Key points at a glance
- AI assistants mainly mention companies that they can describe and classify unambiguously on the web.
- In robots.txt, OAI-SearchBot controls ChatGPT search, whereas GPTBot only controls training.
- Clear facts about services, target group and region on your own website are the foundation.
- Mentions on trade portals, by associations and in the press strengthen the picture from outside.
- A mention can't be forced: test regularly with fixed questions and document the results.
How do AI assistants find companies in the first place?
ChatGPT and comparable assistants draw on two sources of knowledge. The first is what the language model learned during training. This knowledge is not up to date and often patchy for smaller companies.
The second source is a web search that the assistant carries out live for suitable questions. Pages are retrieved, evaluated and often linked as a source. For your company, this second route is the more important one, because you influence it directly with an easily findable website.
It is similar with Google: according to Google Search Central, there are no additional requirements for AI Overviews and AI Mode. A page must be indexed and eligible to appear in search with a snippet. Good classic search engine optimization therefore remains the basis, including for AI answers.
What matters is that an assistant can read from your content quickly and without room for interpretation: who the company is, what it offers, for whom and where. The more unambiguously this answer appears in several places on the web, the more likely a system is to mention you for a suitable question.
Let the right crawlers onto your website
A common mistake is a robots.txt that blocks all AI crawlers across the board. At first glance this seems cautious, but it can mean that your pages don't appear in ChatGPT search at all. It is important to distinguish between the individual crawlers, because they have different tasks.
According to its own documentation, OpenAI separates clearly: OAI-SearchBot makes sure websites appear in ChatGPT's search features. GPTBot collects content that may be used for training models. ChatGPT-User retrieves pages when a user specifically asks for them in a conversation. So you can allow search and still exclude training.
At Google, the corresponding control is called Google-Extended. It governs whether content may be used for training Gemini models and for certain grounding features. According to Google, Google-Extended has no influence on inclusion in Google Search and is not a ranking factor. For Claude, Anthropic also distinguishes between ClaudeBot for training, Claude-SearchBot for search and Claude-User for retrievals on behalf of a user.
Also check whether your hosting provider or a service such as a firewall or CDN blocks these bots regardless of robots.txt. This happens more often than you'd think and goes unnoticed without a targeted test.
| OAI-SearchBot | allow (visibility in ChatGPT search) |
|---|---|
| GPTBot | your own decision (model training only) |
| Claude-SearchBot | allow (visibility in Claude search) |
| Google-Extended | your own decision (no influence on Google Search) |
| Googlebot | allow (prerequisite for Google and AI Overviews) |
Illustrative example. Whether you allow training crawlers is a fundamental decision, not purely an SEO question.
Clear facts on your own website
Many SME websites describe their offer with general phrases such as innovative solutions or individual concepts. A human may still be able to infer something from that, a language model hardly. Instead, state concretely what you manufacture or provide, for which industries and in which region.
A page that bundles the most important facts compactly is helpful: company name, year founded, location, services, target industries, certifications and contact persons. This can be the About Us page, complemented by separate service pages, each with one clear topic.
Keep important content as real text and not only in images, PDFs or sliders. Google explicitly names this as a recommendation for AI features, and the same applies to other crawlers. Also make sure that information is identical everywhere, such as service names, address and company name.
A good test is the question: could someone from outside the industry say in one sentence what you do and for whom after one minute on your homepage? If not, a language model will have a hard time too.
- A separate page for each core service with a clear title
- Name specific industries, applications and fields of use
- Answer the questions customers actually ask in sales conversations
- Consistent spelling of company name, address and services
Structured data: helpful, but no magic trick
Structured data according to schema.org describes content in machine-readable form. For companies, the Organization markup or a suitable LocalBusiness type is particularly interesting. With it, you store name, logo, address, contact options and, via sameAs, the links to your profiles, for example on LinkedIn.
Google makes clear that no special markup and no special AI text files are needed for AI Overviews and AI Mode. However, structured data should match the visible text. So it is not a lever to force a mention, but a clean addition that reduces misunderstandings about your company.
After adding the markup, check it with Google's Rich Results Test. Faulty or contradictory information does more harm than good.
Why mentions on other websites count
When an AI assistant formulates a recommendation, it rarely relies only on the provider's own website. It draws on sources in which your company is mentioned in connection with a topic. These include industry directories, association websites, trade articles, regional press, trade fair appearances and review platforms.
For SMEs, there is often untapped potential here. Memberships in associations, references with partners, talks or expert articles often already exist, but can't be found online or contain outdated information. Check where your company is already mentioned, and make sure that name, service and website are correct there.
Bought link lists or mass-produced directory listings with no connection to your industry don't help, on the other hand. Relevant, credible mentions in your professional environment are worth more than many random ones.
How to test your visibility in ChatGPT
Define ten to fifteen questions that potential customers would realistically ask. For example: which providers of industrial coatings are there in England? Or: what should I look for when choosing a contract manufacturer for turned parts? Ask these questions in ChatGPT with web search activated, as well as in Perplexity and Gemini.
Note whether your company is mentioned, which competitors appear and which sources are linked. The sources are particularly revealing, because they show you which pages the systems use for the topic. Repeat the test every few weeks with the same questions.
Expect the answers to fluctuate. The same question may be answered differently the next day or in a different account. Individual results are therefore not very meaningful; only the trend across several tests gives a picture.
A look at your web analytics is also worthwhile: visits via links from AI assistants usually show up there as referrals from the respective domains. At Google, according to Google, clicks from AI Overviews and AI Mode are counted in Search Console's performance report under the Web search type, but not shown separately.
Realistic expectations of GEO
Be skeptical of promises that your company will be guaranteed a recommendation in ChatGPT. The answers of AI assistants can't be controlled directly, and the systems change constantly. Nobody outside the providers knows the exact weighting of the sources.
What you can influence are the basics: technical accessibility, clear content, consistent information and credible mentions. These measures also improve your visibility on Google and make your website more understandable for human visitors. So the effort is well invested even if AI answers still play a small role in your industry.
So don't plan GEO as a separate project, but as part of your ongoing work on your website and search engine optimization. Short-term tricks, such as hidden text for bots, are risky and violate search engine guidelines.
A pragmatic roadmap for the coming weeks
You don't have to tackle everything at once. For most mid-sized companies, a manageable sequence that can be implemented with existing resources is enough.
Start with the technology, because it is the quickest to check and is a prerequisite for all further steps. Then comes the content, because this is usually the biggest lever. Involve sales and management: nobody knows the actual customer questions better than the people who write quotes and hold conversations every day.
Schedule a fixed date each month for the test and record the results in a simple table. This way you can spot changes without being unsettled by individual answers.
- Check robots.txt and firewall settings for the crawlers mentioned
- Revise the homepage and About Us page for clear, factual statements
- Create or sharpen a separate, understandable page for each core service
- Add Organization markup with sameAs links and test it
- Update existing mentions with associations, partners and directories
- Create a fixed set of questions and test regularly