A certificate can be genuine while the marketing claim wrapped around it is false. The certificate may cover one person while the copy accredits the whole company. One successful project may become a “typical result”. One fast attendance may become an “always within 60 minutes” promise. The evidence has not necessarily failed. The sentence has asked it to carry more weight than it can support.
Evidence has a load limit. Every word that expands the subject, scope, population, period or certainty adds weight. When the wording crosses the record’s boundary, putting the record beside the claim does not make the claim honest.
TL;DR
- Start with the exact sentence the buyer will meet, not the proof folder.
- Most evidence failures are boundary failures: subject, scope, population, period or certainty has been stretched.
- Match the proof to the job; quotation marks do not launder an objective claim.
- Keep a claim record with one decision, never a trust score, and make the limitation travel with the claim.
Start with the sentence, not the proof folder
Businesses often review evidence backwards. Marketing opens a folder containing certificates, reviews, case studies, screenshots and reports, then asks what impressive claims can be extracted. I start with the exact sentence the buyer will encounter. Copy the heading, body text, caption and qualification that shape its meaning. Then identify the named subject; the atomic proposition; the service, product, place or population covered; the applicable date or period; the level of certainty; the evidence being offered; the material condition; and the person responsible for maintaining the fact.
This discipline matters because claims travel. A sentence leaves its paragraph and enters a proposal, advert, search result, social image or generated answer without the contextual language that once made it defensible. Google’s people-first guidance asks publishers to consider clear sourcing, evidence of expertise and easily verified factual errors (Google Search Central: helpful content). Those are useful editorial questions. They do not produce a machine-trust score, and they do not absolve a business from checking the actual proposition.
Most evidence failures are boundary failures
A claim and its proof must concern the same thing. I test five boundaries.
| Boundary | Question | Typical failure |
|---|---|---|
| Subject | Does the record concern the company, one employee, one product or one project? | One director’s qualification becomes “our accredited team” |
| Scope | Does it cover the same service, geography or configuration? | Approval for one component becomes approval for the complete service |
| Population | Does one observation represent one case, a measured sample or every case? | One result becomes “customers typically save 30%” |
| Period | Was it valid then, and is it current now? | An expired status remains written in the present tense |
| Certainty | Does it support can, usually, will or guaranteed? | A possible outcome becomes a promise |
The document can be real. The issuer can be legitimate. The result can have happened. The public sentence can still be false because it changed one of those dimensions. This is why “we have evidence” is not a useful final answer. The useful question is: does this evidence permit these exact words?
Small words create large proof burdens
The most expensive language is not always flamboyant. Often it is short. Each of these words expands what the record must prove:
Suppose a dispatch record shows that an engineer reached one emergency call in 47 minutes. Subject to the quality of the system, that record may prove the timing of that call. It does not establish the percentage of eligible calls reached within an hour, the measurement period, excluded territories, rota conditions, future capacity or a contractual service level. The evidence supports neither “we always arrive within 60 minutes” nor “we cannot respond quickly”. It supports one bounded observation: the dispatch record for this call shows 47 minutes between the recorded request and recorded arrival.
If the company wants a general response-time claim, it needs a defined service standard and records capable of measuring the relevant population. My editing habit is to circle every word that makes a claim larger. Those circles show where the proof burden is hiding.
Match the proof to the job
An authoritative source can still be the wrong source for a proposition. An official register may establish the recorded status of a named holder. It does not prove customer satisfaction. A testimonial can report what one customer experienced. It does not prove technical compliance. A method can describe an intended process. It does not guarantee perfect execution.
| Claim type | Evidence must establish | Common wrong substitute |
|---|---|---|
| Identity | Current legal, trading or role fact | Copied directory profile |
| Qualification | Exact holder, issuer, scope and validity | Unexplained badge or logo |
| Service scope | Current inclusions, exclusions and delivery conditions | Positive testimonial |
| Method | Approved process, responsibility and relevant output | Promised result |
| Performance | Metric, denominator, method, period and limitations | Selected success |
| Comparison | Shared criterion, market, date and method | Confident preference |
| Customer experience | What the attributed customer reported | Universal objective fact |
I call this the claim-proof fit test. A piece of evidence passes only when its job matches the sentence’s job. Prestige does not repair mismatch.
Quotation marks do not launder an objective claim
“I enjoyed working with the team” is an attributed opinion. “The treatment permanently removed all damp” contains objective propositions about cause, effectiveness, completeness and duration. Permission to publish the customer’s words does not prove those propositions. UK CAP guidance states that objective advertising claims need documentary substantiation and that objective claims appearing inside testimonials still require sufficient objective evidence (ASA/CAP: substantiation; ASA/CAP: claims in testimonials). Applicability depends on the communication and circumstances; this is an editorial control, not legal advice. Keep three approvals separate:
- Is the testimonial genuine?
- Does the business have permission to publish it?
- Can appropriate evidence substantiate every objective claim inside it?
A yes to one does not answer the others. I would preserve the customer’s authentic opinion while removing or qualifying an objective claim the business cannot carry. That is not censoring the review. It is refusing to outsource the advertiser’s proof burden to a customer.
Make the limitation travel with the claim
A claim may appear accurate on its original page only because a distant paragraph supplies the missing condition. When the sentence is extracted, the condition disappears. Run a detachment test. Copy the claim and adjacent proof into a blank document. Remove navigation, earlier sections and footer disclaimers. Ask whether a cautious reader can still identify the subject, scope, evidence role and limitation.
Weak: “The treatment prevents future damp.” Stronger: “The specified treatment addresses the diagnosed moisture pathway; it does not prevent damp caused by unresolved leaks, condensation or external defects.” The stronger sentence is not timid. It tells the reader what the treatment is intended to address and what it cannot establish. The boundary survives when the sentence travels.
A qualification belongs beside the words it changes. A disclaimer three screens away is not a structural substitute for an honest sentence.
Want to know which sentences on your site are carrying more than their evidence can hold? We audit the consequential claims page by page and hand back a claim record, not a trust score.
Use a claim record, not a trust score
For every commercially consequential sentence, create one compact record: approved wording; rejected broader wording; evidence source and owner; evidence type; applicable scope and date; material limitation; public verification route, where appropriate; permission or confidentiality decision; maintained page; and the event that triggers review. End the record with one decision: publish, narrow, support, investigate, remove or leave alone.
Do not total the fields into “87% trustworthy”. One unresolved contradiction can invalidate the complete statement. A missing denominator and an expired certificate are different failures requiring different owners. An average hides the work. Leave alone matters too. An evidence review that always creates new content is a production target disguised as governance.
Public proof must also be safe
Not every valid record belongs on a website. Evidence may contain personal data, confidential terms, client information, security-sensitive material, unpublished pricing or details the company lacks permission to release. Narrow the public wording to what can be disclosed responsibly. Use a genuine controlled review route before purchase where one exists. Remove the claim when neither route can carry it.
Do not hint that secret evidence proves something larger than the reader is permitted to inspect. “Evidence available on request” is useful only when the evidence exists, the route works and the disclosed material actually supports the claim. Permission and substantiation remain separate. Permission to show a case does not make it typical. Permission to reproduce a certificate does not extend its scope. Redaction must not remove the condition that makes the evidence meaningful.
Search visibility cannot upgrade weak evidence
Clearer, crawlable and more extractable claims may be easier for people and systems to encounter. None of those qualities makes the claim true. An unsupported statement is not repaired by repeating it across more pages, surrounding it with generic trust badges, adding structured data later, publishing an unrelated case study, obtaining one answer mention or calling it a first-party fact. The sentence still has to fit its evidence.
This is where the AI-trust conversation becomes a distraction. We cannot inspect whether an answer system “trusts” a brand in the human sense. We can inspect whether the public proposition, proof and limitation agree.
Make the claim smaller before the campaign gets larger
Take the most consequential sentence on the page and run this seven-step audit:
- Remove decorative adjectives
Strip the words that add tone but no proposition.
- Name the exact subject
The company, one person, one product or one project.
- Circle every expanding word
Anything that widens scope, population, period or certainty.
- Split bundled propositions
One sentence often carries three claims.
- Match each proposition to evidence
The evidence must prove the same thing, not something adjacent.
- Attach the condition that changes the answer
Beside the words it qualifies, not three screens away.
- Choose one decision
Publish, narrow, support, investigate, remove or leave alone.
If the proof is narrower, narrow the sentence. If the record concerns another subject, replace it. If the evidence has expired, investigate or remove the present-tense claim. If everything already agrees, stop editing. I will sign a modest claim that survives inspection before a spectacular one held together by layout and hope.
Evidence has a load limit. Never ask one record to represent a population it did not measure, a person it did not qualify or a future it cannot guarantee. Claim-to-evidence fit, not vague AI trust, is the publishing standard a business can control.
Good to know
FAQs
Yes. A real certificate, review or result can be stretched across the wrong subject, scope, population, period or certainty. The document is true; the sentence has asked it to carry more than it measured.
Under UK CAP guidance, objective claims appearing in testimonials still need sufficient objective substantiation. Genuineness, permission to publish and substantiation are three separate approvals; a yes to one does not answer the others.
Short ones: all, every, always, fully, guaranteed, nationwide, independent, approved, typical, fastest, leading. Each expands what the record must prove. Circle them to find where the proof burden is hiding.
Only when the evidence exists, the route works and the disclosed material actually supports the public claim. Do not hint that secret proof establishes something larger than the reader is allowed to inspect.
Next step
Want claims that survive inspection?
We run the seven-step audit on your most consequential pages, narrow what the evidence cannot carry and leave you a claim record with an owner and a review trigger.