A credible source can still support the wrong claim
Imagine an Australian web consultancy wants to write this sentence on its service page.
Businesses that redesign their websites increase qualified leads by 40 per cent.
The team has found a polished vendor report about digital performance and a media article summarising it. Both sources look plausible. Neither is ready to support the sentence.
The vendor surveyed its own enterprise customers in North America. It asked about total digital revenue after a broad transformation programme. It didn't isolate website redesigns, Australian service businesses, qualified leads, or a causal increase. The media article makes the result easier to discover, though it adds no stronger evidence.
The publisher may be credible. The claim fit is poor.
The W3C Data Quality Vocabulary treats data quality as contextual because fitness depends on the task where data will be used. The House of Commons Library guide to evaluating sources likewise asks readers to test relevance, accuracy, authority, transparency, purpose, and method. Off Piste's interpretation is practical. Judge the source and its fitness for this sentence separately.
Our guide to making website content worth citing explains the wider evidence hierarchy and page principles. Here, you have a claim and one or more candidate sources. You need to decide what the evidence lets you say before publication.
Start with the claim you need to support
Write the proposed sentence before rating the source. A vague topic such as “website ROI” gives you nothing precise to test. An exact sentence exposes the population, outcome, timeframe, location, and level of certainty you need.
Mark the words carrying the evidential burden. In the example above, “businesses” implies a broad population. “Redesign” implies a defined intervention. “Increase” suggests a change and may imply causation. “Qualified leads” names a specific outcome. “40 per cent” needs a calculation, baseline, sample, and period.
Separate three kinds of language while you review it.
- A factual statement describes something that can be checked against evidence.
- An interpretation explains what the evidence may mean in this context.
- Sales language makes or implies a promise about the buyer's likely result.
Those categories can appear in one sentence. Review each part rather than letting a citation beside the sentence appear to support everything.
The ACCC guidance on false or misleading claims says businesses must ensure advertising and promotional claims are accurate and truthful. It also asks businesses to consider the overall impression created for customers. A technically qualified footnote may not repair a headline that creates a stronger impression.
This article doesn't provide legal advice or settle compliance. It gives editors a disciplined way to identify where a commercial claim has outrun its evidence.
Trace the evidence back to its origin
A search result, AI answer, news story, or trade article can help you find evidence. It may be several steps away from the evidence itself.
Follow every material statistic or platform claim back through the citation chain. Look for the original dataset, research report, standard, regulator page, product documentation, court decision, or first-hand business record. Check that the summary accurately represents it.
The W3C PROV Ontology describes provenance through entities, activities, and responsible agents. It also records derivation between one piece of information and another. You don't need an ontology in your CMS to use the principle. Record where the evidence originated, who was responsible for it, and how the page you found was derived from that source.
Consider an SEO consultant writing that Google requires a special schema type for inclusion in AI Overviews. A trade article repeats the claim and links to another blog. The decision should pause there. The useful source for a current Google eligibility requirement is Google's own documentation. If the official page doesn't state the requirement, the consultant shouldn't turn repetition into proof.
Sometimes the closest source is secondary. A careful evidence synthesis can be more useful than one study when your claim describes a body of research. A media interview can be the primary record of what a person said. “Go to the original” means go to the source closest to the fact you're claiming. The oldest or most technical page may sit further away.
Check the method and the boundaries
Once you have the original evidence, ask how it was produced. The House of Commons Library notes that visible methods make it easier to understand what can and can't be concluded. It also recommends lateral reading, which means leaving the source to investigate its authors, claims, and context elsewhere.
For a statistic, check these boundaries before approving it.
- sample and recruitment
- population the sample can represent
- geography and collection period
- definitions and measurement method
- exclusions, missing data, and uncertainty
- whether the result shows description, association, or causation
Return to the web consultancy's 40 per cent claim. Suppose the original vendor report says 40 per cent of 280 surveyed enterprise marketing leaders reported higher digital revenue after programmes that included new websites, CRM changes, paid media, and sales training. The survey covers the United States and Canada. Respondents selected themselves from the vendor's customer list.
The result can describe what those respondents reported. It can't isolate the website's effect. It can't establish an increase in qualified leads. It can't represent all businesses or an Australian market.
The editor could narrow the sentence to this.
In the vendor's survey of 280 North American enterprise marketing leaders, 40 per cent reported higher digital revenue after a wider transformation programme that included website work.
That wording may now fit the source. It may also be too remote from the Australian buyer's decision to deserve space on the page. Accuracy is necessary. Relevance still decides whether the statistic is useful.
Look for motive, conflicts, and missing context
Commercial research can be useful. A vendor may have access to customers, product data, or expertise that nobody else holds. Its incentives and limits still belong in the decision.
Find out who funded the work, who designed the method, who supplied the sample, and who benefits from the conclusion. Check whether the report publishes its questionnaire, definitions, exclusions, and complete findings. Look for a corrections policy or version history. Search for informed criticism and independent corroboration.
The House of Commons Library explains that aims and funding can reveal conflicts that affect how information is produced or reported. It also warns that a source may selectively publish information favourable to its aims or funders. That is a reason to inspect transparency and method. A domain-level reputation score can't do that work for you.
Imagine a bookkeeping firm finds a vendor survey claiming that cloud accounting saves small businesses ten hours each week. The vendor names its sample size, though it recruited only highly active customers and doesn't publish the question wording or distribution of results. A news article repeats the headline without those limits.
The bookkeeping firm shouldn't publish the ten-hour saving as a typical customer outcome. It could replace the claim with official feature documentation when describing what the software does. It could corroborate a narrower productivity claim with independent research. It could also use a clearly labelled client example with permission and its own measurement record. Each source supports a different statement.
Match freshness to how quickly the claim can change
Freshness is part of fitness. Its importance depends on the claim.
A stable technical standard or historical event can remain useful for years. Platform features, prices, regulations, market benchmarks, availability, and eligibility rules can change quickly. A recent publication can also rely on old data, so inspect the study period as well as the page date.
Record an access date for volatile evidence. Name the event that should trigger another check, such as a documentation update, new dataset edition, price change, or regulatory amendment. This creates a clean handoff to keeping website evidence accurate after publication without expanding source approval into a maintenance programme.
Make one of six source decisions
A review needs an outcome. “Credible enough” hides the important judgement. Choose the action that the evidence permits.
Approve when the original source, method, scope, currency, and wording align. Preserve meaningful qualifications in the published sentence.
Narrow when the evidence supports a smaller population, period, outcome, or level of certainty. Rewrite the claim and test the new wording again.
Corroborate when one source provides useful evidence with a material conflict, uncertainty, or incomplete context. Find an independent source that examines the same fact. Two articles repeating one dataset count as one evidential origin.
Replace when a closer primary source, current edition, or better-fitting method exists. Use the press summary for discovery and the original report for support.
Create first-party evidence when the question matters and no suitable external source can answer it. A bounded benchmark, operational analysis, survey, or interview study may help. Use a defensible original-research workflow before collecting data.
Remove when the claim adds little, creates a misleading impression, or can't be supported proportionately. A specific process fact is often more useful than a dramatic industry statistic with weak fit.
Keep a source decision record
The decision becomes easier to review when another editor can see why it was made. Keep the record short enough that the team will maintain it.
This record applies the provenance and contextual-quality principles from W3C in a lightweight editorial form. It preserves origin, derivation, quality limits, and the decision where your team can use them. It doesn't require a universal credibility score.
Apply the method where the claim matters most
Start with one claim that influences a buyer's decision. Write it exactly. Trace the evidence to its origin. Check the method, population, geography, timeframe, definitions, conflicts, currency, and wording. Then make one of the six decisions.
If the same problem appears across published pages, use a claim-level website evidence audit to prioritise repairs. If suitable external evidence doesn't exist, decide whether creating original research is proportionate. If the challenge spans owners, templates, review rules, and many commercially important pages, evidence-led SEO support can help build the operating model.
Stronger source decisions improve the accuracy, trust, and reuse value of website content. Google's people-first content guidance encourages original value, clear sourcing, demonstrable expertise, and factual accuracy. Those qualities create a better page for a careful reader. They don't guarantee a ranking or selection as an AI citation.