How to Measure Your Brand's Visibility in ChatGPT
To measure your brand’s visibility in ChatGPT, run the questions your buyers actually ask, log four fields every time, and repeat on a fixed cadence. Visibility is not one number. It is a pattern you build by testing the same prompts over time under controlled conditions.
This is also the cheapest useful thing you can do in AI SEO. It costs an afternoon to set up and it turns every later decision into an evidence-based one.
Why you have to measure this deliberately
Three properties make casual checking useless.
Answers vary between runs. The same prompt can name different brands minutes apart. A single answer is an anecdote.
Most AI answers are zero-click, so your analytics will not show what happened. Semrush measured roughly a 93% zero-click rate for Google’s AI Mode on informational queries. If you wait for traffic to tell you, you will conclude nothing is happening while your competitors get named.
Two different mechanisms produce answers. With web search active, ChatGPT retrieves live pages. Without it, it answers from trained memory. Those respond to completely different work and on completely different timescales, as covered in LLM SEO.
Step 1: Design the prompt set
Build 30 to 50 prompts across four intents. Write them the way a person talks, not as keywords.
| Intent | Example shape | What it tells you |
|---|---|---|
| Category | “What tools do X for Y?” | Are you in the consideration set at all |
| Comparison | “A vs B for Z” | How you fare against a named rival |
| Problem | “How do I fix X?” | Whether your content earns the citation |
| Recommendation | “Who should I hire for X in Y?” | The commercial question that matters most |
| Brand | “What is [your brand]?” | Accuracy, not presence |
That last row is the one most people omit and the one that most often finds a real problem. An inaccurate mention can cost more than no mention.
Freeze the set. Changing prompts between cycles destroys the trend. Add new prompts as a separate tranche and track them from their own baseline.
Step 2: Control the run
- Run each prompt three to five times and treat the runs as a sample.
- Start a fresh conversation each time, so prior turns do not contaminate the answer.
- Record whether web search was active, and ideally run both ways.
- Note the model version and the date.
- Turn off personalization or memory features where you can, or note that they were on.
Consistency matters more than sophistication. The same method, run the same way, beats a better method run inconsistently.
Step 3: Log the four fields
For every run, record:
- Named. Did your brand appear at all?
- Cited. Was a link attached, and was it to your site or a third party talking about you?
- Accurate. Was the description correct? Note any specific error.
- Competitors. Which other brands appeared.
Write your counting rule down before you start. Does an unlinked mention count? Does appearing in a list of eight count the same as being the recommendation? There is no right answer, only a consistent one.
A workable sheet has one row per run: date, engine, model version, prompt, search on/off, named, cited, source cited, accurate, competitors, notes.
Step 4: Turn logs into a cadence
- Monthly: the full prompt set, all engines, several runs each.
- Weekly: a five-prompt subset covering your highest-value questions.
- Event-driven: re-run after a major model release or a significant content launch.
Calculate presence rate and competitive share from the logs, and track direction rather than decimals. The definitions are in share of model, and the wider metric set is in the AI visibility metrics that actually matter.
What the results tell you to do
| Pattern | Likely cause | First move |
|---|---|---|
| Absent everywhere, search on | Retrieval failure | Crawler access, Bing indexation, rendering |
| Absent with search off, present with it on | No presence in trained memory | Long-term corroboration work |
| Present but never cited | Passage not quotable | Restructure to answer-first |
| Competitors cited for your category | Off-site gap | Reviews, roundups, directories |
| Named but described wrongly | Entity problem | When AI gets your brand wrong |
| Strong on category, weak on recommendation | Trust gap | Entity plus corroboration |
That second row is the distinction most measurement setups miss entirely, and it changes the remedy completely.
Do this for the other engines too
ChatGPT is the largest AI referral source, at 78.16% of AI-chatbot referrals per StatCounter in March 2026, but it is not the only surface your buyers use. Run the same set against Perplexity, Gemini, and Google’s AI Mode, and keep the results separate: they retrieve from different indexes and select differently. See tracking your brand in Perplexity, Gemini and AI Overviews.
Keep ChatGPT measurement honest
Four things keep these numbers meaningful: a fixed prompt set, multiple runs per prompt, consistent fields logged every time, and a steady cadence. Collection can be manual or scripted, and the tradeoffs are compared in AI citation tracking tools.
Read trends, not readings. This space is new and inherently noisy, so a three-cycle direction beats any single impressive number. Do not invent precision you do not have: a well-kept directional metric is far more useful than a fabricated exact one, and reporting decimals from a sample of thirty answers implies certainty the method cannot support.
Where ChatGPT measurement leads
The log is not the deliverable. The pattern in it is: it tells you whether your problem is retrieval, structure, corroboration, or accuracy, and those have different fixes and different timescales.
Want this done rigorously and turned into a prioritized plan? Request an AI visibility audit and we will measure how ChatGPT and the other engines describe your brand today, and show you which gap to close first.
Frequently asked questions
How do I check if ChatGPT mentions my brand?
Ask ChatGPT the questions your buyers actually ask, then record whether your brand is named, how it is described, and whether a source is cited. Repeat the same prompts on a fixed schedule so you can see change over time rather than a single snapshot.
How many prompts should I track?
Thirty to fifty for most brands, covering category, comparison, problem and recommendation intents. Fewer than twenty is too small a sample to show a trend; more than fifty becomes hard to sustain manually every month.
How often should I measure ChatGPT visibility?
Monthly for the full set, with weekly checks on your highest-value prompts. Re-measure after major model updates, since answers can shift noticeably when a new version ships.
Does ChatGPT give everyone the same answer?
No. Answers vary by phrasing, prior conversation, model version, whether web search is active, and personalization. Run each prompt several times and record ranges rather than single results.
Should I run prompts with web search on or off?
Both, and log which. With search on you are measuring live retrieval, which you can influence in weeks. With search off you are measuring the model's trained memory, which only changes when a newer model ships. Confusing the two leads to the wrong fix.
What should I record for each prompt?
Whether your brand appeared, whether a source was cited and whose, whether the description was accurate, which competitors were named, and the date and model version. A spreadsheet is enough to start.