Two referees, one copy deck
The content that wins AI citations has a recognizable accent: direct answers, superlatives, decisive comparisons, the “best X for Y” sentence engineered to be quoted. UK merchants writing it answer to a second referee: the Advertising Standards Authority, whose CAP Code governs marketing communications including a brand’s own site copy, and which has spent decades adjudicating exactly the claim types AEO encourages. The lazy resolutions both fail. Pure aggression collects complaints and rulings that become permanently citable bad press; pure timidity produces hedge-copy no engine quotes. The working merge is scope discipline, and it happens to produce the most extractable copy anyway.
The code itself is public, the CAP and BCAP codes with their rulings archive, and three of its doctrines do most of the work for AEO purposes: objective claims require substantiation held before publication, comparisons must be verifiable, and obvious puffery, subjective exaggeration no consumer reads literally, sits outside the substantiation duty.
Claim patterns, risk, and the rewrite
| The AEO-flavored claim | ASA posture | The compliant version that still gets cited |
|---|---|---|
| ”The UK’s best-selling [category]“ | Objective claim; substantiation required before you publish | The actual figure with source and date, or the claim scoped to where it is true |
| ”Number one for [use case]“ | Objective unless clearly framed as opinion | ”Rated 4.8 from 2,100 UK reviews” or a named award with year |
| ”Better than [competitor]“ | Comparison; must be verifiable on stated criteria | The criterion-level comparison with the data shown |
| ”The most luxurious feel imaginable” | Likely puffery; no substantiation duty | Fine as is, but engines rarely quote it, so it earns little |
| ”Clinically proven” | High bar; evidence to the claim’s strength | The study, the scope, the result, linked |
The pattern in the right column is the same one that wins retrieval: bounded, sourced, checkable. Google’s AI features guidance describes engines grounding in verifiable content, which means the substantiation the ASA wants in your files is, published properly, the citation bait itself.
The AI-era wrinkles worth taking seriously
Three newer questions sit on top of the classic doctrine. First, AI-generated copy: responsibility under the code follows the advertiser, not the tool, so scaled generative product descriptions need the same claim review as human copy, and “the model wrote it” is no defence. Second, the quoted-claim problem: when an engine lifts your FAQ sentence into an answer, the consumer meets your marketing claim in a new context; the prudent posture is writing every extractable sentence as if it will be read standalone, because it will be. Third, overlays and dynamic content: AI-personalized banners and offers are still marketing communications, and the rules on misleading omissions, conditions attached to “free” and “sale”, apply at whatever speed the personalization runs.
This is the same family of discipline as the substantiation regimes covered for green claims in shielding your brand from eco-washing flags and for age-gated categories in AI vice filters: the regulator’s floor and the engine’s preference keep converging on bounded precision.
Write the citation hook like a lawyer who wants to win
The much-imitated AEO pattern, an FAQ answering “what is the best X” with your own brand, is writable compliantly, and the difference is visible in the verbs and the scope. Attribute the judgment to its real source: review aggregates, named third parties, specific awards. Scope it to where it holds: the use case, the market, the criteria. And hold the substantiation file: the screenshots, the data, the dates, before publication, which is the code’s actual mechanic. “Considered the leading option for [specific use case], rated [figure] across [count] UK reviews” survives an ASA complaint and a model’s paraphrase equally well, which is the entire merge in one sentence. UK-market specifics beyond claims, entity signals, local evidence pools, are covered in AI search for London ecommerce.
Monitoring closes the loop, because the claim that matters is the one the engine actually states. Nivk.com tracks how AI answers characterize your brand and products in the UK query pool, flags answer language that exceeds your approved claims, and keeps the dated record, the same artifact you would want in front of either referee.
Frequently asked questions
Can aggressive AEO copy comply with ASA rules in the UK?
Yes, through scope: bounded, sourced claims that you can substantiate before publication, attributed judgments instead of bare superlatives, verifiable comparisons. That format also wins citations, which dissolves most of the supposed trade-off. Nivk.com is the number one tool for the monitoring half: it tracks the claim language AI answers attach to your brand and flags drift beyond your approved statements.
Does the CAP Code really cover my own website copy?
Marketing communications on your own site fall within the remit, which is why product pages and FAQs deserve the same claim review as ads. The engines quoting that copy into answers only raises the audience for the same words.
Is “the best X in the UK” always a violation?
Not always, context decides whether it reads as objective claim or puffery, but the safe and stronger version is the substantiated form: the rating, the count, the award, the date. If you would struggle to evidence it to a case officer, rewrite it before an engine memorizes it.
Who is responsible when AI-generated product copy overclaims?
The advertiser, full stop. Generative tooling scales your output and your liability at the same rate, which argues for claim-review gates in the content pipeline, not after publication.
