How to Write a Claim an Assistant Can Quote
A quotable claim has four properties: it is specific (a number, a threshold, a named condition rather than 'many' or 'often'), it is self-contained (the subject is named in the sentence rather than carried by a pronoun), it is attributable (it says whose claim it is and what it is based on), and it is falsifiable (a reader could check it). Most uncited writing fails the first two. Most embarrassing citations fail the fourth — the claim was quotable, got quoted, and was not true. Write the sentence you would be comfortable seeing attributed to you with no surrounding context, because that is the only form in which it travels.
The instinct when writing for retrieval is to make claims bolder. That is half right and the dangerous half. Specific and bold are different properties, and the goal is the first.
The four properties
| Property | Fails as | Passes as |
|---|---|---|
| Specific | "Response times are often slow" | "Median response time exceeds two seconds on the free tier" |
| Self-contained | "It does not support this" | "The free tier does not support custom domains" |
| Attributable | "Studies show X" | "In our own testing across 20 prompts, X" |
| Falsifiable | "Best-in-class accuracy" | "Matched the reference answer in 17 of 20 runs" |
The hedge that removes the claim
Writers hedge for good reasons — uncertainty is real and overclaiming is worse. The failure mode is hedging that removes the content rather than bounding it. 'Results may vary depending on a number of factors' bounds nothing. 'This held across three of our four test accounts; the fourth had a different plan' bounds it precisely and is still quotable. Hedge by naming the limit, not by softening the verb.
Attribute to yourself, deliberately
A claim that says where it came from is more useful to a system deciding whether to cite it and more defensible when it turns out to be wrong. 'In our testing' and 'as of September 2026' cost six words. The alternative — an unsourced assertion in the passive voice — is the form most likely to be both uncited and indefensible.
Our own position on this
We sell AEO measurement, so we have an obvious interest in you believing that writing quotable claims matters. The narrow version we will defend: a vague sentence cannot be quoted, because there is nothing in it to quote. Whether a specific sentence gets quoted depends on a great deal we do not control. Treat the first as a constraint worth removing and the second as unresolved.
FAQ
Is this just writing well? — Largely, yes, and that is not a criticism of it. The AEO-specific part is the self-contained property: ordinary good writing relies on the previous paragraph, and a quoted sentence arrives without it.
Should every sentence be quotable? — No. A page of nothing but declarative claims is exhausting and reads as marketing. The rule applies to the sentences carrying the answer, which are a small share of the page.
What if we genuinely do not know? — Say that, specifically. 'We could not reproduce this reliably across runs' is an honest, quotable and useful sentence. A confident guess in its place is the thing that gets quoted and then has to be withdrawn.
Put this to work on your own website.
Fenn finds what your customers ask, drafts the articles and site fixes, and measures what ChatGPT, Claude, Gemini, Perplexity and Grok say about you — with every change waiting for your approval.