Someone on the leadership team asks ChatGPT for the best vendors in your category. Your brand isn't in the answer. A screenshot lands in Slack before lunch.
Two days later, someone else asks the same question. You're third on the list. Another screenshot. Now everyone's confused.
Both screenshots are accurate. Neither tells you much.
AI answers change every time you ask
SparkToro tested this directly. In research published in January 2026, 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI, 2,961 times in total. The prompts asked for brand and product recommendations, from cloud computing providers to Volvo dealers in Los Angeles.
There was less than a 1 in 100 chance of getting the same list of brands twice. The same list in the same order? Under 1 time in 1,000.
So "where do we rank in ChatGPT?" has no stable answer to report. What you can measure is how often you appear across many prompts, run many times. SparkToro's own conclusion was that visibility percentage across dozens to hundreds of prompts, each run multiple times, is a reasonable metric.
That's a frequency. Treat it like one.
Why it's worth measuring
The clicks are moving. Pew Research Center tracked how Google users browsed in 2025 and found that people who saw an AI summary clicked a traditional search result on 8% of visits. Without a summary, it was 15%. Links inside the AI summary itself got clicked on 1% of visits.
For B2B research, a buyer can now build a shortlist without visiting your site. If your brand isn't in the answer, that buyer may never appear in your analytics. Your SEO report can look healthy while you lose ground.
Build a prompt set your buyers would type
Start with 30 to 50 prompts. Write them the way a buyer talks, which is rarely how a keyword tool phrases things.
Split them by stage:
Problem prompts. "Why aren't our trial users converting?" Early research, no vendor in mind yet.
Category prompts. "Best [category] tools for mid-size companies." Shortlists form here.
Comparison prompts. "[Your brand] vs [competitor]." Late stage, and where a wrong fact hurts most.
Brand prompts. "What does [your brand] cost?" or "Is [your brand] good for [use case]?"
Ask sales for the questions prospects bring to first calls. Those make the best prompts you'll get.
Run each prompt more than once
One run per prompt just repeats the screenshot problem. Run each prompt at least 5 times per tool, in fresh sessions, logged out where you can. Cover the tools your buyers use. For most B2B teams that's ChatGPT, Google's AI Overviews and AI Mode, Perplexity, and Microsoft Copilot.
For every response, log four things in a spreadsheet:
Were you mentioned? Yes or no.
Which competitors were mentioned?
Was what it said about you correct? Pricing, features, who you serve.
Which sources did it cite?
Then work out your visibility rate per prompt group. Appeared in 14 of 50 category runs? That's 28%. Do the same for your top three competitors, and you have a share-of-voice comparison sitting next to your own number.
Read the citations, then go earn them
The fourth column is the useful one.
AI answers lean hard on third-party pages: review sites, comparison articles, industry publications, community threads. When the same few domains keep getting cited for your category prompts, that's your target list. Getting mentioned accurately on those pages tends to move visibility more than another rewrite of your own homepage.
Watch the accuracy column too. An answer that names you with last year's pricing, or describes a product you've retired, can cost a deal. Fix the source pages it's pulling from. Start with your own site, then the review profiles and directory listings you control.
Measure monthly, report the trend
Rerun the full set once a month. Same prompts, same number of runs, same tools. Chart visibility rate by prompt group over time, with your competitors on the same chart.
Expect noise. A swing of a few points from one month to the next is probably randomness. A steady climb across a quarter is a signal.
Be careful with paid tracking tools. Some are useful. SparkToro's advice was to make any provider show its math before you spend money, since a tool that runs each prompt once inherits the screenshot problem at scale.
Start with 10 prompts this week
You can start without a platform. Pick 10 category prompts, run each 5 times in ChatGPT and in Google's AI Mode, and log the results. That's 100 responses and about two hours of work.
At the end you'll know your rough visibility rate and which competitors keep showing up in your place. You'll also see which sites the AI tools trust in your category, and that list is where next quarter's work starts.
