Answer engine optimisation best practices, in priority order

Nearly every best-practice list in this category is a flat list of twenty items. That is the problem with it. Two of those items decide most of your result and the other eighteen are rounding, and a list that presents them as equals will send you to spend three months on the rounding.

Why the order is the practice

These items are not independent. They are a sequence with hard dependencies, and treating them as a checklist to work through in any order is how people spend a quarter on schema markup while a firewall quietly returns 403 to every AI crawler.

That is not a hypothetical. It is what happened to us. Our own site scored zero because bot protection we never consciously switched on was blocking every AI crawler at the door. Every piece of content work we had done up to that point was worth exactly nothing, and no amount of additional content would have changed it.

1. Let the crawlers in

Request your homepage with the user agents GPTBot, ClaudeBot, PerplexityBot and ChatGPT-User and confirm each returns a 200. Then read your robots.txt for Disallow rules under those agents, and check your CDN or firewall for a managed AI bot rule.

This is a binary. There is no partial credit and no clever content that routes around it. It is also the practice most often failed by exactly the businesses who think they are ahead, because the blocking usually arrives as a security default rather than a decision anyone made.

The full method, with the commands, is at are AI crawlers blocked on your site.

2. Be quotable

Engines lift statements. They do not lift atmosphere, and they do not reward the slow build-up that works on a human reader.

• Answer in the first two sentences. Put the direct answer at the top of the page, then support it. Burying the conclusion under six paragraphs of context is a human convention that costs you here.
• Use the buyer's words, not your internal ones. If buyers type how do I get my business mentioned in ChatGPT, a page headed Generative Discovery Solutions will not match it. Take the phrasing from real queries, not from a brand workshop.
• Write headings as the questions people ask. A heading that is a question, answered directly underneath, is the single most liftable structure on a page.
• Make the page readable without JavaScript. Many fetchers read raw HTML. If your sentences appear only after a framework boots, they may not be seen at all.

More on this at writing content AI will quote.

3. Be verifiable

Engines cross-check facts before repeating them, and a fact they cannot confirm tends to get dropped rather than guessed. That makes consistency a ranking input in a way it never quite was for traditional SEO.

• Same business name, same one-line description and same offer on your site, your LinkedIn, and every directory you appear in. Read the three side by side and ask whether a stranger would say they describe the same company.
• Organization schema on the homepage and FAQPage schema where you genuinely answer questions. Not everywhere. Marking up a page as an FAQ when it is not is the kind of thing that gets discounted.
• An llms.txt at your root. It is a young standard and it costs ten minutes, so the expected-value calculation is easy even though adoption is still early.

Detail at schema for AI search.

4. Be mentioned somewhere engines already cite

This is the one that matters most and the one almost every best-practice list under-weights, because it is the only item that cannot be completed in an afternoon.

Roughly 85% of the sources AI engines cite are third-party pages: roundups, directories, review sites, trade press, communities. Your own site sets your ceiling. Other people's pages decide where you sit under it.

The practical version: ask an engine the main buying question in your category, open every source it cites, and search each one for your name. The pages that come back are your actual target list, and they are specific rather than generic. Getting into three of them will move your number further than another twenty pages on your own site.

Method at how to get mentioned in AI answers.

5. Re-measure on a schedule

Answers move as models update and as competitors publish. A number from March tells you very little in July, and the failure mode is not that you get the wrong answer, it is that you stop having one and do not notice.

Monthly is enough. Measure the same questions each time so the series means something, and record the date next to the score.

Practices worth skipping

A short list of things that circulate as AEO advice and do not hold up.

• Mass-producing near-duplicate pages. This is Google's own definition of scaled content abuse, and the penalty is not a dip, it is removal. One strong page per topic beats thirty variations of it, and the variations also compete with each other.
• Stuffing schema onto pages that do not warrant it. Markup describes a page, it does not upgrade one.
• Optimising for a single engine. Overlap between what ChatGPT, Perplexity and Gemini cite is smaller than most people assume, so tuning to one is a narrow bet.
• Treating AEO as a rename of SEO. If it were the same measurement, a site with the worst technical score in our 24-site study could not have been named in 6 of 6 answers while a site scoring 90 was named in none. Both happened. The comparison is set out at AEO vs SEO.

Questions, answered