ChatGPT visibility
ChatGPT visibility: TrustInsider asks OpenAI’s models the questions in your panel through the official API, with and without web search, and for each answer keeps mentions, recommendations, cited pages, country, city and model.
Country and city
With search on, every question carries the country of the panel’s market and, in a city panel, the city as well. The data is always sent: without it, OpenAI’s engine assumes a search from the United States.
The location is approximate and declared: it is not an address or a person, it is the market you chose for those questions. Without search no location is sent, because the model does not consult pages.
The three ways a panel reaches ChatGPT’s web search: by country, by city, and without search, with the location data each one sends.
| Panel | Location sent | What changes in the reading |
|---|---|---|
| By country | The market’s country, approximate. | The pages found are the ones the search offers for that country. |
| By city | The panel’s country and city. | Each city is a cell of its own: two cities are not added together. |
| Without search | None: the model does not consult pages. | Read how the model knows the brand, wherever you are. |
Cited pages
With search, ChatGPT ties links to precise points in the text. TrustInsider keeps each link next to the sentence that carries it, so you see which statement rests on your site and which on a third-party page.
Only the links that the answer declares as citations of the text are counted. A site named without a link is a mention, not a source: the two stay in different columns.
A sample ChatGPT answer with search on, about fictional brands: the sentence that recommends the product, the citation of the brand’s site, the competitor without a link and the electronics shop cited as a source.
QuestionWhich bookshelf speakers should I buy for a small living room?
ChatGPT with search · sample answer about fictional brands
In a small space the Ilmarde Audio Sulvene S work well, compact and with sound adjustable from the app, as explained on the product page on ilmarde-audio.example. Varolde Audio also has similar models; up-to-date prices are on an electronics shop.
The sentence recommends the brand’s product for that space: it counts as a recommendation.
The link to the brand’s site is tied to this sentence: the site is a cited source.
The competitor is named without a link: it is a mention, not a source.
A third-party page cited for prices: it enters the external sources with its domain.
Data to OpenAI
Only the text of the question leaves, with the chosen model and, with search, the market’s location. No description of the brand is added to the question: it would distort the very visibility being measured.
Every request asks OpenAI not to save the answer among those that can be retrieved later (store: false); OpenAI may retain the data for a limited period for security purposes, under its terms. The model’s reasoning is not requested: if a model reasons, the receipt reports only its number of tokens, never the text.
The information in a measurement question, split between what reaches OpenAI’s API and what stays only in the customer’s collection.
| Sent to OpenAI | Kept only in the collection | |
|---|---|---|
| The question | The exact text, with nothing added about the brand. | The text, the panel and the intent it belongs to. |
| The market | Country and city, only if search is on. | The panel’s market for each cell. |
| Storage | The answer is not saved among those that can be retrieved. | The answer, the cited pages and the outcome. |
| The reasoning | The reasoning summary is not requested. | Only the count of reasoning tokens. |
Cost per answer
Every answer arrives with its bill: input and output tokens, reasoning tokens, searches run and the service tier with which OpenAI served it. The cost in credits shown next to the command sets the cap; the receipts say how much was actually spent.
The sample receipt of a ChatGPT answer with search, about a fictional brand: model, service tier, tokens, reasoning, searches and outcome.
Frequently asked questions
The differences between the API and the app people use, between a refused answer and an absent brand, and between a measure by country and one by city.
Not always. The app uses the memory and settings of the person writing, the API does not: TrustInsider measures OpenAI’s official API and writes so next to every measure.
No: a refusal is an outcome of its own. It stays in the log with its label and does not enter the count of appearances, neither for your brand nor for competitors.
With search on, ChatGPT receives the panel’s city, and the pages it finds can change from one city to another. Each city stays a cell of its own, and comparison holds between periods of the same city.
The free check asks the installed assistant a single recognition question. ChatGPT needs the questions in your panel: repeated, with and without search, and read by country or by city.
Related paths
ChatGPT’s answers change with web search and with location. To compare them with the other assistants’ answers, open AI Visibility, then explore the platform’s steps, the sampling methodology and support.
See when ChatGPT, Gemini and the other assistants name, recommend or cite your brand: the answers, question by question.
ReadLearn how to use TrustInsider: choose the brand and the questions, read the collected answers, update your content and compare the results.
ReadHow every TrustInsider measure is built: sampling of AI answers, sources and rights, index calculation and declared limits.
ReadHelp with access, credits, collections, integrations and data: guides, FAQs, the site assistant and support requests from your workspace.
Read