The AEO Playbook

1. What is Answer Engine Optimization (AEO)?

AEO is the practice of structuring web content so AI answer engines — ChatGPT, Perplexity, Google AI Overviews, Claude, and Gemini — quote it directly inside generated answers. It overlaps with SEO on E-E-A-T but diverges sharply on formatting: AEO favours short, self-contained paragraphs, question-based headings, and rich structured data over long-form prose.

• AEO vs SEO

  SEO optimises for ranked lists of blue links. AEO optimises for being the sentence an AI engine lifts into its answer.

• AEO vs GEO

  Generative Engine Optimization is the same discipline under a different name; treat them as synonyms.

• Why now

  By mid-2026, roughly 42% of informational queries in the US resolve inside an AI-generated answer before a user clicks through. Not being cited is not being seen.

2. Build a machine-readable entity graph

AI engines resolve everything to entities — a real Organization, a real Person, a real Product. If your entity isn't disambiguated, you get generic mentions instead of citations. The entity graph is the highest-leverage AEO investment, full stop.

• Organization schema

  Publish Organization JSON-LD on every page with a stable @id, logo, sameAs (LinkedIn, X, Crunchbase, Wikidata), and founder references.

• Person schema for authors

  Every article needs an author with a Person node linking to a real author page. sameAs on the Person outranks backlinks for AI Overview inclusion.

• Consistent @id URLs

  Reuse the same @id string across pages so engines merge the graph. A shifting @id creates duplicate entities.

3. Format content to be quotable

Quotability is a property of the paragraph, not the page. AI engines lift 40–80 word blocks that stand alone semantically. Long, hedged, or narrative paragraphs almost never get quoted verbatim.

• Question headings

  Phrase H2/H3s as the exact question a user would ask an AI. Question headings paired with a 40–60 word answer paragraph lift citations 34% (Internet Monkeez benchmark, July 2026).

• Definition-first paragraphs

  Open each section with a self-contained definition. Save nuance, caveats, and stories for later paragraphs the AI engine will skip.

• Enumerated lists

  Numbered ordered lists (<ol>) plus ItemList schema get quoted more often than prose — Claude in particular lifts entire bullet sequences.

• Tables for comparisons

  Comparison tables (with a caption) are Perplexity's favourite quotable block. Use them for X-vs-Y content.

4. Layer the right schema on every page

Every page should carry at least three schema types: a site-wide layer (Organization + WebSite), a page-type layer (Article, FAQPage, HowTo, DefinedTerm, Product…), and a navigational layer (BreadcrumbList). Add Speakable to the paragraphs you actually want quoted.

• Site-wide

  Organization, WebSite, and a SearchAction. Put these once in the root layout so every URL inherits them.

• Page-type

  Match the schema to intent: Article/BlogPosting for editorial, FAQPage for Q&A, HowTo for playbooks, DefinedTerm for glossaries, Product/Offer for commerce, Review/AggregateRating for social proof.

• Speakable

  Add SpeakableSpecification with cssSelector targeting the exact paragraphs you want lifted into voice/AI answers. Keep them to 40–80 words.

• BreadcrumbList

  Every non-root page needs a BreadcrumbList. It is the cheapest way to earn a citation-worthy path label.

5. Ship /llms.txt and /llms-full.txt

The llms.txt standard is now a hard gate for at least one revenue-sharing AI engine (Perplexity Publisher API) and a strong preference signal for others. Split the two files correctly: llms.txt is a curated markdown index of your most citable URLs; llms-full.txt is the corpus.

• llms.txt

  A short markdown file linking to your best pages with one-line descriptions. Group by section (Products, Guides, Research, Answers).

• llms-full.txt

  The full text of those pages concatenated so an AI can ingest your corpus in one request.

• Link both from the footer

  Crawlers discover /llms.txt at the root but confirm it via footer links. Link both files publicly.

6. Signal freshness ruthlessly

Since Gemini 3's live-web launch and Google AI Overviews' 2026 shift to fewer, more recent citations, freshness is a first-class ranking signal. Undated pages effectively don't exist to modern answer engines.

• Dated everything

  Every article needs datePublished AND dateModified in JSON-LD, plus a visible date near the H1. Prefer ISO 8601.

• Fast crawls

  Keep TTFB under 800ms, allow Google-Extended and GoogleOther in robots.txt, and set Cache-Control that permits 5-minute revalidation.

• Newsroom cadence

  Ship a dated /news feed at least weekly. AI engines re-crawl active domains far more often.

7. Measure what actually gets cited

Rank tracking is dead for AI queries — there is no rank. Instead, track citation share: how often each engine names you in an answer for the prompts you care about.

• Prompt list

  Maintain a spreadsheet of 30–100 prompts spanning definitional, commercial, and comparison intent. Re-run monthly.

• Engines to test

  ChatGPT, Perplexity, Google AI Overviews, Claude, and Gemini — at minimum. Each has different heuristics.

• Publish your proof

  When you're cited, screenshot it and publish it. Citation proof pages themselves rank and earn more citations.

FAQ