How to track your brand in ChatGPT

To track whether ChatGPT recommends your brand, send the questions your buyers ask to the ChatGPT model through the OpenAI Responses API with the web_search tool enabled, then check each answer and its url_citation annotations for your brand names and domains — every day, on a prompt set you hold constant. ChatGPT has the largest reach of any AI assistant, which makes it the single most important answer surface to measure.

Last updated 2026-07-28

How do you query ChatGPT for measurement?

Through the Responses API with the built-in web_search tool switched on. That combination is what produces a grounded answer plus url_citation annotations — a structured list of the pages the model actually retrieved, rather than URLs written into the prose.

The distinction matters more than it sounds. An OpenAI completion without web_search answers from training data and will happily produce citations for pages that do not exist, which corrupts a visibility score invisibly. AI Visibility Tracker runs ChatGPT through the Responses API with web search on, reads citations only from the annotations, and reads the answer text separately for brand and competitor names. Why that separation matters is covered in why web-grounded measurement is non-negotiable.

Which model should represent "ChatGPT"?

The flagship consumer tier — the model a normal ChatGPT user actually gets. Measuring with a cheaper mini model measures a product almost nobody uses, and the resulting number is not comparable to anything.

This is a real temptation because grounded flagship calls are the most expensive part of a run. AI Visibility Tracker resolves it by fixing the measurement model to a consumer-representative choice with no free-text override, and by controlling cost elsewhere instead: prompt tiers so only core prompts run daily, same-day deduplication, a hard answer-token ceiling, and a per-run budget cap that skips a run whose projected cost exceeds it.

What does ChatGPT actually cite?

For discovery questions it leans heavily on third-party comparison articles, review directories and recent roundups; for brand-specific questions it reads your own site and major coverage. Because each answer synthesizes only a handful of retrieved pages, presence in those specific sources — not general ranking strength — decides whether you appear.

The practical consequence is that your own site is the wrong lever for "best X for Y" prompts. A vendor asserting it is the best is exactly the claim synthesis discounts; a third-party comparison naming you is the claim it repeats. How to get cited by ChatGPT works through the moves that follow from this.

What is ChatGPT-specific about reading the results?

• Highest reach of any assistant — if you weight engines by audience rather than treating them equally, ChatGPT carries the most weight in your overall score.
• Grounded and ungrounded answers differ sharply, and the difference is not random: ungrounded answers over-favour brands with a large training-data footprint. Always measure with search on.
• Answers vary between runs even with identical inputs. Daily sampling on a fixed prompt set is the only honest way to read it; a single answer is an anecdote.
• Real users' answers are personalised by memory, custom instructions and location. Your API measurement is deliberately the neutral baseline — it will not match any individual user's experience, and should not try to.
• A brand can be recommended by name with no link attached. Checking only the citation list undercounts ChatGPT visibility noticeably.

How much does daily ChatGPT tracking cost?

It depends on prompt count and model tier, but for a core set of ten to fifteen prompts it is typically a few cents per day in API usage — you pay the provider directly, at cost. Grounded calls cost more than bare completions because they carry search fees and read retrieved content.

AI Visibility Tracker shows a projected cost before every run and enforces a per-run budget cap (default $5) that skips a run projected to exceed it, so spend cannot run away while you sleep. The app itself is $29 once, no subscription, with a 7-day free trial; there is no per-prompt or per-seat meter on top.