Be wary of anyone who promises to “train” ChatGPT or Gemini on your brand, guarantees AI citations, or sells llms.txt or special schema as the key to AI answers. Google says optimizing for its AI features is still SEO, that its search ignores llms.txt, and that no special markup is needed. What’s real is retrieval: AI answers pull from pages that are crawlable, indexed and genuinely useful, and SEO influences that.
Why this needs saying
AEO and GEO are real terms for real work. They also attract a lot of overselling. Google’s AI optimization guide now has a whole section titled “Mythbusting generative AI search: what you don’t need to do,” and tells site owners considering third-party AEO or GEO services to review its guidance on evaluating SEO advice. When a search engine publishes a myth list, it’s seeing the claims too.
For the mechanics behind everything below, see how AI answers pick their sources.

The claims, checked
“We train AI on your brand.”
False. Foundation models are trained by the labs that build them, on datasets they choose. An agency can’t insert your brand into that process. What crawlers like GPTBot or ClaudeBot may collect for training is up to the lab, and your only real control is whether to allow them, per OpenAI and Anthropic. AI search answers are grounded in live retrieval, and that’s where SEO actually works.
“We guarantee you’ll be cited in AI answers.”
No one can guarantee that. Google says that even when a page meets every requirement, “indexing and serving aren’t guaranteed.” AI answers also vary from run to run. In BrightLocal’s study of 200,085 local searches, repeated runs of the same prompt shared only 21% to 33% of the businesses named. A guarantee on something that changes every time you ask is a sales line, not a result.
“llms.txt will get you into AI answers.”
Not in Google. Google says you don’t need llms.txt, AI text files or special Markdown to appear in Search or its AI features, and that its search ignores them. Creating one “will neither harm nor help” your visibility in Google Search, as Search Engine Journal also covered. We haven’t found a statement from OpenAI, Anthropic or Perplexity saying their answer engines read third-party llms.txt files. It’s a harmless file, not a strategy. See our llms.txt entry.
“You need special AI schema.”
No. Google says structured data isn’t required for generative AI search and there’s “no special schema.org markup you need to add.” Standard structured data is still worth having for rich results. The Toronto citation study, arXiv 2610.15033, found structured data and FAQ blocks helped only when comparing different websites, not pages within one site.
“Chunk your content so AI can read it.”
Not required. Google says there’s no need to break content into tiny pieces and no ideal page length. Clear headings and paragraphs help readers, and that’s reason enough.
“Rewrite everything for AI.”
Not required. Google says AI systems understand synonyms and meaning, so you don’t have to cover every long-tail variation. The useful version of this advice is narrower: make sure titles and paragraphs state real questions and answer them plainly.
“We’ll get you mentioned across the web.”
It depends how. Earned coverage, genuine reviews and accurate listings help, because models favor brands they recognize. Bought or fake mentions don’t. Google says seeking inauthentic mentions “isn’t as helpful as it might seem,” because its AI features depend on the same spam systems as the rest of Search. Fake reviews also break the law: see the FTC’s Consumer Reviews and Testimonials Rule, which explicitly covers agencies and reputation firms.
“We’ll publish hundreds of AI pages for every question.”
Risky. Google says creating separate content for every fan-out variation primarily to manipulate rankings or AI responses violates its scaled content abuse policy. That’s a fast way to hurt the whole site. See our note on black-hat SEO.

What is real
- Retrieval depends on search. Google says the best practices for SEO continue to apply because its AI features are built on core ranking systems.
- Crawler access matters. Blocking a search crawler such as
OAI-SearchBotorClaude-SearchBotcan remove you from that assistant’s answers. - Matching real questions helps. It’s the one within-site factor the Toronto data supports.
- Local listings feed local answers. Google Business Profile, Maps and Yelp show up heavily as sources. See reviews and AI answers.
- Measurement now exists. Search Console, Bing and ChatGPT all report some AI visibility data. See how to measure it.

Seven questions to ask any AEO or GEO provider
- What exactly will you change on our site, and why do you expect it to matter?
- Which numbers will you report, from which tools, and how do you handle answers that change from run to run?
- What do you promise, and what don’t you promise?
- Do you buy links, mentions or reviews, or use networks that do?
- Will you mass-produce pages? How many, and who reviews them?
- How do you use AI tools in your own writing, and who checks facts?
- What happens if results don’t move in six months?
A good provider answers these plainly and won’t mind being asked. Our publishing principles set out how we work, and our guide to choosing an SEO company covers broader vetting.
Our position
We work on AI search visibility every day, and we’ll tell you what it is: the same discipline as SEO, applied to how retrieval systems find and quote pages. We don’t train AI, we don’t guarantee citations, and we don’t sell files that search engines ignore. We build pages and reputations that deserve to be found, then measure what happens. It’s a marathon, not a sprint. Our AI Search Visibility service has the details.
