Writing for the Reader Who Asks a Machine First

A growing share of the people your content is written for never actually reach your page. They ask a chatbot a question, the chatbot reads several sources including yours, and it hands back an answer with your brand mentioned — or paraphrased, or misattributed, or left out — somewhere inside it. Your words are still doing work. They're just doing it once removed, through a system you don't control and can't proofread.

A different game from SEO

The industry has settled on two overlapping terms for this: Generative Engine Optimisation (GEO), structuring content so tools like ChatGPT, Perplexity and Claude cite it in an answer, and Answer Engine Optimisation (AEO), aiming to be the source quoted directly in a featured snippet or an AI overview. The useful distinction, however you label it: classic search rewards the page that ranks, and a ranked page earns a click. This rewards the source that gets quoted, and a citation earns a mention inside the answer — the reader may never click through at all (Writer's guide to the distinction is a clear starting point).

That changes what "working" means for a piece of content. A page can perform badly by every traditional metric — low traffic, few clicks — while still being the source an AI system quotes verbatim to thousands of people who never landed on it. Analytics dashboards built around visits don't see that at all, which is its own problem for anyone trying to prove this is worth doing.

Most of your visibility here isn't on your own site

The uncomfortable finding in most of the research on this: a large majority of brand mentions in AI search answers come from third-party pages, not from the brand's own website — one widely cited estimate puts it at around 85% (also from Writer's guide; treat any single precise percentage in this space with some caution, because the underlying measurement methods vary and the field is moving fast, but the direction of the finding is consistent across the write-ups we've checked). Press coverage, directory listings, review sites, forum threads and third-party comparisons are feeding these answers as much as, or more than, a brand's own copy does. For a corporate content team, that means the brief can't stop at your own site. Coverage you don't directly control — earned media, partner pages, industry directories — is now part of your visibility surface in a way it wasn't for classic SEO.

What actually changes in the writing

• Put the direct answer where it can be lifted. A short, plain-language answer to the obvious question, placed directly under a clear heading, gives an answer engine something it can extract cleanly. Burying the actual answer three paragraphs into scene-setting works against you here in a way it never quite did for a human reader either — this is really just good writing, made more important.
• Cut the hedging. Phrases like "we believe" and "in our opinion" signal uncertainty to a model doing next-token prediction, and content that reads as confident and specific is easier for an engine to lift as a clean answer than content that reads as tentative. This is a genuinely useful discipline for house style generally, whatever it does for a machine.
• Back claims with something checkable. Content built around specific statistics, named sources and expert quotes is reported to earn substantially more AI citations than content without them — some studies put the gap at 30–40% (Frase's AEO guide discusses the research behind this). It's also, not coincidentally, the same standard that makes writing trustworthy to a human reader — which is worth remembering before treating this as a separate discipline from writing well.
• Be indexable by Bing, not only Google. ChatGPT's search feature draws heavily on Bing's index rather than Google's, so a site that has only ever optimised for Google search visibility may be invisible to a meaningfully large slice of AI-mediated traffic without anyone noticing.

The part worth being sceptical about

This is a genuinely new field, most of the research is vendor-produced, and the specific percentages circulating — including the ones cited above — should be read as directional industry estimates rather than settled facts. There's also a real risk in over-indexing on any of it: content engineered purely to be machine-extractable, stripped of tone, nuance and the things that make writing worth reading, is a worse outcome for your actual human audience even if it technically performs better in an engine's citation count. The safest position, and not just because it's the one we'd take anyway: write clearly, back your claims, structure for a reader who's skimming — and treat the AI-visibility benefit as a useful side effect of writing well, not a separate discipline that overrides it.

Checking it, rather than guessing

Unlike classic search, there's no equivalent of Google Search Console for most answer engines yet — no dashboard a brand can log into and see exactly how it's being cited. The practical workaround most teams use is manual: ask the same handful of likely customer questions directly of ChatGPT, Perplexity and Copilot, in each market's language, on a regular cadence, and note whether the brand appears, how it's described, and whether the description is accurate. It's basic and a little tedious compared with a proper analytics tool, but it's the closest thing available today to knowing what's actually being said on your behalf inside an answer you'll never see rendered.

For a multilingual brief specifically

One consequence worth flagging for anyone commissioning content in more than one language: this third-party visibility effect is language-specific. Being well cited in English answer engines says nothing about whether the same brand is being cited correctly, or at all, when the question is asked in French or German. If AI-mediated visibility genuinely matters to a client's strategy, it needs checking market by market, in the language that market actually searches in — not assumed from the English result.