AI search can mention a business using newly retrieved sources or model knowledge. Business owners can check whether information is accessible and accurate, and whether visits follow. This work is commonly called GEO, Generative Engine Optimization; it does not guarantee recommendations.
How AI search picks who to mention
Different services retrieve and select sources differently. They do not publish their full ranking rules, and a citation cannot be promised.
1. Live search. For AI Overviews and AI Mode, Google specifies indexing and snippet eligibility. It may run several related searches, so source selection is not simply the first ten results for the original query. ChatGPT may use search partners; OpenAI recommends allowing OAI-Searchbot and its published IP addresses through network rules. Bing registration is not a published ChatGPT requirement.
2. Answer context. A brand mention can come from newly retrieved sources or model knowledge. Without checking the links, you cannot assume the origin of a particular claim. Accurate information on your site and independent profiles helps identify the business; no fixed mention count guarantees a recommendation.
GEO vs. SEO: what changes and what doesn't
Google's SEO fundamentals remain relevant. Other services have their own retrieval and selection systems; do not assume identical ranking factors.
An answer and a link are different outputs. An assistant may summarize information and link to a source. Clear wording helps readers check meaning, but questions, tables and structured data do not guarantee source selection.
The brand needs to be identifiable. Keep your name, services and contacts accurate and consistent across your website and relevant profiles. There is no substantiated universal ratio under which ten mentions outperform a hundred content pages.
What to actually do (in order of importance)
Six steps to a findable business
- Classic search indexing and snippet eligibility
- Crawler access
robots.txt, hosting and CDN let the crawlers you want reach public content - Citable content prices or a pricing process, FAQ, structured data, clear answers
- llms.txt a voluntary experiment, not a citation requirement
- Entity consistent details in profiles, directories and reviews
- Measurement visits, enquiries and a repeated sample of specific queries
1. The foundation: rank in classic search
For Google, verify indexing and snippet eligibility; for other services, check their own crawling requirements. Useful content and healthy SEO support discoverability, but a top-ten position is not a universal AI citation prerequisite. I cover the process in my SEO services and SEO pricing guide.
2. Verify crawler access and use diagnostics
Check robots.txt, hosting and CDN rules to ensure the search crawlers you want can retrieve public content. OpenAI separately specifies OAI-Searchbot access and allowing its published IP addresses. Use Bing Webmaster Tools to manage and diagnose Bing visibility; registration alone guarantees neither indexing nor ChatGPT visibility.
3. Content that can be cited
Review each page from the reader's perspective: can they find an answer, understand its conditions and verify it?
What makes content citable
- Actual prices or a clear pricing process including scope and exceptions; publish only valid figures
- FAQ sections phrased the way people actually ask
- Structured data Organization, Service: a machine-readable base that mirrors visible content, not a special AI lever
- Clear answers up front a summary in the first paragraph, detail below
4. llms.txt, a business card for AI crawlers
llms.txt is a voluntary proposal for a text overview, not a generally required AI-search standard. Google explicitly says no new AI text files are needed for its AI features. Evaluate potential benefits for other services separately. Treat it as a maintained experiment, not a citation requirement. This site provides such an overview, including an llms-full.txt version.
5. Entity: mentions, reviews, profiles
For local businesses, keep information discoverable and consistent. The following sources are useful to customers; this is not a published ranking of AI factors:
- Google Business Profile with reviews, current information and real customer experiences
- Directories, consistent name, address, website
- Relevant industry mentions and case studies with genuine context; do not create mass footer links for ranking
- Profiles (LinkedIn, Facebook), current contacts and verifiable business information
6. Measure what you can
Track visits and enquiries from identifiable sources, indexing status and a repeated sample of specific queries with a date and service recorded. Search Console includes Google AI features in the overall Web search type; it is not a separate inventory of every AI citation. Bing Webmaster Tools describes Bing, not everything ChatGPT sees. Individual answers can change, and a manual query does not measure the whole market.
Worked example: Mention ≠ citation ≠ visit
worksheet
| What I track | Evidence | Record |
|---|---|---|
| Business mention | Saved answer + date | To record |
| Website citation | Cited URL in answer | To record |
| Website visit | Analytics + identified source | To record |
| Enquiry | Received contact + source | To record |
This worksheet does not combine the results into one AI visibility score.
Where the Czech market actually stands
In the 22 August 2026 measurement of a purposive sample of 30 Czech e-shops, 5 of 25 readable robots.txt files explicitly mentioned a tracked AI crawler; no JSON-LD was found in the initial HTML of 7 of 23 homepages. An absent agent name does not establish owner inaction: wildcard User-agent: * groups and network rules can apply. Methodology and limits are in 30 Czech e-shops vs. AI crawlers.
Worked example: Availability is not a recommendation
OWN MEASUREMENT · PURPOSIVE SAMPLE
| Observation | Result |
|---|---|
| Sampled domains | 30 |
| llms.txt returns HTTP 200 | 8 / 30 |
| Text response rather than HTML | 4 / 30 |
| Causal effect on AI citations | Not measured |
The file alone does not prove an assistant will recommend you.
What to avoid
Do not trust guaranteed citations. A supplier can improve content and availability but cannot promise selection for a particular answer. Ask about measurement and its limitations.
Write for a real reader task. Explain the offer, conditions and evidence. A few manual queries do not reveal a private ranking algorithm.
Summary
Start with discoverable content, indexing and verified access for the crawlers you want. Then make the offer, contacts and sourceable information clear to people and search engines. Top-ten positions, Bing Webmaster Tools registration and llms.txt are not universal tickets into AI answers. Monitor specific queries and real enquiries; a citation cannot be guaranteed.
Want help with this? GEO preparation is a standard part of my SEO optimization work. And if you want your whole team to master AI tools, see company AI training.
Sources and verification
Requirements checked on 2 October 2026.