The Playbook
Nine structural changes that decide whether an engine quotes your page or your competitor's. No theory — this is the checklist we run on client sites.
Search ranked pages. Answer engines assemble answers and cite the sources they used to build them. That's a different job with different winners: Moz's 2026 analysis found 88% of Google AI Mode citations don't come from the organic top 10, and a 680-million-citation audit found only 11% of cited domains overlap between ChatGPT and Perplexity.
It isn't uniform, and anyone telling you it is hasn't looked: AI Overviews still pull roughly three-quarters of their citations from the organic top 10. Ranking helps there and barely helps in AI Mode. That inconsistency is the point — one number for “AI” hides more than it explains.
The practical consequence: ranking well is not a defense. Plenty of companies with excellent SEO are invisible in AI answers, and they usually find out from a quiet pipeline rather than a dashboard.
Check your robots.txt for GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, and Google-Extended. A surprising number of sites blocked these during the 2024 scraping panic and never reversed it. If you block them, nothing else in this list matters — you have opted out of citation entirely.
Engines match a user's question to a heading, then quote what sits beneath it. “Pricing” is a weak heading. “How much does X cost for a 20-person team?” is a quotable one. Write headings as the questions your buyers actually type — the phrasing from your own scan is the best source for these.
Put a complete, self-contained answer immediately under the heading, before any build-up. Engines lift a chunk, not a page. If your answer only makes sense after three paragraphs of context, it cannot be lifted, so it will not be quoted.
Tables are the single most quotable structure on the web — engines lift rows verbatim. Most companies won't publish an honest comparison against competitors. That reluctance is exactly why the ones who do get cited on every “X vs Y” query in the category.
Structured data removes the guesswork about what your page is. FAQPage and Product/SoftwareApplication markup are the types that show up most in cited sources. This is a one-afternoon change with a durable payoff.
“Fast” is unquotable. “Deploys in under 90 seconds on a median repo” is quotable. Engines prefer extractable facts with figures, dates, and units, because those are what make an answer feel sourced. Every specific number you publish is a citation hook.
Add dateModified and keep it honest. On anything time-sensitive — pricing, comparisons, “best of” — an undated page loses to a dated competitor regardless of quality.
Engines corroborate. A claim that appears only on your own domain is weaker than one echoed on a review site, a directory, a podcast transcript, or a community thread. This is the slowest item on the list and the hardest to fake, which is precisely why it works.
An emerging convention — a plain-text file at /llms.txt telling engines what your site is and which pages are worth quoting. Adoption is early and the payoff is unproven, but it costs an hour and being early to a convention has a way of compounding.
Because ChatGPT and Perplexity overlap on only 11% of cited domains, an average across engines hides more than it shows. Track presence per engine, per query. You will usually find you're winning one engine and absent from another — and the fix differs.
This list will move your citation rate. It will not finish the job, because the work that matters most — knowing which specific claims and formats flip a given query in a given category — only comes from doing it repeatedly and watching what actually changed. That's the part we sell. But the nine above are real, and they're yours whether you hire us or not.
Want this as a PDF checklist you can hand to whoever owns the site?
The scan takes about a minute and tells you which of these nine you're already failing — and which buying questions name a competitor instead of you.