How to Test AI Visibility for Your Website

Test AI visibility in two stages: check whether your public pages are accessible, then record whether search-enabled AI services actually cite or recommend them for relevant buyer questions. Passing a technical check does not establish that your business appeared in an answer, and appearing in an answer does not establish that it produced a customer.

This guide provides a repeatable review for a small marketing or operations team. The examples are suggested tests, not reported results for Rebel Online or its clients.

What should you measure before changing the site?

Keep access, visibility and acquisition as separate records. Otherwise it is easy to describe a successful page fetch as an improvement in customer demand.

• Access: whether an important public page loads, exposes useful text and permits the relevant search crawler.
• Observed visibility: whether a recorded answer mentions your company, cites your URL or recommends your service, and whether the information is accurate.
• Acquisition: relevant visits, qualified enquiries and booked calls where the source can be established.

Choose questions that a prospective customer would ask before buying. For an AI operations company, useful starting points might include choosing a client-onboarding workflow, comparing an internal build with an existing tool, or deciding who should maintain an automation after launch. Avoid building the whole test around your brand name: that mainly tests whether a service can identify a company it has already been told to find.

How do you check technical access?

Start with the pages a buyer needs: the service page, a useful implementation guide, company information and the contact route. Confirm that each page loads publicly, contains the expected information and links to a relevant next step.

Review the canonical URL, any indexing directives, robots.txt and hosting restrictions. A public page may load for you while a firewall or challenge prevents a crawler from reading it. Where logs are available, verify requests using the provider's published guidance; a user-agent name alone does not prove who made a request.

For Google, use Search Console to inspect the important URLs when you have property access. A public search can reveal examples of pages being shown, but absence from a site: search is not a complete diagnosis. Record missing access to Search Console as a measurement gap rather than declaring the site unindexed.

The free Rebel AI Search Readiness Audit checks one public page's technical foundations. Its report does not test live answers from ChatGPT, Google, Perplexity or Claude, so complete the observation step separately.

How do you test ChatGPT search?

Ask a relevant question in a session where web search is available, request current sources and inspect whether search was actually used. ChatGPT has a built-in web search tool; the old description of requiring third-party browsing plugins is outdated. Availability can depend on workspace settings. See OpenAI's web search documentation.

Record the response and open any cited source that appears to be yours. Check whether it links to your domain or merely mentions your business while citing someone else. If you supply your own URL and ask for a summary, log that as a direct retrieval test, separately from an unaided buyer question.

For crawler controls, OpenAI distinguishes OAI-SearchBot, used for search, from GPTBot, used for potential model training. ChatGPT-User supports certain user-triggered visits. Use OpenAI's crawler documentation to review these controls without assuming that all OpenAI requests have the same purpose.

A response without your citation does not reveal whether your content was in a model's training data. Do not infer training inclusion, a knowledge cutoff or a crawl failure from that observation.

How do you test Google's AI search features?

Record Google Search AI Overviews and AI Mode as distinct surfaces. Note which feature appeared, the question and any supporting links. Keep a Gemini app test in its own record instead of treating every Google product as one search result.

Google's guidance connects Search AI eligibility to its search foundations, including indexing and snippet eligibility. It does not guarantee selection for a particular answer. Google's AI search guidance explains this distinction.

In Search Console, check the property's Search generative AI control and whether it inherits a choice from a parent property. Google's control documentation explains inclusion and exclusion. Do not change an existing publisher preference as a side effect of an audit.

Where available for the property, review the Generative AI performance report for reported impressions. It reports supported Google features, not all AI services. Keep those impressions separate from analytics visits and qualified leads. A missing report does not by itself prove there were no mentions elsewhere.

How do you test Perplexity and Claude?

Use search-enabled sessions and inspect the sources attached to each answer. Save the exact question and distinguish a citation to your own page from a mention supported by another website.

Perplexity documents PerplexityBot for search discovery and Perplexity-User for user-initiated requests. Their roles and access behaviour differ. Review its crawler documentation when diagnosing a suspected block, including any hosting restrictions.

Claude does offer web search. Anthropic announced it in March 2025 and subsequently expanded availability; when it uses web information, it can provide citations. See Anthropic's web search announcement. Confirm what is available in the session you use, then log a search-based result separately from a response generated without retrieval.

For either service, pasting your content into the conversation proves that the service can use supplied content in that conversation. It does not prove that a prospective customer will discover the page independently.

What should the observation log contain?

Use the same small set of questions each week, and keep exploratory questions in a separate group so the comparison remains understandable. Record:

1. Date, exact question, language and intended buyer need.
2. Platform, mode, relevant session context and whether web search ran.
3. Company mentioned, URL cited and whether there was a service recommendation.
4. Accuracy of the business description and whether the citation supports the answer.
5. Next action, with the page or access issue it addresses.

For an internal-systems provider, an example question is: "What should a small agency include in a scope for automating client onboarding?" Follow-up questions can cover integrations, approval steps, ownership and maintenance. These are buying topics, not evidence that any particular service currently recommends Rebel.

How do you turn the findings into customer acquisition work?

Choose the next change from the problem you observed. Fix an access failure at the relevant layer. Correct a misleading company description. When the site lacks a useful answer to a buyer question, publish a documented method or worked example with a relevant service link.

If a page receives relevant visits but few enquiries, inspect its offer and contact route before adding more content. State who the service helps, what information is needed to scope the work and what the prospective customer should do next. Use authorised customer evidence when available; label demonstrations and estimates clearly.

For AI operations buyers, Rebel's AI Operations service is the relevant place to discuss an internal workflow or system. For a search visibility implementation, see AI Search Optimization. Route the reader to the service that matches the problem they are trying to solve.