AEO myths: the tactics we never recommend
Our rules engine is not allowed to suggest any of these. Each one is sold somewhere as a way to get cited by AI assistants, and each one has evidence against it. The evidence is other people's research, linked on every item, not a measurement of ours. This is the same list the product reads, version aeo-2026.07.2.
1.“Adding llms.txt gets you cited by AI.”
No major AI search feature reads llms.txt as a ranking signal; Google says no AI-specific text file is needed. Ship it only as harmless agentic-web documentation, never as a citation lever.
Evidence: seranking.com ↗2.“Structured data directly increases AI citations.”
A controlled study (1,885 pages) found no significant citation lift and a small decline in AI Overviews. Ship schema for entity clarity and rich results, not as a citation driver.
Evidence: seroundtable.com ↗3.“Manually chunk content into AI-sized pieces.”
Google explicitly says you can ignore chunking; engines handle full-length pages. Write for humans and structure normally.
Evidence: developers.google.com ↗4.“Increase keyword density for LLMs.”
Keyword stuffing performed ~10% worse than baseline in the Princeton study and degrades fluency (a real positive factor). Never recommend it.
Evidence: arxiv.org ↗5.“Longer content earns more citations.”
Word count correlates near-zero with citations and grounding plateaus; padding can hurt retrieval. Recommend density, not volume.
Evidence: ahrefs.com ↗6.“Confident, persuasive phrasing wins citations.”
Authoritative phrasing WITHOUT supporting evidence had small-to-negative impact in Princeton. Confidence only works paired with stats and citations.
Evidence: aisearchglossary.com ↗7.“Get the brand mentioned everywhere by any means.”
Fabricated or astroturfed mentions are filtered like link schemes and add no genuine entity signal. Pursue earned, authentic mentions only.
Evidence: developers.google.com ↗8.“GEO/AEO is a new discipline separate from SEO.”
Google's position: what wins AI citations (real expertise, citations, clear structure, entity clarity) is the same work as good SEO. Frame AEO as authority/entity hygiene, not a distinct ritual.
Evidence: searchenginejournal.com ↗9.“You need a Wikipedia article to be cited.”
A Wikidata item plus consistent multi-source corroboration matters more and is achievable without meeting Wikipedia notability. Prioritize Wikidata.
Evidence: overthetopseo.com ↗10.“Buying clicks to inflate your search click-through rate (SerpClix-style) raises rankings.”
Google's engineers state CTR is not a direct ranking signal; the 2024 leak that manipulators cite actually reveals the filtering infrastructure Google built to normalize and strip manipulated clicks. Efficacy is fundamentally unmeasurable (you cannot separate paid clicks from noise), so by Evidence over Opinion it is effort not to spend. Improve legitimate CTR with better titles and answers instead.
Evidence: seroundtable.com ↗11.“Paying humans or bots to seed prompts and engage with AI answers that mention your brand will bias the engine toward citing you.”
This is CTR click-buying one layer up: manufactured engagement is exactly the abuse pattern answer engines filter, its effect is unmeasurable, and it risks penalties. We refuse to sell engagement-manufacturing. The only durable path is becoming genuinely more citable (structured, sourced, extractable content that earns the mention). Named pre-emptively so we never drift into recommending it.
Evidence: developers.google.com ↗
What we check instead
The free audit reads your pages for the things that do decide whether an assistant can read and quote you: crawler access, content in the server HTML, direct answers, sourced facts. It asks for no account. Run the free audit, or see what we refuse to do.