GEO Myths, Graded: llms.txt, Schema, FAQ Pages and 15 More

No, llms.txt doesn’t help: Google says its Search ignores the file, and server logs from 137,000 sites show 97% of llms.txt files were never requested in a month. Schema and FAQ markup don’t buy citations either; a controlled test of 1,885 pages found no meaningful lift, although structured data still matters for rich results. GEO itself isn’t a scam, but most tactics sold under the name are unproven, and the ones with evidence are unglamorous: let AI search crawlers in, serve content in plain HTML, be indexed and snippet-eligible on Google, keep pages fresh, and earn mentions on sites you don’t own. Below are 18 common claims, each with a grade, the evidence and what to do instead.

How we grade

We use the same seven grades as our methodology, applied here to each claim as people actually make it:

• official: the platform says so in its own documentation.
• strong: a direct mechanism, or consistent results across large studies.
• moderate: supported by several studies, with real caveats.
• weak: little support, or small and inconsistent effects.
• unstudied: plausible, but nobody has measured it.
• practice: standard web practice, low stakes on its own.
• myth: platform guidance or a controlled or direct test says it doesn’t work as claimed.

Where the evidence file behind our audit already grades a tactic, the grade here matches it. Where it doesn’t, we grade conservatively: a claim is only called a myth if a platform’s own documentation or a direct test contradicts it; otherwise the lowest it gets is weak. Most studies in this field are run by vendors and are correlational, and we say so each time.

The scorecard

Files, markup and formatting

1. “Add an llms.txt file so AI can find you” — myth

Google’s guide is explicit: “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search”, and such files will “neither harm nor help” because “Google Search ignores them.” Usage data agrees. Ahrefs’ June 2026 analysis of server logs from 137,000 sites found 97% of llms.txt files received zero requests, and no AI bot went looking for the file where none existed; SE Ranking found no correlation with AI citations across about 300,000 domains. Two nuances. Google’s Lighthouse now has an llms.txt audit, which marks a missing file “Not Applicable”. And Shopify serves /llms.txt on every store by default: when we checked four large Shopify stores on 13 October, three returned one, so adoption counts increasingly measure platform defaults, not intent. Do instead: nothing, unless you publish developer docs that coding agents read.

2. “Schema or FAQ markup gets you cited” — myth

As a citation lever, the best test available says no. Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 against 4,000 matched controls: citations moved +2.4% in AI Mode and +2.2% in ChatGPT, both “statistically indistinguishable from zero”, and fell 4.6% in AI Overviews. A searchVIU test found AI chatbots “exclusively extract visible HTML content” when they fetch a page. Google says “there’s no special schema.org markup you need to add”, and FAQ rich results stopped appearing in Google Search on 7 May 2026. Do instead: keep structured data for rich results and to describe your organisation clearly, and put the answer in visible text. Our schema evidence review goes deeper.

3. “Rewrite pages into short Q&A blocks and FAQ sections” — myth

Google: “There’s no requirement to break your content into tiny pieces for AI to better understand it”, and “You don’t need to write in a specific way just for generative AI search.” The largest correlational dataset points the same way: SE Ranking found pages averaging 120–180 words between headings drew 4.6 ChatGPT citations on average, against 2.7 for sections under 50 words. In C-SEO Bench, a NeurIPS 2025 benchmark, most rewriting tactics were “largely ineffective” and often hurt ranking once several sites used them; the much-quoted “up to 40%” gain comes from a 2024 simulator, not live engines. Do instead: lead with a real answer in the first paragraph (moderate evidence: citations cluster near the top of pages), then write sections as long as the topic needs. An FAQ that answers real customer questions is fine for readers; just don’t expect it to win citations.

Access and inclusion

4. “Submit or list your business on ChatGPT” — myth

There is no submission form for websites or local businesses. OpenAI’s documented route into ChatGPT search is its crawler: “To help ensure your site appears in search results, we recommend allowing OAI-SearchBot in your site’s robots.txt file.” Local data also arrives through partners: Yelp’s August 2026 filing says its ratings and reviews “recently began powering ChatGPT’s local experience in relevant categories”, even though Yelp’s robots.txt blocks OAI-SearchBot from its business pages (we checked on 13 October). The one real application is for products: merchants apply at chatgpt.com/merchants, and Shopify and Etsy catalogs are already included. Our evidence file has no entry for this claim; we call it a myth because crawling is the only route OpenAI documents for websites. Do instead: allow OAI-SearchBot, then get listed where the data comes from (see our local guide and products guide).

5. “ChatGPT just uses Bing, so optimise for Bing” — weak

OpenAI runs its own search crawler, OAI-SearchBot, which it says “is used to surface websites in search results in ChatGPT’s search features”, and it doesn’t document any single outside engine. Third-party analyses describe a blend of OpenAI’s own index and external providers; none is official, and we couldn’t verify their figures. Bing indexing still helps (Copilot runs on it, and it may feed ChatGPT), so the claim isn’t worthless, just incomplete. Our evidence file doesn’t grade it; we put it at weak because no platform confirms it and the one official signal points elsewhere. Do instead: be crawlable by OAI-SearchBot and indexed by both Bing and Google.

6. “Blocking GPTBot or Google-Extended removes you from AI answers” — myth

Each is a separate control. OpenAI: “Each setting is independent of the others”; GPTBot governs training, OAI-SearchBot governs ChatGPT search. Google: “Google-Extended does not impact a site’s inclusion in Google Search nor is it used as a ranking signal in Google Search”, and AI Overviews and AI Mode are part of Search. Google’s real opt-out is the Search Console Search generative AI control, rolled out “to all websites worldwide” by 31 August 2026, or the nosnippet rule. The reverse mistake is real, though: Cloudflare’s stricter Block setting for AI training will “stop Applebot, Bingbot, and Googlebot from reaching your site — search included.” GPTBot vs OAI-SearchBot has the full matrix.

Rankings, reviews and recommendations

7. “We rank #1 in ChatGPT” — myth

AI answers don’t have stable positions. In SparkToro and Gumshoe’s test, hundreds of volunteers ran 12 recommendation prompts 2,961 times across ChatGPT, Claude and Google’s AI, and there was less than a 1-in-100 chance that ChatGPT or Google’s AI returned the same list of brands twice. The study isn’t peer-reviewed and Gumshoe sells AI tracking, but repeating the same prompt is exactly the test this claim needs, and the claim fails it. Microsoft’s 2026 Bing guidelines also say GEO doesn’t guarantee citations, as Search Engine Journal reported. Do instead: report the share of answers that mention you across many runs, per engine. How to measure AI visibility shows how.

8. “If I’m in the Google 3-pack, AI will recommend me” — weak

Partly true for Google’s own AI: Google says Business Profiles “can help your products and services to be visible in both AI responses and other Google Search results.” Much less true elsewhere. In SOCi’s 2026 study of more than 350,000 locations, “ChatGPT recommends just 1.2% of brand locations, compared to an average 35.9% appearance rate in Google’s local 3-Pack.” SOCi is a vendor and its sample is multi-location brands. Do instead: treat Google, Apple Maps, Bing and Yelp as four separate listings, and find which directories each assistant cites in your city.

9. “More reviews means AI will pick you” — weak

Reviews matter, but the data supports rating more than volume. Locations ChatGPT recommended averaged 4.3 stars in SOCi’s data, which looks like a threshold rather than a race to the biggest count. In forum threads, owners regularly describe losing to newer rivals with fewer reviews but more press and community coverage. No controlled study has isolated review count. Do instead: keep a steady flow of recent reviews on Google and your industry’s main platform, reply to all of them, and work on coverage beyond reviews.

10. “Publish your own ‘best X’ list with yourself at #1” — weak

It gets you cited more than it gets you recommended. Lily Ray tracked 100 B2B “best [category]” queries in Google’s AI Overviews at three dates between April and June 2026 and found that when a self-promoting brand’s own listicle was cited, the brand was left out of the recommendation “roughly two-thirds (69%) of the time.” It is observational, one researcher and Google’s surfaces only, hence weak rather than myth. Do instead: earn places on lists other people write, and keep comparison pages honest.

11. “Seed Reddit to get into ChatGPT” — weak

Genuine Reddit discussion does show up in AI answers; manufactured threads are a poor bet. Reddit said in July 2026 that its own AI catches about 25,000 spammy posts a day. Promptwatch’s tracking showed reddit.com falling from 3.83% of ChatGPT search citations (18 July to 7 August 2026) to 0.52% (14 to 17 August), a drop Promptwatch itself calls provisional. Google says “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem.” Do instead: answer questions in your field under your real name, and leave astroturfing alone.

12. “AI Overviews killed SEO” — myth

Google’s guide answers this directly. Is SEO still relevant? “In short, yes!”, because its generative AI features “are rooted in our core Search ranking and quality systems.” What changed is clicks. Pew’s analysis of 900 US adults’ browsing found people clicked a traditional result in 8% of visits to search pages with an AI summary, against 15% without one, and clicked a link inside the summary in 1%. So SEO is the foundation of AI visibility, and each ranking sends less traffic than it used to. Both are true.

Tools, ads and shopping

13. “A visibility tool shows what customers see” — weak

Tools sample prompts, usually logged out and from one location. Real answers shift with location, wording, memory and account history, and they vary run to run (see 7). Google warns: “Be wary of third-party tools that promise ranking success or claim to use ‘internal’ Google metrics. No third-party tool has access to our internal ranking or AI systems.” Trackers are useful for trends across many prompts, not as a window into customers’ screens. Do instead: pair any tracker with first-party data (Search Console, Bing AI Performance, GA4) and a “How did you hear about us?” question.

14. “Paying for ChatGPT ads will get us into the answers” — myth

OpenAI’s ads principles, published 16 January 2026, say: “Ads do not influence the answers ChatGPT gives you. Answers are optimized based on what’s most helpful to you. Ads are always separate and clearly labeled.” Its product-feed spec treats ad eligibility as “independent of search eligibility”, and its crawler documentation says OAI-AdsBot “only visits pages submitted as ads.” Ads buy a labelled placement, not a place in the answer.

15. “ChatGPT shopping pulls from Google Merchant Center” — myth

OpenAI’s merchant pages describe its own Agentic Commerce Protocol feeds, Shopify Catalog and Etsy. Its feed spec accepts a Google-compatible file format, but “only after OpenAI confirms it for your registered feed”; it doesn’t read your Merchant Center account. Shopify’s routing shows the split: ChatGPT and Copilot get products through Shopify Catalog, Google AI Mode and Gemini through Merchant Center. A related belief is out of date: since 24 March 2026, ChatGPT sends shoppers to the merchant’s own checkout by default rather than Instant Checkout. Details are in getting products into ChatGPT shopping.

Asking the AI, and what counts as success

16. “Tell ChatGPT about your business and it will recommend you to others” — myth

Memory and custom instructions are personal settings. OpenAI’s Memory FAQ places them under Settings > Personalization in your own account, where they shape your answers and nobody else’s. (OpenAI’s help centre blocks automated requests, so we read that page through its search-indexed text; our evidence file doesn’t cover this, and the grade rests on how the feature is documented.) The side effect is worse than useless: an owner who has told ChatGPT about their business sees it recommended in their own chats and assumes it’s working. Do instead: test logged out or in a temporary chat.

17. “Ask the AI why it didn’t recommend you, and it will tell you” — weak

A model will offer a plausible reason, but it has no reliable view of its own retrieval. Research on chain-of-thought explanations (Turpin et al., NeurIPS 2023) showed that models’ stated reasons can be systematically unfaithful to what actually drove their answers. The trustworthy evidence is external: the sources an answer cites, and Bing Webmaster Tools’ grounding queries, which Microsoft describes as “the key phrases the AI used when retrieving content that was referenced in AI-generated answers.” Do instead: read the citations and grounding queries, then fix those pages or get onto them. Why ChatGPT recommends your competitor walks through it.

18. “Cited means recommended means traffic” — myth

These are three outcomes, and studies measure the gaps directly. Cited isn’t recommended: in Lily Ray’s data, self-promoters were cited but left out of the recommendation 69% of the time. Cited isn’t traffic: Pew found users clicked a link inside Google’s AI summary in 1% of visits, and Bing’s AI Performance report counts citations without any click data. Do instead: measure mentions, citations, AI-referred sessions and leads separately, and judge GEO work on the last two.

What the evidence does support

These are the grades our audit uses for the same factors. They are less exciting than a file you can upload, and they work across every engine.

Our methodology lists what our audit deliberately doesn’t score (llms.txt, schema and FAQ markup as citation levers, chunking and rewriting for AI) with the sources behind each decision.

So is GEO a scam?

No, but a lot of what is sold as GEO is. The useful part overlaps heavily with good SEO and digital PR: let the right crawlers in, serve real content in HTML, keep it current, and get talked about on sites you don’t control. Google puts it bluntly: optimizing for generative AI search “is optimizing for the search experience, and thus still SEO.” The snake oil is the part that guarantees placement, sells llms.txt and FAQ schema as fixes, or reports a ChatGPT “rank”. One test works on any proposal: ask for the source behind each recommendation. For a start that checks only what has evidence, run a free audit, then see how to get ChatGPT to recommend your business for the off-site work.