Practice / 320px

How to check what AI answers say about your business

By the 320px team · 17 November 2026

An AI answer can sound confident while naming the wrong service, an outdated price or another developer. The mistake may be in the answer, your website or an external listing. Each needs a different response.

The aim is not to invent a visibility percentage from one day’s results. It is to understand what a potential customer sees, what evidence is available and which source your team can actually correct.

Start with customer questions

Choose a few questions your website should answer: who maintains an existing website, how enquiries reach a CRM, or what a particular service includes. Add a question with your company name, but do not put it in every prompt. Repeating a name supplied in the question is different from discovering a business.

Save the exact question, date, service and mode, including whether web search was used when that is visible. Keep search-assisted answers separate from answers without search. If a service provides no sources, record them as unknown; do not assume it did or did not use a particular website.

Compare the answer with its sources

Save the complete answer, open the available links and compare the answer with the current pages. An old price in an answer calls for a check of its source; an outdated price on your own page calls for a correction there. A third-party listing that credits the wrong developer needs an approach to that listing’s owner. If no source is available, do not guess where the error came from.

For example, our team found a third-party catalogue that named another developer for our product and sent a correction request. At the time of the request the change was not confirmed. This is a documented observation, not a claimed improvement in AI answers.

Check the page a customer should find

Can a visitor see the current offer, limitations, provider and contact route? Does the link work? If an answer depends on a plan, version or separate piece of work, explain that next to the main statement.

Check the server response, canonical URL and crawl rules for public pages. Do not expose accounts, enquiries or personal data, or disable site protection wholesale for one crawler. A User-Agent can be imitated. To confirm a real visit, use the crawler owner’s verification method and the logs available to you. A page opening in an ordinary browser is not proof that a particular bot visited it.

Correct the source and check again

Update confirmed facts on your own site and check related pages. For an external listing, use the existing account or contact the publisher; do not create a duplicate while ownership is unresolved. For a stale search result, compare the last crawl with the page change and use the available webmaster tools. A re-crawl request being accepted does not mean the search index has already changed.

Record the corrected URL and date. Repeat the same questions in the same services and modes. Answers can vary between runs: one improvement, or one answer without a mention, does not establish a market-wide visibility share.

Give the team an actionable record

A short table is enough: question and mode, date, answer, available sources, the discrepancy, the person responsible for the next action and the result of the repeat check. Distinguish completed corrections from external requests and unknowns. Leave out keys, private correspondence and customer data.

320px can agree a suitable set of questions, sources, observation period and report format for an existing website. Search positions and inclusion in AI answers cannot be guaranteed.

Sources: OpenAI crawler documentation, Yandex important-page monitoring and Yandex re-crawl status.

Discuss your website →