocking the action the page exists for is 2.4. Trust findings — no reviews, no refund policy, no privacy policy — sit at 1.8. Tidying work is 0.7. Anything the engine cannot classify keeps a neutral 1.0 rather than being quietly promoted or buried.

Where it was found. The same defect is weighted 1.6 on a pricing or checkout page, 1.4 on a contact page, 1.3 on the homepage, and 0.5 on a privacy or terms page — required, but nobody buys from it. An informational finding is capped at 1.8 and can never reach the top of the list. Each ranked finding carries a sentence naming the signals that placed it, so any position is something you can argue with.

What this engine says about this website

The report published at /sample-report is this engine pointed at this site. Trust signals scored 66 out of 100, on three findings, and all three are the ordinary ones.

No customer reviews or testimonials — critical, and true. It is marked overruled by hand with one line: it stays true until there are real ones, and inventing them is the single thing this business will not do. No phone number found on the page — a warning, and it stands. No pricing visible — informational, and it stands too. A sample report in which the seller's own site passes everything is an advertisement, not a sample.

What is worth doing first

In the order the engine weights them: give people a way to ask a question before they commit — an email address, a form, or both. Put real customers' words on the page where the decision is made, not only on a third-party profile. If you sell to consumers in the UK or the EU, link a returns policy; that information is required before a distance sale in any case.

Naming a real person is the one signal here that cannot be scored. The engine sees whether an About page is linked. It cannot tell whether anybody is named on it. That part is read by hand before the report reaches you.

What the engine checks here

Each of these is a rule in the code, not a category heading.

Customer reviews or testimonialsThree ways to satisfy it: an embedded widget from a known platform (Trustpilot, Yotpo, Judge.me, Feefo, reviews.io, Okendo and others), Review or AggregateRating structured data, or testimonial wording in the copy. None of the three present raises a critical finding — the strongest severity in this category. The check reads the rendered page where one exists, because these widgets are usually injected by JavaScript after the HTML arrives.
A phone number a person could dialA tel: link is trusted outright. Otherwise a candidate must carry 7 to 15 digits, and is rejected if it is a date, a decimal, an identifier, a bare integer or a run of identical digits. Absent, it is a warning — unless the page reads as a digital business with no premises to visit.
Any direct route to a humanAn email address, a mailto: link, or a form. If none of the three exists, that is critical: a buyer with a pre-purchase question has no way to reach you, so they buy elsewhere. WhatsApp and live chat are recorded separately as an additional route, not a substitute.
A returns or refund policyDetected from links whose text or address contains return, refund or exchange. Missing, it is a warning: a first-time buyer has no stated way to undo the purchase, so all the risk of trying you sits with them. It is also required pre-contract information for consumer sales in the UK and the EU.
A physical addressLooked for in the visible text, and reported as informational rather than a fault, because an address also feeds local search. Suppressed entirely for a business showing digital signals and no physical-presence wording, so a remote team is not penalised for having no shopfront.
The supporting pages: About, shipping, terms, FAQChecked as links, each recorded separately. A missing About page is raised as informational — for a brand nobody has heard of, it is one of the pages a first-time visitor goes looking for before deciding.
Accepted payment methods, shownCard schemes, PayPal, Stripe, Apple Pay and Google Pay, Klarna and Afterpay, and the local rails a Pakistani buyer looks for — JazzCash, easypaisa, cash on delivery. Not showing them is informational: it does not stop the sale, it adds hesitation at the last step.
One obvious next step, and a visible priceNo call to action of any kind, worded or implied, is critical. A form or tel: link with no clearly worded action button beside it is informational, as is a page carrying more than 25 competing actions — usually a sign it is asking for too many different things. A page with no price or range anywhere is flagged informational too.

Questions

Our reviews are all on Google and Facebook. Do those count?

Not as an on-page trust signal. The check reads your page the way a visitor's browser receives it, and looks for a review widget, Review or AggregateRating structured data, or testimonial wording. If none is there, the finding is raised, because a first-time visitor on your page sees exactly what the engine sees. Bringing the reviews you already have onto the page is usually the cheapest fix in the whole report.

We work remotely and have no office. Will that be held against us?

No. The analyser decides whether the page reads as a digital business before it applies the address and phone findings, and skips both when nothing suggests visitors are meant to turn up anywhere. If it gets that wrong on your site, say so — every finding is checked by hand before the report is sent, and the reasoning is written down so you can point at the part that is off.

Do we have to fix all of them?

No, and the ranking exists so you do not have to. A finding on a privacy or terms page is weighted at less than a third of the same finding on a pricing page, and an informational finding is capped so it can never reach the top of the list. Most sites have two or three findings that matter and a long tail that does not. Separating those is most of the value.

Sources

Public and checkable. Nothing here rests on a number we cannot point at.

This is one of fifteen categories. You can read a real report this engine produced on this very site — including the findings it raised against us — or ask for an audit of yours.