To improve your chances of being cited by ChatGPT, make the page that answers a real buyer question accessible, accurate and supported by evidence. If your business is missing, first save the question, answer and linked sources. Check whether the question asks for the kind of provider you are before commissioning more content.
Then inspect the relevant page: does it explain your service or product, its limits and the proof a buyer needs? Fix the gap you can substantiate and repeat the check under recorded conditions.
Count mentions, citations and recommendations separately. None guarantees a qualified enquiry, and no page edit can guarantee that ChatGPT will choose your source.
Check whether you asked the right question
Suppose you run a Salesforce consultancy. “Which CRM should we use?” asks about software. “Which consultancy can implement Salesforce for our business?” asks about a delivery partner. If your test returns Salesforce or HubSpot as the answer, first check whether it tested the service you sell.
The same distinction matters for equipment. A question about how pneumatic cylinders work can be answered without naming a supplier. A question about sourcing a suitable replacement needs different evidence: model compatibility, availability, application limits and delivery coverage.
Start with the questions your sales team hears. Write down the buyer's situation and what would make a recommendation useful. If you construct a question yourself, label it as a test question. It does not become evidence of customer demand just because an AI system answers it.
Record more than the missing name
Save the exact question, answer, date, product and target market. Note whether you used the consumer app or an API tool, whether search was used, and whether personal context was present. Where you cannot establish a setting, record it as unknown.
For each answer, ask what it recommends, which claims it makes, and which linked pages support those claims. Follow the links. A source can explain the category without supporting the recommendation of a particular provider.
Keep branded questions separate. “What does this company do?” checks whether the business is represented accurately. An unbranded shortlist checks whether it is suggested for a particular job. A good branded answer does not establish visibility in supplier discovery.
Record each separately. A source link can support a fact without recommending the company.
A real report shows why this distinction matters
A saved report for a New York–based Salesforce consultancy, dated 5 September 2026, records ChatGPT responses collected from India. Its five broad category questions did not name the consultancy. Several category answers placed software platforms first.
That gives us an observation to investigate. It does not establish how often New York buyers search, or that a missing service page caused the absence. The report describes each question as asked once, and the original prompt and answer records were not available for our review.
The first useful action is to check whether the questions sought consulting firms consistently. Only then should we compare the business's service evidence with the sources in the answer. The business location and the test market are separate facts.
| What we have | What to inspect next | A useful output |
|---|---|---|
| A report showing absence in broad CRM categories | Whether the question asks for software or a consultancy | A clearer project-buyer question |
| A current service description | Whether it explains the required work and its limits | An evidence-backed page brief |
| A source link in an answer | Whether it supports the specific claim | A source-to-claim note |
Check access before rewriting everything
Ask your web team to inspect the important public pages. Can they be reached through normal links? Do they return the expected page? Are they accidentally excluded from indexing? Is the main information readable text rather than an image or an interaction that hides it?
For ChatGPT search, OpenAI identifies OAI-SearchBot as the search crawler and GPTBot as a separate training control. Allowing search access is a discovery consideration, not a promise of citation. OpenAI crawler documentation
Keep a record of what was actually checked. A successful browser visit is not a verified crawler request, and a crawl log is not proof that a page was used in an answer.
Improve the page that should answer the buyer
A useful service page tells a prospective customer what you do, who the work suits, what you need from them and where the service ends. It connects those statements to inspectable evidence.
For a consultancy, that could be a permissioned project explaining the Salesforce cloud, migration constraints and delivery decisions. For an equipment supplier, it could be a model table supported by current manufacturer documentation and a clear way to confirm compatibility.
When a page is vague, make the missing facts available. When the facts are already there, investigate another explanation instead of adding a second page that says the same thing. The source-selection guide explains why a useful page still may not be selected.
A diagnosis connects these records. It does not establish that one page change will produce a citation.
Correct inaccurate details at their source
If an answer gives the wrong location, service or product detail, identify the cited source where one exists. Update incorrect information you control and request a correction from an external publisher when appropriate. Keep the request factual and retain the supporting record.
Do not assume every wrong statement has a visible source or can be changed immediately. Record the disputed claim and repeat the check after the source correction. The result you can demonstrate may be that the source is now accurate, even if a later AI answer has not caught up.
Turn the diagnosis into a small work plan
Use one row per issue: buyer question, observation, supporting evidence, proposed change, owner and verification. “We were not mentioned” belongs in the observation column. “We need a new comparison page” is a proposed response that needs justification.
Start with the issue that most affects a buyer's decision. Fix it, retain the page version and repeat the same question under documented conditions. Show the individual rounds rather than keeping only the most favourable answer.
A better answer after an edit is encouraging. It does not isolate the effect of the edit from other changes. Measure qualified enquiries separately from answer visibility.
Questions owners ask
Should I create a page for every question?
Usually, one substantial page can answer several related questions. Create another only when the buyer has a different task and you have useful material to support it.
Does a citation mean ChatGPT recommends us?
No. The link may support a fact or a general explanation. Read the sentence attached to it.
Where should I begin this week?
Choose one important buyer question, save the current answer and inspect the relevant page and cited evidence. That gives you a clearer next step than a broad visibility score.
Request an AI visibility audit if you want help with that investigation. If the evidence already points to implementation work, see how TCC handles it.