Keeping Your Cited Sources Defensible

Why the risk changes when AI quotes you

On a webpage, a claim sits in context: a reader can see the heading, the date, the surrounding argument and the footnotes. A model that lifts a sentence out of that context and repeats it in an answer removes the frame. A conditional claim becomes unconditional, a dated figure becomes current, and a summary becomes a statement of fact. The exposure is not that the claim was wrong; it is that the claim was true only under conditions that the quote dropped.

That changes what defensible means. The standard is not 'the number is accurate' but 'the number remains accurate in the form it can be quoted in, and there is a record of where it came from'.

• Write claims so that the conditions travel with them: the period, the sample, the scope and the unit
• Avoid phrasing that only makes sense after reading the paragraph above it
• Prefer a specific, bounded statement over a general one, because a bounded statement survives being quoted. The citation report shows which of your pages are being lifted, which is where a claim that only holds in context becomes a problem first.

Citing third-party data

Third-party figures are usable, and often the most credible material available. The exposure is in the details that get dropped: the date the data was published, the population it covers, the method behind it, and whether it has since been revised. When a model quotes the headline and drops the rest, the claim has been changed in substance.

• Record the source, the publication date, the population or sample, and the retrieval date for every third-party figure you publish
• State the coverage next to the number rather than in a footnote, so the qualifier is part of the quotable sentence
• Link the original source where licence terms allow it, and check whether the terms permit republication at all
• Note whether the figure has been revised since publication, and remove or re-attribute it when it has

Comparing competitors

Comparison pages are the most-quoted pages on a site, because a question about which option to choose is exactly the question people put to an assistant. The risk is asymmetric: a claim about your own product is a claim you can support, while a claim about a competitor is a statement about someone else's business that they will dispute.

• Compare on criteria you can evidence, and cite the evidence for each criterion rather than asserting a general ranking
• Date the comparison and state what it was based on, because both change. The recommendation check records which pages get cited in comparison answers, so you can see whether a comparison page is being read out of context.
• Avoid absolute superlatives without qualification; a qualified claim survives quotation, an absolute one invites challenge
• Keep a record of the source for each factual claim about a competitor, including the date you checked it

Publishing performance numbers

Performance figures are usually the most persuasive material you have and the least qualified. A number quoted without its measurement conditions becomes a claim about your business that you cannot later support, and it is the kind of claim that is picked up and repeated.

• State the measurement conditions in the same sentence as the number: the period, the baseline, the sample and the method
• Distinguish what was measured from what was observed, and never present a sample as a population
• Publish the methodology alongside the figure, so a reader can tell what the number does and does not cover
• Keep the underlying record, and be able to reproduce the calculation on request

The record itself is the deliverable

None of the above needs a heavy process. What it needs is a small, consistent record kept as the content is written rather than reconstructed later. A claim log with six fields per entry is enough, and it is what makes a disputed quotation answerable within a day instead of turning into an argument.

• Claim text, the page it appears on, the source, the date checked, the conditions that limit it, and who verified it
• Keep the log with the content, so it survives staff changes and redesigns
• Review entries when a source is revised or a figure ages past the period it describes. The query tool keeps the dated response, so a revised figure can be checked against what was said when it was quoted.
• Make the log available to whoever handles a claim dispute, including legal and communications

None of this is about being cautious in what you publish. It is about writing claims that remain true when a model lifts them out of the frame you wrote them in. Sources reviewed on 2026-10-20. This article records a method with verifiable boundaries; it does not claim that any search or AI platform will index, cite or recommend anything.