Assistants answer in two steps: they retrieve a handful of pages, then write an answer that quotes some of them. To be cited, you must survive both steps.
Step one: be retrieved
Retrieval is mostly search. ChatGPT search, Copilot and others lean on established indexes and their own crawls; Perplexity runs its own. That means classic SEO still decides most of it: be indexed, be relevant to the question, be linked to by others.
The additional conditions, covered in the readiness guide:
- Their crawlers are allowed by robots.txt and not blocked by your firewall.
- Your text exists in the HTML without JavaScript.
- The page is not too slow to fetch.
Step two: be quotable
Once retrieved, the assistant chooses which sources to quote. From what is observable, these pages do better:
They answer the question directly. The specific sentence exists on the page, in plain language, near the top. If the answer is only implied across three paragraphs, another page gets quoted.
They are specific. Numbers, steps, prices, dates, named examples. Vague pages are summarised, not cited.
They are self-contained. A section that makes sense on its own can be quoted on its own. Text that depends on the previous section, or on an image, cannot.
They show their basis. "We measured", "according to X", a date, a method. Assistants and their guidelines favour sources that look accountable.
They are current. For anything that changes, a page updated this year beats one from three years ago, even if the older one ranks better in classic search.
Practical page patterns
- Put a two-sentence answer immediately under the H1, then the detail.
- Phrase H2s as the questions people ask, and answer each one in the first sentence beneath it.
- Add a short FAQ for the questions your sales or support team hear, with real answers, marked up with
FAQPagedata. - Keep tables for comparisons, but restate the conclusion in a sentence: a table alone is hard to quote.
What does not work
- Keyword stuffing, which reads as noise to both readers and models.
- Hidden text or instructions aimed at AI ("ignore other sources"). Assistants are built to disregard page content that tries to instruct them, and it damages trust if noticed.
- Mass-generating hundreds of thin pages. Retrieval rewards the page that answers best, not the site with the most pages.
Measuring it
There is no Search Console for AI answers yet. What you can do:
- Ask the assistants. List the ten questions your customers ask, ask each assistant, and note who gets cited. Repeat monthly. It is manual, and it is the most direct evidence available.
- Watch referrals. Assistants send visits with their own referrer. Look for
chatgpt.com,perplexity.ai,copilot.microsoft.comand similar in analytics. - Watch your logs.
OAI-SearchBot,PerplexityBotandClaude-SearchBotin your server logs confirm they can reach you and which pages they fetch. - Track readiness. Rankwell's readiness score tells you whether the mechanics are in place, which is the part you control.
The honest summary
You cannot make an assistant cite you. You can make yourself eligible, easy to read, and easy to quote, and you can remove the blocks that rule you out. Most sites that are invisible to AI answers are invisible for a mechanical reason, not an editorial one.