Every one of these is said out loud in sales calls. None of them survived a hundred audits.
myth 1 — "our SEO is strong, AI will pick us up"
A brand with 8,336 visits a month holds no recommendation slot. A vendor with 29 visits holds it in all three engines. This is the single most important finding of the hundred: there is no correlation between organic traffic and entering an LLM answer. None.
| benchmark example | seo performance | llm visibility |
|---|---|---|
| niche B2B vendor | 29 visits/mo · 6 keywords | recommended by all 3 engines |
| its traffic-leader competitor | 8,336 visits/mo · ~3,000 keywords | holds no recommendation slot |
| event agency | 1 visit/mo | named by AI in 2 verticals |
| the biggest brand in its category | 270,000+ visits/mo | only #13 for its own head query |
| infrastructure company | DA 1–5 · ~2.5 visits/mo | #1 pick in ChatGPT and Claude |
A model builds its answer not from rankings but from sources it trusts — best-of listicles, industry directories, Wikipedia, readable whitepapers. Placement decides, not positions. Bad news if you spent five years buying only classic SEO. Very good news for everyone else: entry into the answer is cheap today and does not require years of domain authority.
myth 2 — "we're too small for AI answers"
In our data, categories were regularly won by businesses getting 1 to 41 visitors a month from Google. Their shared trick is page format: a dedicated page per vertical or use case instead of one generic services page, a comparison page instead of a product page. One structurally correct page beat competitors' year-long content plans.
myth 3 — "AI knows our brand, so it recommends us"
The gap showed up in 87% of audits. Branded recall runs at ~93% — the model knows the brand and describes it accurately. Category recall is 4.5% across 66+ live prompts. AI knows exactly who you are, and in 95.5% of answers it recommends someone else.
In one audit the category slot was held by a competitor whose website did not even resolve. Its name simply lived in the sources the model cites. Knowledge is not recommendation. The recommendation has to be built separately.
myth 4 — "the site is fast, Lighthouse is green, bots are fine"
Lighthouse measures a human's experience in a browser with JavaScript executed. AI crawlers don't execute JavaScript. That is how Lighthouse 100/100 and an empty shell for bots coexist on the same page — we watched it happen. The only honest test is a curl as GPTBot. It takes a minute, and the result almost always hurts.
myth 5 — "just publish more content"
| case from the audits | volume | result |
|---|---|---|
| site with 1,422 URLs | 1,000+ articles | 0 top-3 positions; a competitor 6.8× smaller gets 6× the traffic |
| blog with 136 posts | 136 URLs | a canonical bug collapsed all of them into 1 indexed URL |
| 896 articles of ~4,500 words | ~4M words | stuck on page 2 — every author is anonymous |
Volume without structure, markup and authorship isn't an asset, it's ballast. The winner isn't whoever writes more. It's whoever the machine can read, verify and quote.