Something quietly new appeared in analytics: sessions from AI assistants — ChatGPT search, Perplexity, Copilot — arriving as referrals, sometimes converting better than search because the visitor arrives pre-sold on your relevance. The optimization playbook for this channel is young, but its patterns are already legible.
How assistants choose sources
Retrieval-based assistants search the web, read candidate pages, and cite the ones that answer cleanly. The selection favors what good editors favor: direct answers near the top, original data, clear structure, freshness, and the credibility signals of a real site. Pages that bury their answer under preamble get skipped — the model reads like an impatient researcher with a quota.
Make pages quotable
Answer-first formatting: the question as a heading, the standalone answer in the first sentences — the same structure that wins featured snippets wins citations, for the same reason. Original specifics: your test results, your numbers, your methodology. Models synthesize the consensus; they cite the primary sources. A page that exists only to restate consensus has nothing worth citing. Clean semantics: proper headings, tables as tables, dates visible — machine-readability is literal here. Entity clarity: an about page, author identities, consistent naming — assistants weigh source credibility like cautious librarians.
Track it and feed it
Watch referrers for the AI domains in analytics — the traffic is small but growing and the conversion behavior is worth knowing. Notice which pages get cited (some tools show this; the referral URLs imply the rest) and build siblings to those winners. And keep the traditional SEO floor strong: retrieval systems lean on search indexes, so pages that rank poorly tend to get read rarely.
The honest framing: this is a lottery with better odds than most. You can't force a citation; you can make your site the kind of source that deserves one — which, conveniently, is the same thing search engines have always rewarded.
