ChatGPT visibility work helps you understand whether buyers receive accurate information about your offer during a defined research task. TANTU AI reviews the relevant website evidence, records answers under agreed conditions and identifies corrections or content gaps your team can act on. The service distinguishes an unprompted brand appearance from an answer that mentions the brand only after the user introduces it. It also separates factual accuracy, linked sources and recommendation context, giving you a more useful basis for action than a single screenshot or an assumed ranking.
Start with the missing buyer decision
A brand can be absent because the question is outside its market, its offer is poorly explained, its pages are inaccessible or other sources answer the question more directly. Repeating a branded prompt does not reveal which problem matters. Start with the products you can actually deliver and questions a qualified prospect would ask before contacting you.
For a B2B supplier, useful examples include category selection, compatibility with an existing system and implementation constraints. These are research prompts, not evidence that the platform recommends a supplier. We agree a balanced set of branded and unbranded questions before collecting the baseline.
Check search access and authoritative facts
OpenAI identifies OAI-SearchBot as its search crawler and separates its controls from GPTBot’s training-related controls. We inspect the owner’s crawler policy and hosting access against OpenAI: Overview of crawlers. Allowing access is a discovery prerequisite, not a promise of inclusion.
The website review then checks whether your legal identity, offer, availability, specifications and contact details are consistent. A comparison article cannot compensate for an unclear product page. We prioritize the source page closest to the buyer’s unresolved question.
What the work contains
| Workstream | What you receive | How we review it |
|---|---|---|
| Question design | A market-specific set organized by buying stage | Each question has an audience, intent and owner. |
| Observation | Dated answers, search context and source URLs where available | Missing responses are recorded as unavailable. |
| Content improvements | Approved page briefs, corrections and decision content | Claims trace to product evidence or an approved source. |
| Review | A comparison against the original sample and change log | Different modes and markets remain separate. |
For the initial diagnostic, ChatGPT can be one of the three agreed platforms. The $990 diagnostic covers 15 questions in one market and language, with two observations per question per platform. Additional implementation is scoped separately.
Turn findings into changes a buyer can use
- Missing category fit: explain who the product serves and when another solution is more suitable.
- Incorrect brand facts: reconcile the first-party source, identify outdated references and record a correction request where appropriate.
- Competitor cited: examine the underlying source and buyer need before deciding whether a new page is justified.
- Useful citation but no inquiry: review the destination page, offer clarity and next action.
We retain the baseline so later reviews can distinguish changed content from changed observation conditions. We do not submit confidential customer data to public research interfaces or imply that publishing a page immediately changes an answer.
Agree the scope before production
Share your website, a priority offering, one target market and the questions your sales team hears. We confirm platform access, evidence ownership, review responsibilities and the reporting sample. A managed program can then connect website corrections, new content, source research and monthly decisions over 365 days.
Your team remains responsible for approving product facts and receiving inquiries. The program does not include guaranteed placement, paid platform endorsement or a way to purchase a fixed recommendation.
Distinguish discovery from prompted recognition
A category question asks whether the business appears before the buyer supplies its name. A branded question tests what the answer says once that name is provided. A follow-up question can test a particular concern, such as implementation fit. All can be useful, but they answer different research questions and should be labeled separately in the agreed sample.
For an illustrative software vendor, “Which tools support this workflow?” examines discovery. “Does Brand X support this workflow?” examines representation after recognition. If the brand appears only in the second answer, report that distinction. Do not merge the responses into a stronger discovery claim. Follow-up exploration also consumes research time: agree whether it sits within the question allowance or requires a separate scope. Preserve the original question and conversation conditions so later observations can reproduce the intended buyer task.
Check whether the answer identifies the right business
Shared names, abbreviations and older product names can make a mention ambiguous. Build a short identity reference from the approved company name, owned domain, product names and category. Review whether the answer refers to that entity and whether any source link supports the identification. An exact text match alone may not be enough.
If an illustrative answer attributes another company’s product to your brand, first inspect your own naming and product descriptions. Make the relationship between company, product and domain explicit where it is unclear, then document any conflicting public source. The acceptance check for this work is a consistent, verifiable identity across the pages in scope. A later answer observation assesses whether the ambiguity persists, without promising a correction in the platform. Where identification remains uncertain, flag the observation for review instead of automatically counting it as a meaningful mention.
Distinguish search crawling from a user-requested visit
OpenAI documents separate roles for OAI-SearchBot, GPTBot and ChatGPT-User. Search crawling and training controls are independent. A user-triggered visit is a different mechanism and does not determine search inclusion.
An access review should examine the relevant policy and real hosting behavior, using the official agent and IP documentation where applicable. A successful fetch demonstrates access under those conditions; it does not establish that a page will be selected for a future answer.
Common questions
Is “ChatGPT SEO” a separate ranking system we can control?
It is a market term for improving discovery in ChatGPT. We scope concrete website, content and measurement work; the platform decides which sources and answers to show.
Why can two tests give different answers?
Prompts, search mode, account context, market, time and follow-up conversation can differ. We record the conditions and repeat the agreed sample rather than treating one answer as a stable rank.
Sources & method
Platform facts are linked below. The planning steps, examples and acceptance criteria are TANTU AI’s operating approach. Examples are illustrative unless identified otherwise.
Reviewed 16 September 2026. Product availability and controls can change.