Every claim below is circulating right now in sales decks, LinkedIn posts, and agency pitches. Each one gets the same treatment: the claim, why people believe it, the actual mechanism, and the verdict. The mechanisms come from this book's own chapters and the numbers from the OppAlerts study documented in The Research Behind This Guide, so every debunk here can be checked rather than taken on my authority.
"SEO is dead"
It spreads because it flatters both sides: vendors selling new tools want the old discipline buried, and practitioners tired of the old discipline want permission to stop. The mechanism says otherwise. AI search products retrieve from search indexes, mostly Google and Bing, and the study finds that brands recommended by AI sit in Google's organic top 10 in 60.9% of memory answers, rising to 65.6% once search turns on. Google visibility predicts AI visibility at a partial correlation of about 0.17 even with backlinks and the other signals held fixed. Verdict: false. Classic SEO is a load-bearing input to AI visibility; What Carries Over from SEO, and What Breaks maps exactly which parts transfer.
"Blocking GPTBot protects your content"
It spreads because blocking a named bot feels like a concrete act of control. The mechanism is narrower. GPTBot is OpenAI's training crawler, and disallowing it signals only that your content should be excluded from future training; OpenAI documents it as independent from OAI-SearchBot, which powers search results, and from ChatGPT-User, which fetches pages during live conversations.OpenAI's bot documentation, which states each bot's robots.txt setting is independent of the others. Blocking GPTBot does nothing about models already trained on your pages, nothing about other providers' crawlers, and nothing about your text appearing in third-party pages and datasets that do get crawled. Verdict: mostly false, and for most brands strategically backwards: the businesses in this book's audience are usually fighting to get into training data, not out of it. Where the Machines Get Their Data: Crawlers, Indexes, and Training Pipelines covers the full bot roster and what each one feeds.
"Schema makes AI cite you"
It spreads because schema is concrete, cheap, and auditable, which makes it easy to sell as the fix. The mechanism: structured data is a standardized format that gives search engines explicit clues about page content, and Google documents it as enabling rich results in classic search.Google's structured data introduction, which describes structured data as clues about page meaning that can enable rich results; it makes no claims about AI citation. No major AI platform documents schema as an input to citation selection, and the models write answers from retrieved page text, not from markup. Schema still earns its place indirectly, by strengthening the classic rankings that retrieval draws from and by feeding knowledge graphs. Verdict: unproven as a citation lever, useful as infrastructure. The honest case for it is in Structured Data and Machine-Readable Facts.
"llms.txt is a proven standard"
It spreads because it pattern-matches to robots.txt, a real standard, and because adding a file feels like early-mover advantage. The mechanism: llms.txt is a proposal from September 2024 for a file that helps LLMs use a website at inference time.The llms.txt proposal, authored by Jeremy Howard, published September 3, 2024; it describes itself as a proposal. No major AI platform documents consuming it. Verdict: it is a proposal, cheap to adopt and harmless, and any vendor presenting it as a required standard is ahead of the evidence. Current adoption status is tracked in Structured Data and Machine-Readable Facts.
"AI search rewards fresh content everywhere"
It spreads by generalizing from one true case. Freshness matters in live retrieval, where the model reads current pages. But memory answers come from weights frozen at a training cutoff months or more in the past, and the study's news finding runs opposite to the myth: press coverage predicts memory answers (rho 0.169) more strongly than searched answers (0.139), because coverage works as accumulated reputation absorbed into training, not as a recency signal. Verdict: false as stated. Freshness is a retrieval tactic; the two staleness clocks and where updating actually pays are in Freshness, Cutoffs, and Temporal Accuracy.
"You can rank #1 in ChatGPT"
It spreads because it imports SEO's most familiar deliverable into a channel that sells better with a familiar deliverable. The mechanism breaks the frame: an LLM samples each token from a probability distribution, with a randomness setting called temperature, so the same prompt on the same model produces different answers on different runs, by design. There is no index holding a position between queries. What exists is a distribution: your brand appears in some share of sampled answers, at some average prominence. Verdict: false as a concept. Measure share of answers over repeated runs instead, per Measuring AI Search Visibility, and treat any single-query "ranking" screenshot as one coin flip.
"Video is a major citation driver"
It spreads because video correlates with AI visibility, and because video is expensive, which makes the correlation profitable to sell. The mechanism behind the correlation is fame: famous brands dominate the video shelf and the AI answers at the same time. In the study's citation logs, video is 0.5% of what search-grounded models cite, against 9.2% for Reddit and 2.6% for Wikipedia, and most of that sliver comes from one model. Verdict: false as a general lever. Where video genuinely earns its cost is scoped in Multimodal: Images, Video, and Audio.
"Mentioning your brand more on your own site teaches the model"
It spreads because it is the one tactic entirely inside your control. The mechanism disagrees on both layers. Model memory forms from broad training corpora where the signal is many independent sources discussing you; the study's leaderboard is topped by backlink authority and Reddit discussion, both third-party signals, and your own site's self-description carries no independent weight in it. On the retrieval layer, stuffing self-mentions does nothing for whether your pages get retrieved, and scaling out low-value pages to do it collides with published spam policies.Google's spam policies, which define scaled content abuse as generating many pages primarily to manipulate rankings rather than help users. Verdict: false. What third parties say moves the model; that argument runs through Links: The Signal That Refuses to Die and Reviews, Reddit, and Communities.
The filter for the next claim
New myths will outrun this page, so apply the test that produced these verdicts. Ask which layer the claim works through: training data, retrieval, ranking inside the search index, or generation. Then ask what would have to be true mechanically at that layer, who documents it, and what experiment would show it. A claim that cannot name its layer and mechanism is marketing.
Most hype dissolves at the first question. The rest usually dissolves at "who documents it," because the platforms publish more about their machinery than the people selling shortcuts ever quote.