7 min read
5 GEO myths that can cost you money
Five false ideas about GEO, checked against documentation from Google, OpenAI, Anthropic and Microsoft and against published studies.
Short answer. Five GEO myths can cost you money: that it doesn’t need SEO, that you need an llms.txt file or special schema, that blocking every AI bot is free, that asking ChatGPT once is enough, and that someone can guarantee you the top spot. All five clash with official documentation from Google, OpenAI, Anthropic and Microsoft, or with studies that publish their method.
We checked every fact in this article against its source on 25 October 2026.
Can GEO work without SEO?
No. When an AI assistant searches the web, it answers from what it finds in a search index, and that index is fed by the same things as SEO: pages that can be crawled, are indexed and are clear. Google says it plainly: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” The page has to be indexed and eligible to appear in Search with a snippet.
It’s the same outside Google. Microsoft explains that Copilot, with web search turned on, sends queries to the Bing search service to ground its answers. And when ChatGPT doesn’t search, it answers from what it learned in training.
34.5% of ChatGPT prompts triggered a web search in February 2026, down from 46% at the end of 2024. Semrush, US sessions from its clickstream panel.
What it costs: paying for “GEO” that forgets about indexing. If Google or Bing don’t have your page right, the AI that searches them doesn’t either.
Do I need an llms.txt file or special schema for AI?
No. Google says so in its guide to AI features: “You don’t need to create new machine readable files, AI text files, or markup to appear in these features.” It adds that there’s no special schema.org structured data you need to add either.
Independent studies agree. SE Ranking analysed around 300,000 domains and found no link between having an llms.txt file and getting cited by AI. Ahrefs compared 1,885 pages that added schema with a control group, 30 days before and 30 days after the change. These were pages AI already cited a lot.
−4.6% in AI Overview citations after adding schema, compared with the control group. In AI Mode (+2.4%) and ChatGPT (+2.2%), the effect can’t be told apart from zero. Ahrefs, May 2026.
Ahrefs warns that both groups were already declining beforehand, so the drop may reflect that trend rather than the schema. Its conclusion: schema had no clear effect, positive or negative.
What it costs: development hours spent on files that don’t move the needle today.
Does blocking AI bots protect my site at no cost?
It depends on which bot you block. OpenAI, Anthropic and Google separate the crawlers that train models from the ones that power their search.
| Crawler | What it’s for | If you block it |
|---|---|---|
| OAI-SearchBot (OpenAI) | ChatGPT search | Your site isn’t shown in ChatGPT search answers |
| GPTBot (OpenAI) | Content that may be used to train its models | You ask for it not to be used for training; OpenAI says each setting is independent |
| Claude-SearchBot (Anthropic) | Improving Claude’s search results | Anthropic doesn’t index your content for its search, which may reduce your visibility |
| ClaudeBot (Anthropic) | Content that may be used for training | You ask for your future content to be left out of training |
| Google-Extended (Google) | Training future Gemini models and grounding answers in the Gemini app and Vertex AI | Doesn’t affect your inclusion in Google Search and isn’t a ranking signal |
OpenAI spells it out: you can allow OAI-SearchBot to appear in search results and block GPTBot. Changes to robots.txt take about 24 hours to show up in its search.
What it costs: a “block anything that sounds like AI” rule can take you out of ChatGPT and Claude without anyone noticing. If you don’t want your content used for training, block the training crawlers and let the search ones through. We explain how in our guide on whether your website is blocking AI.
Is asking ChatGPT enough to know whether I show up?
No. AI answers change from one time to the next, and a single question tells you almost nothing. SparkToro and Gumshoe analysed 2,961 answers from ChatGPT, Claude and Google’s AI to 12 prompts, with around 600 volunteers.
Less than 1 in 100: the chance that ChatGPT or Google’s AI repeats the same list of brands in two answers. In the same order, the chance drops below 1 in 1,000. SparkToro and Gumshoe, January 2026.
On top of that, everyone asks in their own way. In that study, 142 questions written by people on the same topic (choosing headphones for a relative who travels) barely resembled each other: their semantic similarity was 0.081.
That’s why you have to measure with many answers, always to the same questions, and give every percentage with its margin of error.
If you appear in 3 out of 10 answers, your real presence could be anywhere between 11% and 60% (95% Wilson interval). With 29 out of 96, the range narrows to 22–40%.
What it costs: making investment decisions from a screenshot. One day you show up and the next you don’t, without anything having changed.
Can anyone guarantee I’ll come first in AI?
No one. Google says it about classic search: “No one can guarantee a #1 ranking on Google.” With answers that change on every query, even less so.
There are no shortcuts either. The famous “up to 40% more visibility” comes from the paper that coined the term GEO (KDD 2024), and it was measured in a benchmark, not in real businesses. A later benchmark, C-SEO Bench (NeurIPS 2025), concluded that most of the methods it tested are largely ineffective and often hurt rankings.
And tricks carry risk. Google’s definition of spam includes “attempting to manipulate generative AI responses in Google Search”.
What it costs: paying for a promise no one can keep and, at worst, exposing yourself to a penalty.
So what does work?
What works is less flashy, but measurable:
- Access: search crawlers can read your site as text.
- Consistent facts: name, services, prices and opening hours that match on your site, your Google Business Profile and directories.
- Content that answers: one page per real question, with the answer up front and a visible date.
- Honest measurement: a fixed panel of questions, with its margin of error.
A typical case: a language school publishes its prices only in a PDF and, without knowing it, blocks OAI-SearchBot. ChatGPT search won’t show its site in its answers. Unblocking the crawler and putting the prices in text is a better starting point than any new file.
Frequently asked questions
Is GEO the same as SEO?
No, but it needs it. SEO aims to get your page into the results; GEO aims to get your business named inside the AI answer. According to Google, there are no additional requirements: the page must be indexed and eligible to show with a snippet.
Should I create an llms.txt file?
Google says it is not necessary, and SE Ranking found no link between having one and getting cited across around 300,000 domains. If someone pitches it as a priority, ask what improvement they expect and how they will measure it.
Does blocking GPTBot take me out of ChatGPT?
No, according to OpenAI: each setting is independent. GPTBot relates to training; what counts for appearing in ChatGPT search answers is OAI-SearchBot.
How many answers do you need to measure AI visibility?
It depends on the margin you accept. At 95% confidence and in the worst case, 10 answers give about 31 points either way; 96 give 10 points; and 384 give 5 points. Those answers can come from many different questions or from repeating the same ones.
Does any of these myths affect you?
Our free diagnostic checks in a few seconds whether AI crawlers can get into and read your site. How the main assistants describe you today is what we measure in the full diagnostic, with data and its margin of error, and with no ranking promises.
If you then want us to make the changes and improve them month by month, we explain how in services. And if you’re missing the basics, start with what GEO is.
Sources
[A]: official source or peer-reviewed paper. [B]: industry study or open dataset with a published method.
- [A] Google Search Central: AI features and your website (updated 10 Dec 2025) accessed on 25 October 2026
- [A] Microsoft Learn: Data, privacy, and security for web search in Microsoft Copilot (updated 18 Aug 2026) accessed on 25 October 2026
- [B] Semrush: ChatGPT traffic analysis: Insights from 17 months of clickstream data (7 Apr 2026; US sessions) accessed on 25 October 2026
- [B] SE Ranking: Does LLMs.txt impact your AI visibility and citations? (15 Oct 2026; around 300,000 domains) accessed on 25 October 2026
- [B] Ahrefs: We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved. (11 May 2026) accessed on 25 October 2026
- [A] OpenAI: Overview of OpenAI crawlers (no visible date) accessed on 25 October 2026
- [A] Anthropic, Claude Help Center: Does Anthropic crawl data from the web… (7 Apr 2026) accessed on 25 October 2026
- [A] Google Search Central: Google's common crawlers, Google-Extended (updated 15 Oct 2026) accessed on 25 October 2026
- [B] SparkToro and Gumshoe: AIs are highly inconsistent when recommending brands (28 Jan 2026; updated 28 Jun 2026) accessed on 25 October 2026
- [A] Google Search Central: Do you need an SEO? (updated 5 Jun 2026) accessed on 25 October 2026
- [A] Aggarwal et al., “GEO: Generative Engine Optimization”, arXiv 2311.09735 (KDD 2024) accessed on 25 October 2026
- [A] Puerto et al., “C-SEO Bench: Does Conversational SEO Work?”, arXiv 2506.11097 (NeurIPS 2025, Datasets & Benchmarks) accessed on 25 October 2026
- [A] Google Search Central: Spam policies for Google web search (updated 28 Aug 2026) accessed on 25 October 2026
Also available in: Castellano Català Français