To see your share of voice in ChatGPT, run a panel of category prompts, count every brand mention across all the responses, and divide your mentions by the total. ChatGPT has no analytics dashboard and OpenAI publishes no prompt logs, so this sampled number is not a proxy for the real figure; it is the only figure anyone has, including your competitors.
The concept itself, and why share of voice beats raw mention counts, is covered in what share of voice in AI answers means. This page is the how.
Design the prompt panel
Share of voice is only as meaningful as the prompts behind it. Build a panel of at least 20 prompts spread across the four buyer families: recommendations ("best CRM for a small agency"), comparisons ("HubSpot vs Pipedrive"), alternatives ("alternatives to Salesforce"), and fit or pricing questions ("affordable CRM with good email automation"). Use the vocabulary your buyers use, not your internal category jargon, and include two or three phrasings of the highest-stakes questions, because ChatGPT can answer a reworded prompt with a different brand list.
Counting rules that keep the number honest
- Count each brand at most once per response, however many times it is repeated.
- Record position too: first-named and last-named are both one mention, but they are not the same outcome.
- Run each prompt 3 to 5 times in fresh sessions; ChatGPT's answers vary between runs, and single runs produce coin-flip data.
- Keep memory out of it: use sessions without personal history, since ChatGPT's memory can tilt recommendations.
- Note whether search mode was active, and keep it consistent across measurements; with and without retrieval are different surfaces.
- Pin the model version where you can, and record it where you cannot.
What a defensible sample looks like
The test for sample size is stability: if rerunning the whole panel tomorrow would swing your share of voice by ten points, the sample is too small to act on. In practice, 20 prompts at 3 to 5 runs each, meaning 60 to 100 responses, is a workable floor for a directional read on one provider, and bigger panels buy tighter error bars. Report the number honestly as what it is, a sampled estimate with variance, and resist decorating it with false precision. Trends across weeks matter more than the third decimal place.
Reading the result
Benchmark against competitors, not against 100 percent. In a category ChatGPT answers with five-brand shortlists, nobody gets 60 percent share of voice; the question is whether you outscore the rivals you actually lose deals to, and whether the gap is closing. Split the number by prompt family before drawing conclusions, because a healthy overall share often conceals a zero on alternatives prompts. A free scan computes competitive share across ChatGPT and four other providers in one pass, with the same prompts on the same day, which is what makes the comparison fair.
Get your ChatGPT share of voice measured properly
The free scan runs organic buyer prompts across five AI engines and returns your competitive share, so you can see exactly how much of the ChatGPT conversation your brand owns.
Frequently asked questions
What is a good share of voice in ChatGPT?
There is no universal benchmark, because it depends on how many brands ChatGPT typically names in your category. The practical standard: outscore the competitors you actually lose deals to, and trend upward quarter over quarter. A brand named in most shortlists alongside four rivals is doing well at roughly 20 percent.
Does ChatGPT's memory affect share of voice measurement?
Yes. Memory and chat history can personalize which brands ChatGPT recommends to a given account, which contaminates measurement. Sample from sessions without personal history so you are measuring what a generic buyer would see, not what ChatGPT has learned about you.
How often should I re-measure share of voice?
Weekly while you are actively publishing content or fixing entity signals, monthly once things are stable. Use the same prompt panel and the same counting rules every time; the trend across identical measurements is the insight, and any change to the method breaks comparability.
Related answers