A company appeared in an AI answer, but the link led to a directory. A month later, the same question produced no mention. These observations are useful when you save the question, service, date, and answer. Without them, the number of “AI mentions” is difficult to verify and compare.
Generative Engine Optimization, or GEO, generally refers to work on how information about a company and its pages appears in AI answers. There is no single set of factors that applies to all systems. It is easier to start with things you can verify: what information has been published, whether pages are accessible, and what exactly appears in the answers you receive.
This article was reviewed on 7 October 2026.
Distinguish mentions, links, and visits
A mention is the company or website name in the text. A link is a URL that can be opened. A citation connects a claim to a stated source. A visit is an actual visit to the website from the answer. One observation does not prove the others.
In a hypothetical example, an answer names the store “Sever,” describes its delivery, and links to a business directory. The name is there, but there is no link to the store itself, and the number of visits to the store from that answer is unknown. Record all three results separately.
| What you observe | What to save | What it does not prove |
|---|---|---|
| Company name in an answer | Exact excerpt and question | That it refers to the company you mean |
| Link to the company’s own website | URL and destination page | That the user visited the website |
| Link to a third-party source | URL, source excerpt, date | That all information in the answer came from it |
| Visit in analytics | Visit, source, conversion action | That all answers mentioning the company bring visitors |
If the name is absent, it is absent from that specific saved answer. You cannot conclude from one check that “AI never knows about the company.”
Check page accessibility and content
Choose a page where a visitor can verify the relevant information: a service, operating region, price, or terms. Open it without signing in, and check the HTTP response, whether the main text is accessible, robots, canonical, and internal links. If the source is blocked or outdated, first fix the confirmed issue.
AI Overviews and AI Mode in Google Search follow the usual search requirements. Google explicitly states that no special AI markup or separate text file is required for inclusion. However, meeting the requirements does not guarantee that a page will be included. These conditions apply to Google Search products; they should not be assumed to apply to all chatbots. See Google’s official requirements for AI features in Search.
Checking a single page with the reChecker technical audit can help identify barriers to its accessibility and SEO settings. It does not measure the likelihood that a particular model will select the page in the future. Compare the results with the content that visitors and search crawlers can access; for JavaScript sites, see the guide to indexing dynamic content.
Account for each crawler’s purpose
Search access and the use of data for training are different things. For example, OpenAI describes OAI-SearchBot as a crawler associated with discovering content for ChatGPT search, and GPTBot as being associated with possible use of content for training. They should be managed separately, using the current rules in OpenAI’s publisher FAQ.
Check robots.txt, CDN restrictions, and anti-bot protection. Simply spoofing a User-Agent in a manual request does not prove that the real crawler is allowed access from its IP address. Confirmation requires actual server logs and the crawler verification methods described by its operator.
Do not make service pages public for the sake of AI visibility. If information is available only to customers, first decide which parts can be made public. Robots.txt controls crawling and does not replace authentication.
Googlebot and Google-Extended also have different roles: the second token controls specified Gemini/Vertex AI use cases and does not determine inclusion in Google Search. Check Google’s crawler list. Perplexity’s search crawler and its user-requested fetcher are distinct, as described in Perplexity’s documentation. A practical process for checking robots, noindex, nosnippet, and HTML is covered in the AI site audit.
Make company information verifiable
On a service page, describe who the service is for, what the work includes, what terms apply, and what the price depends on. On the company page, state the company name, domain, contact details, and region of operation. If there are organizations with the same name, add genuine distinguishing details.
A hypothetical statement such as “we implement CRM for small online stores; integration with the listed systems is priced separately” gives readers more to work with than “the best digital solutions.” Specific information is useful regardless of whether the page appears in an AI answer.
Keep information consistent across website sections and publicly available company profiles. When a price changes, remove outdated values from your own pages or clearly explain the date and terms of an archived offer. Ask the owner of a third-party directory to correct it. Even after a source is updated, you cannot promise an immediate change to an AI answer.
Markup should describe what readers can actually see. Do not add fabricated reviews, ratings, or FAQs to try to influence a model. FAQs are useful to readers, but Google FAQ rich results were discontinued on 7 May 2026 — official announcement. See Google’s general guidelines for structured data.
Create a repeatable set of questions
For monitoring, choose several tasks that reflect what a real customer might do: find a provider in a specific region, compare options against a condition, or check a service’s limitations. Ask a separate question using the company name if you want to check brand recognition and factual accuracy. An answer to a branded question is not proof of a recommendation for a general query.
Save the wording and note the date whenever you change it. If one set of questions was about building a website and the next was about promoting a store, differences in mentions may be explained by the change in task. Comparisons require the same questions and comparable conditions. A newly added question starts a new line of observation.
Manual tracking template:
Дата и время:
Сервис / доступный режим:
Точный вопрос:
Название компании в ответе:
Совпадают ли домен, регион и деятельность:
Ссылки и страницы назначения:
Неточные утверждения:
Следующая проверка / необходимое действие:The reChecker subscription includes monthly checks across seven services: ChatGPT, Alice, DeepSeek, Gemini, Claude, Grok, and Perplexity. You can set up to five questions for a website and see which questions and answers were included in the check. This monitors the specified set of questions; it does not cover all user questions and is not a separate check of Google AI Overviews.
Make decisions based on confirmed findings
If an answer confuses the company with another, check its identifying details and cited sources. If it gives an old price, find the current offer and any potentially outdated publications. If there is no link to the website, check whether there is a suitable public page with the relevant answer. If no source is stated, do not guess which source the model used.
Record page changes and the date of the next check. An increase in mentions after an edit is an observation, not proof of a causal relationship: the answer may have changed because of its sources, mode, model, or other conditions. A decrease also requires reviewing the saved answers.
Search rankings, AI answers, and visit analytics complement one another. Do not use an increase in total Search Console impressions as confirmation of how often chatbots mention a company. For Google AI features, see the Generative AI performance report (Search): Google announced its global launch on 31 August 2026. It lets you analyze impressions by page, date, country, and device; the report may be unavailable if there is insufficient data. It has not been verified here whether this report is available in the reChecker account. It does not cover all answers from third-party chatbots.
The practical outcome is an accessible page with accurate information, a saved set of questions, and a list of confirmed discrepancies. The next action should address a specific factual or technical issue, rather than promise to “get into every AI system.”