Documentation

AEO and GEO

Optimising for AI assistants and generative search — with an honest account of the limits.

The honest framing, first

Be careful here

Crawlix improves the inputs AI systems read about you. It cannot make ChatGPT, Claude, Gemini, Perplexity or Google's AI results mention you. That decision belongs to the engine, and any vendor telling you otherwise is selling something they do not control.

What the terms mean

TermMeaning
SEOTraditional search. A ranked list of links
AEOAnswer Engine Optimisation — being used as a source when an assistant answers a question
GEOGenerative Engine Optimisation — being surfaced in AI-generated search results

They overlap heavily. Almost everything that helps an AI system understand your page also helps a search engine, which is why the SEO work comes first.

How AI search differs from Google

GoogleAI assistant
OutputA ranked list of linksA synthesised answer
Your goalBe clickedBe quoted and attributed
Unit that mattersThe pageThe passage
RewardsRelevance, authority, linksClarity, extractability, verifiable claims
FeedbackRank positionAlmost none — this is the hard part

The practical consequence: AI systems lift passages. A page burying its answer in paragraph nine is worse than one answering in the first sentence, even where both rank identically on Google.

What Crawlix can actually improve

These are real detections with real fixes.

LayerWhat Crawlix doesWhat you must do
AccessDetects whether AI crawlers are allowed in. Generates llms.txt, opt-out headers, content versioning headersDecide whether you want AI crawlers at all — blocking them is a legitimate choice
StructureFAQ, HowTo and Speakable schema. Flags missing Q&A structure and absent direct-answer blocksActually answer the question, near the top, in plain language
CitabilityCitation content markup. Flags claims without supportHave real facts, real data, real expertise
Entity signalsEntity schema, author entity bios, E-E-A-T signalsBe a recognisable entity off-site too — consistent naming, real authors
Structured factsProduct, Image, Video, Audio and organisation markupVerify anything asserted as fact before approving it
LocalLocalBusiness markup and completeness, NAP consistency, geo fieldsYour Business Profile, reviews and physical location — none of which any tool controls

Worth knowing

Direct-answer blocks always require your review. They insert prose that asserts something, so a person reads them first.

What Crawlix can currently measure

Crawlix includes citation checking: it drives four engines with a browser and reads the answer — ChatGPT, Claude, Gemini and Perplexity. Three things to understand before you plan around it.

Be careful here

1. It is off unless enabled for your plan. Citation checking requires operator configuration and a verification step. Until then it reports "not configured" — which means we did not check, not "no citations found". Those are different facts and Crawlix will not conflate them.

2. It is four engines, sampled, with named failure modes. When a check cannot run, Crawlix tells you why: the engine demanded a login, it changed its layout and our reader lost the answer, the request failed, or the figure came from an unverified source. Tools reading six to nine engines from licensed feeds cover more ground. We do not match that and do not claim to.

3. The AI Readiness Score is not a citation predictor. It checks whether one page can be reached, read, quoted and attributed. Its bands are explicitly not calibrated against citation outcomes.

What Crawlix cannot measure

Can measureCannot measure
Whether AI crawlers can access your siteWhether an engine actually used your page
Whether your content is structured for extractionYour “share of voice” in AI answers
Whether claims are citable and attributedWhy an engine chose a competitor
Whether four named engines mentioned you, when enabledThe other engines, or answers not sampled
AI crawler visits to your siteWhat the crawler did with what it took

Not currently supported

There is no single AI visibility score in Crawlix, and there will not be a fabricated one. There is no honest way to compute a share-of-voice figure from four sampled engines with known failure modes. If you see a competitor publishing one, ask what it is measured against.

What to actually do

Do the SEO work first. An unindexable page is not an AEO problem.

  • Check AI crawlers are not blocked in robots.txt
  • Generate llms.txt
  • Answer the question in the first paragraph on every page meant to answer one
  • Add FAQ schema where you genuinely have FAQs — not invented ones
  • Add real author identities to anything requiring expertise
  • Make claims specific and verifiable, and cite sources
  • Fill in entity and organisation markup
  • If local: complete LocalBusiness markup and keep your details consistent

Be careful here

No amount of markup fixes having nothing worth citing. AEO rewards substance and is worse than useless at faking it — an AI system quoting a claim you invented has published your fabrication under your name.

Honest expectations

  • Improvements here are slower and harder to attribute than technical SEO
  • You may do everything right and still not be cited — engines have their own preferences and licensing deals
  • The field changes fast; assume some of this ages

The reason to do this work is that it makes your content clearer, better structured and better sourced. That helps traditional search, helps human readers, and gives AI systems something worth using. Those benefits are real whether or not any engine ever cites you.