If your brand does not appear in a ChatGPT answer, first check what was asked and whether the answer used web search. One missing mention does not establish a technical fault or lost revenue. I work through access, content, sampled answers and the enquiry journey before recommending a change.
These seven checks help separate a fault you can reproduce from a visibility hypothesis. They apply to existing pages: you do not need a new page for every variation of a prompt.
1. Check access for the relevant search crawler
Search access and permission to use content for training are separate decisions. OpenAI's crawler documentation assigns ChatGPT search to OAI-SearchBot, training to GPTBot, and user-triggered fetching to ChatGPT-User. Allowing a training crawler is not a prerequisite for search visibility.
Other providers also distinguish roles. Anthropic separates Claude-SearchBot, Claude-User and the training crawler ClaudeBot. Perplexity uses PerplexityBot for search and Perplexity-User for user requests. Google Search uses Googlebot; Google-Extended does not control Search inclusion.
Check robots.txt and the actual response to verified crawler requests. A robots allow rule does not establish that your CDN or firewall lets the request through. A logged block or challenge is an observed access fault. Without request logs or a reproducible fetch, crawler blocking remains a hypothesis. Review a specific verified rule rather than disabling security protection wholesale.
2. Check indexation and the content returned
Confirm that the intended page returns successfully, its canonical points to the right owner and its indexing controls match your intention. Use Search Console's URL Inspection for Google, and inspect the content the server returns as well as the rendered page.
If the answer is missing from both, you have a content-delivery fault to investigate. If it appears only after interaction, check the behaviour of the particular search system rather than assuming all JavaScript is invisible. A successful fetch alone does not prove indexation or citation eligibility on every platform.
I include these checks in a technical SEO audit. I cannot diagnose an indexing problem from a missing brand mention alone.
3. Make the relevant answer clear and support it
Give a reader enough information to judge whether your business fits the question. For a local service, that might include the area you serve, what the work covers, who carries it out and the published price or pricing basis. Link evidence to the claim it supports, and explain any limits.
Read the section without the rest of the page. If the subject, condition or source disappears, repair that context. This is a useful editorial check, without a fixed opening word count or a requirement to rewrite everything as a FAQ.
Google's guidance requires no special AI schema or fixed chunk size. Keep standard markup consistent with visible facts. A missing FAQPage block is not, by itself, an explanation for missing AI visibility.
Diagnosis
An observed fault
A verified search request is blocked, a canonical points to the wrong page, or the published service information is incorrect. Record the evidence and fix that specific problem.
Diagnosis
A hypothesis to test
The brand is absent from a sampled answer. Access may be fine, and the result may depend on the prompt or mode. Save the answer before deciding what needs changing.
4. Repeat the check with recorded prompts
Use realistic questions from customers or search data. Record the full prompt, product, mode, date and relevant locale. Repeat a small comparable set and save the returned answers. Label generated research prompts separately from observed customer questions.
Record whether your URL was cited, your brand was mentioned and your brand was recommended. A citation supporting another provider's recommendation is still a citation, but it is not a recommendation for you. A failed or unavailable run is missing evidence, not a negative answer.
Repeated checks can show what happened in that sample. They cannot reveal a platform's complete selection logic or establish your share of all answers. Keep those limits attached to the result.
5. Check the cited sources and business facts
Open the sources an answer actually cites. Do they support the claims, describe the right business and contain current service details? An incorrect name, service area or price is a specific accuracy problem. The reason an assistant selected one provider over another may still be unknown.
Correct information you control and pursue genuine corrections on external listings where appropriate. Consistent entity information helps people verify your business. It does not guarantee a citation, and a quantity of forum posts or mentions is not a proven threshold.
Useful external coverage should reflect actual work, research or customer experience. Do not manufacture reviews or publish unsupported claims just to seed an answer.
6. Measure exposure, visits and actual enquiries
Use each report for what it measures. Search Console's Generative AI report includes AI Overview and AI Mode impressions. Bing AI Performance reports citations across supported surfaces and sampled grounding queries. Neither number is a count of enquiries.
In GA4, inspect session sources such as chatgpt.com and perplexity.ai. Visits without identifiable referral information cannot simply be relabelled as AI traffic. Ask enquirers how they found you, while keeping self-reported discovery separate from analytics attribution.
A missing brand mention needs investigation. A citation needs checking. A form success needs reconciliation with a real enquiry before it becomes evidence of commercial value.
7. Check the journey to the intended offer
Follow the action from the cited page on desktop and mobile. Check that the offer, deliverables and price are clear, controls work, and a successful submission produces a useful confirmation. Test error states with a mock backend, so the check does not send a real enquiry.
Then review actual records for duplicates, tests and relevant prospective customers. A recorded form event is not evidence that the enquiry was qualified or became paid work. If records are unavailable, keep the commercial outcome unknown.
How long does an improvement take?
There is no guaranteed citation timetable. A corrected page may need to be recrawled or reindexed before a search answer reflects it. Answers based on training are a different case, and editing a website does not schedule a model update.
Record publication and any observed recrawl separately. Compare later answers with the original sample, and log other changes that could affect the result. Better information is worth keeping even if a small sample cannot show a visibility change.
For an initial diagnostic, try the free AI visibility check. Treat its output as that tool's observation rather than a substitute for repeated consumer-product checks. My AI search optimisation service combines access, evidence and measurement. If you want to compare providers first, see the best AEO agencies in the UK.
