A missing mention is a useful reason to investigate. It is not a measurement of lost customers. Our recommendation is to build a small evidence record, check the business identity and public information behind it, and choose the next action from what you can verify.
ACE had the same problem
ACE ran this investigation on itself before recommending it to customers. On September 21, 2026, our first bounded discovery baseline did not confirm an ACE-owned AI citation. Brand answers also pointed at other meanings: generic agentic commerce, American Express ACE, and a different commerce domain. We preserved those as identity failures rather than counting the shared words as successful mentions.
After the accepted Warm-Demand site was deployed and its canonical sitemap was processed, Google Search Console later showed the ACE homepage, About page and AI Visibility Check methodology as indexed. In the next preserved organic brand captures, agentcommerceexchange.com appeared for the two tested brand questions.
The same bounded AI repeat still did not show an identity-bound ACE citation or commercial recommendation. ACE keeps indexing, organic discovery, AI mentions, AI citations, referral sessions, qualified leads and attributable outcomes as distinct evidence states.
We did not treat the timing as proof that a specific page change caused the organic result. We recorded the deployment, sitemap, indexing and query observations separately so the next repeat can show what actually changes.
Start with the question your customer would ask
Write down the service a customer needs and the place you actually serve. Use your current service pages or confirmed business information to establish that scope. Avoid adding nearby towns or specialist services merely to create more search phrases.
Keep branded and unbranded questions separate. A branded question asks whether the system recognizes your company. A service question asks whether it includes your company when someone is looking for help. Treat them as different tests.
For example, consider a hypothetical repair business serving Plano. An owner might test the business name, a question about the repair it actually offers, and a question combining that repair with Plano. This is an investigation template, not evidence about any real repair company.
Record the exact wording. If you later change it, keep the new question as a separate observation. Otherwise, how will you tell whether you are comparing the same thing?
Save the answer before diagnosing it
For each observation, retain:
- The question and date.
- The product or search surface used.
- Whether search retrieval was visibly demonstrated.
- The answer, named businesses and source links actually shown.
- Any sign-in barrier, incomplete response or uncertain identity.
Do not fill a missing AI answer with ordinary search results and label it an AI recommendation. Keep the ordinary results as separate evidence.
Read each business name alongside its domain and service description. Ask whether the answer refers to your company, a similarly named company or a broader concept. Avoid treating a familiar acronym as confirmation on its own.
This record gives you something concrete to discuss with a provider: “Here is the question, here is the answer, and here is the identity we expected.” That is our recommended starting point, rather than asking for a promise that your name will appear everywhere.
Check access without mistaking it for a recommendation
OpenAI distinguishes OAI-SearchBot, its search crawler, from GPTBot, which concerns potential training use. Ask your webmaster to check the relevant crawler access before agreeing to changes. Do not assume that enabling training access is the same task as making content available for search.
Keep the platforms separate here, too. Google says a supporting page in AI Overviews or AI Mode must be indexed and eligible for a Search snippet. It also says eligibility does not guarantee inclusion. Google’s documented eligibility requirements concern supporting links in AI Overviews and AI Mode.
For a Google observation, inspect the relevant page in Search Console. Record whether it is discovered, crawled or indexed. For a ChatGPT observation, record the answer and its visible sources rather than substituting the Google inspection for ChatGPT evidence.
Inspect the information a prospective customer needs
Read your public pages with the buyer’s question beside you. Can you point to a clear explanation of the service, intended customer, confirmed service area and next step? Does the business name agree with the domain and contact information? Are important qualifications or limits easy to find?
Treat gaps as editorial findings first. A vague service page is a reason to improve the explanation; it does not establish why a particular model omitted you.
Google recommends helpful, reliable content, internal links and crawl access for its AI search features. It does not require special AI files or special schema for those features. Before buying an “AI-ready” file or a large page package, ask what documented problem the proposed work addresses.
Turn the findings into a bounded next action
Return to the hypothetical Plano business. Suppose the answer names another company with a similar name, while the owner’s own service page clearly describes the requested repair. Record the identity mismatch. Do not rewrite the service page simply to manufacture activity.
Alternatively, suppose the owner cannot find a public page explaining the service at all. Prepare a useful explanation using confirmed business facts, then review it before publication. Label the expected benefit as clearer information for readers; leave any search benefit as something to measure afterward.
Preserve the original observation. After an approved change, repeat the same question and record what changed, including an unchanged answer. Do not translate a new mention into revenue without separate commercial evidence.
For help structuring the investigation, read what an ACE AI Visibility Check proves. The free ACE Visibility Check uses public evidence; no website credentials are needed to begin. Request your free check.
A worksheet for deciding what to investigate next
Use a separate row for each question. In the first column, write the buyer's task in ordinary language. In the next, retain the exact answer and the source links you could actually inspect. Then describe the unresolved question rather than assigning a cause prematurely.
If the answer names a different company, write “identity unresolved” and compare the domain and service description. If a source is missing, write “source not captured” rather than “no sources exist.” If the result requires sign-in, record that access limit and leave the answer field empty.
For the proposed action, specify what you would check or prepare, who owns the relevant information, and what would count as a completed check. For example, ask the business owner to confirm an ambiguous service area before drafting location copy. Ask the webmaster to inspect an access setting before recommending that it change.
Finally, add a column for the follow-up observation. Keep it empty until that observation happens. This distinction matters because a proposed explanation and a verified result belong in different parts of your decision record.