AI visibility · Measurement
How to check whether ChatGPT recommends your business
Test the questions a customer asks before they know your name. Use the same prompts and conditions to build a record you can compare after making changes.
01 · Purpose
Separate being mentioned from helping a buyer choose.
- Presence: is your business named in an answer to a relevant question that does not name you?
- Recommendation: is it included as a provider the buyer could consider?
- Accuracy: are the service, market, buyer type and limitations described correctly?
- Sources: which pages are linked, and do they support the description?
- Alternatives: which other providers appear, and what reasons does the answer give for them?
A citation, an accurate description and a recommendation answer different questions. Keep them separate from website visits, submitted enquiries and sales.
02 · Preparation
Fix the questions and conditions before the first run.
Choose one service, one market and a defined buyer. Prepare ten to fifteen questions in the language that buyer would use. Keep questions naming your business in a separate brand-accuracy group.
For each observation, record the date and time, system, visible model label, search mode, sign-in state, language, stated location, exact question and complete first answer. Mark any location or setting you cannot control as unknown.
Start each observation in a fresh conversation. Save the first answer before asking follow-ups. If you then reveal your company name, record the follow-up separately rather than counting it as unprompted discovery.
Repeat each discovery question three times per available system. Ten questions across two systems mean sixty planned observations. Record how many you actually complete.
03 · Questions
Cover the decisions a customer needs to make.
| Intent | Example question | What to look for |
|---|---|---|
| Discovery | Which providers of [service] in [market] are worth comparing? | Whether your business appears before the buyer knows its name. |
| Problem | Who can help with [specific problem] in [market]? | Whether the answer connects your offer to a real need. |
| Fit | Which provider suits [buyer type] with [constraint or goal]? | Whether your audience and limits are described accurately. |
| Comparison | Compare providers of [service] using [three buying criteria]. Explain the reasons and sources. | Which differences the answer uses to help the buyer choose. |
| Validation | What should I check before choosing a provider of [service]? | Whether your content supports the decision criteria. |
| Brand accuracy — separate check | What does [company] do, who does it serve, and what does it not offer? | Whether the description is correct. Exclude this from unprompted discovery results. |
04 · Recording
Keep a row for each completed answer.
Use the worksheet to record mention yes/no, recommendation yes/no, description accuracy, other named providers and checked source URLs. Retain the complete answer or a reference to its saved capture.
Open each source and record its final destination. Check that the page supports the relevant statement. A visible source label alone is not a verified citation.
For a simple coverage measure, divide completed discovery answers mentioning the company by all completed discovery answers in the same system and test. For example, 6 of 30 answers is 20% of that sample. This is an illustrative calculation, not an AXO-8 result or market share.
Report completed and planned observations together. A blocked or failed run remains unavailable. Keep named-brand checks outside the discovery denominator, and show results for each system separately.
05 · Interpretation
Use the findings to choose what to inspect next.
Your company and pages do not appear
Inspect access, indexing, the page serving the buyer’s need and the evidence explaining your offer. The answer alone cannot identify the cause of the absence.
Your company appears only when you name it
Check whether your public content clearly connects your company to the service, location and buyer problem. Brand recognition and discovery are different observations.
The description is wrong
Check the linked sources for outdated services, locations or company details. Correct inaccurate information on the pages and profiles you control.
Another provider appears repeatedly
Inspect the actual sources and reasons in the recorded answers. Use them to identify unanswered buyer questions or missing evidence in your own material. Repeated appearance does not establish which factor caused it.
07 · Sources
Check the platform guidance alongside your observations.
- OpenAI: Overview of OpenAI crawlers — distinguishes OAI-SearchBot, used for search, from crawlers with other purposes.
- Google Search Central: optimizing for generative AI features — covers search foundations, useful original content and measurement.
- Google Search Console — inspect Google indexing and search performance.
This English guide adapts our original Slovenian testing guide. The separate 17 October AXO-8 baseline summary records our initial method and observations.
Common questions
Practical answers before you start.
- How do I check whether ChatGPT recommends my business?
- Write the questions a potential customer would ask without knowing your company name. Run the same questions in fresh conversations, save the complete answers, and record mentions, recommendations, description accuracy and verified source URLs separately.
- Why should every observation start in a fresh conversation?
- Earlier messages and account context can influence the answer. A fresh conversation reduces conversational carry-over. Record sign-in status, model, search mode and location settings because a new conversation does not make every condition identical.
- Does a mention without a link count?
- Record it as a mention. A recommendation and a checked citation are separate observations. Open each source, record the final URL, and check whether its content supports the claim.
- What if a system is unavailable or requires sign-in?
- Mark the observation as unavailable or not run. Exclude it from the denominator of completed answers. An access failure is not an answer with zero mentions.
- Does the free AXO-8 scanner run this test for me?
- No. The free scanner evaluates web and search signals. Testing actual AI answers is a separate process using recorded questions, repeated observations and source checks.
Your next step
Turn your findings into a practical next step.
Review the scope, pricing and measurement approach for AXO-8 AEO/GEO implementation. Bring your website, target market and recorded buyer questions to the conversation.