What does an AI rank tracker actually measure?
A sample. The vendor writes a list of questions, often called synthetic prompts (made up by the tool, not typed by your customers). It sends them to ChatGPT, Gemini, Google's The AI-written summary Google shows above some search results, with links to the pages it drew from. More and others on a schedule, then reads each answer for your name, your links and your rivals. Every number on the dashboard comes from those runs.
| Number on the dashboard | What it really is | Trust it? |
|---|---|---|
| Visibility or mention rate | Share of the tool's answers that named you | Yes, as a trend, if each question runs many times |
| Share of voice | Your mention rate next to your rivals' | Yes, on the same condition |
| Citations | Your pages shown as sources in those answers | Yes, for the answers collected |
| Average position or "AI rank" | Where you sat in a list that reshuffles almost every run | No |
| Prompt or search volume | The vendor's estimate; no AI company publishes it | Only if the vendor shows how it's made |
| Sentiment | The tool's own reading of the tone | Read a few answers yourself |
The mention rate is the useful part. Position and volume are the weak parts.
Why is there no stable rank in AI answers?
Because the same question gets a different answer almost every time. In SparkToro's January 2026 study, 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. The authors put the chance that ChatGPT or Google's AI repeats the same list of brands, in any two of 100 answers, at under 1 in 100. The same order: closer to 1 in 1,000. BrightLocal's September 2026 study of local searches agrees: result sets from repeat runs overlapped only 23% on ChatGPT, 21% on AI Mode and 33% on AI Overviews.
Three more things move the answer:
- The person asking. OpenAI's help center says ChatGPT can use your saved memories (if memory is on), and an approximate location from your IP address, when it searches the web. A tracker can't copy any customer's memories.
- The product. In BrightLocal's study, AI Mode and AI Overviews shared only 29% of the businesses they named, and ChatGPT 19–20% with either. Google itself says:
"AI Mode and AI Overviews may use different models and techniques, so the set of responses and links they show will vary." (Google Search Central: AI features and your website)
- The wording. SparkToro asked volunteers to write a prompt for one need: headphones for a family member who travels. Their 142 prompts were nearly unrelated, with an average semantic similarity of 0.081 (1.0 would be identical).
What held steady was who got named. Across 994 answers to those 142 prompts, Bose, Sony, Sennheiser and Apple each appeared in 55–77% of them. The authors conclude that visibility percentages across many prompts are likely "decent proxies" for how brands show up in real AI answers, while position in a list means little. The study discloses that one author works for an AI tracking startup. BrightLocal saw the same in local searches: a business named once came back in about half of later runs.
So count how often you're named across many runs. Ignore where you sit in any one answer.
Can a tracker see how many people ask ChatGPT a question?
No. No AI company shares how often people ask a given question, so any "prompt volume" on a dashboard is the vendor's estimate. Ask how it's made. What the platforms do share, free, about your own site:
| AI tool | What you get | What you don't get |
|---|---|---|
| Google AI Overviews and AI Mode | Search Console's Generative AI performance report: impressions by page, country, device and date. Launched June 2026; all sites by August 31, 2026 | The questions behind the impressions; the Gemini app |
| Microsoft Copilot and Bing's AI summaries | Bing Webmaster Tools' AI Performance report (public preview since February 2026): citations per page, plus "grounding queries". Since June 2026, also in preview: Citation Share, your percentage of all citations for a grounding query | Placement: Bing counts citations "without indicating placement or presentation within a specific answer" |
| ChatGPT | Your claimed business listing's insights (beta since November 2026): mentions and website clicks over the past 28 days | Individual prompts or conversations; local businesses can claim only US listings |
Bing's grounding queries are not your customers' exact words. Bing defines them as "the key phrases the AI used when retrieving content," and says the data "represents a sample of overall citation activity." Google's free tool that shows which searches bring people to your site and flags pages Google can't index. More setup: Search Console for a small business.
Two more real numbers sit in your own systems:
- Visits from AI answers. OpenAI's publisher FAQ says ChatGPT adds utm_source=chatgpt.com to links it sends, so those visits show by name in your analytics.
- Leads. Ask every new customer "How did you hear about us?" It's the simplest way to tie an AI answer to revenue.
For Claude, Perplexity and the Gemini app, a tracker or your own test is how you see the answers.
How big a change in your numbers is real?
Bigger than most dashboards suggest. A mention rate is a sample, and small samples swing by chance. Basic probability (a binomial calculation) shows how much:
| Answers checked | If you're truly named in 20% | Usual range by chance (95%) |
|---|---|---|
| 40 | 8 | 3–13 |
| 100 | 20 | 12–28 |
| 200 | 40 | 29–51 |
So a move from 8 to 11 out of 40 is noise. Believe a change when it holds for two or three months, or when the sample is large. SparkToro's author goes further: to know an AI's set of recommendations, he says, ask the same question at least 60–100 times.
Is an AI rank tracker worth paying for?
It depends on how many questions and AI tools you need covered.
- Probably not yet for one location with a few services. Google's and Bing's reports, your analytics, and a monthly 20-question test cover it free. In the US, add your ChatGPT listing's insights.
- Maybe for several locations or service lines, weekly reporting, or AI tools with no owner report (Claude, Perplexity, the Gemini app). That's where hand testing stops being practical.
Before you pay, ask the vendor:
- How many times does each question run per check? One run is a snapshot, not a rate.
- Can I add questions in my customers' own words?
- Can I read the raw answers and their sources?
- Where does the "volume" number come from?
- Which city and account settings do the runs use?
- Does the report lead with an average position? If so, treat that number as noise.
Deciding whether to pay for Generative engine optimization: work that helps AI assistants like ChatGPT, Gemini and Perplexity find, trust and mention your business. More work at all: is GEO worth paying for?.
Example
Illustrative cases. Numbers are invented for the example.
A Denver roofer whose "AI rank" dropped (US dollars)
His $100-a-month tracker shows his average position falling from 2 to 5. His mention rate across 120 answers held at 31%. Nothing changed but the order of the lists. He keeps watching the rate, plus Bing's citation count, which is free.
A Calgary dental office with no tracker (Canadian dollars)
It can't claim a ChatGPT listing yet. Search Console's AI report shows its implant page earning The number of times your ad or listing was shown, whether or not anyone clicked. in Google's AI answers each month, and analytics shows a few visits tagged chatgpt.com. With "How did you hear about us?" on the intake form, it knows what it needs for $0.
An HVAC company in six cities (US dollars)
Five services in six cities make 30 questions. Across four AI tools, run three times each, that's 360 checks a month. By hand, that's not realistic. A tracker earns its fee here; the company picks one that shows raw answers and accepts customers' own wording.
FAQ
Can I track my ranking in ChatGPT?
Not as a fixed position. Track how often ChatGPT names you across many runs. In the US, a claimed listing also shows mentions and clicks.
Do AI rank trackers use real customer questions?
Mostly their own. AI companies don't hand out customers' prompts; OpenAI's help center says businesses don't receive individual prompts or conversations.
Does Search Console show AI Mode data?
Yes. Its Generative AI report shows AI Overviews and AI Mode impressions by page, for every site since August 2026. It doesn't show the questions.
Are AI rank trackers accurate?
They record what the AI said. What they infer is weaker: a position in a reshuffling list, or a volume no AI company publishes.
Different for you?
- You haven't checked by hand yet: the 20-question test.
- You want Google's AI numbers: Search Console for a small business.
- Someone is selling you GEO with an AI rank report: is GEO worth paying for?.
- You want the rest of your monthly numbers: is my SEO working?.
- Your buyers use Perplexity, Gemini or Claude: showing up in those answers.
- You'd rather have it tracked for you: SEO & GEO Agent.