Prompt tracking is the practice of running a fixed set of questions (prompts) through AI answer engines on a schedule and recording what comes back: which brands each answer names, which pages it cites, and how both change over time. It fills a gap, because the engines' own reports show only part of that picture.
How prompt tracking works
- Pick prompts that match what buyers ask. Two in our own tracked set are "best Profound alternative" and "AI visibility audit tool"; how to choose the prompts to track covers the method.
- Pick the engines to cover, since each one retrieves and cites differently.
- Run every prompt on a schedule, such as daily, and store each full answer with its date.
- Parse each answer for mentions (the brand named in the text) and citations (a page on the brand's domain that the answer lists or links as a source). Our methodology keeps the two apart, because an answer can mention a brand without citing it.
- Report rates across many runs: the share of answers that name you, the share that cite your pages, and the same for rivals. That is the basis of AI share of voice.
Why prompt tracking matters for AI answers as well as classic search
Classic search has Search Console, whose performance report shows the queries, clicks and impressions behind your traffic. For AI answers the first-party data is thinner. Google's generative AI report, rolled out to all sites by 31 August 2026, counts impressions of your links in AI Overviews and AI Mode by page, country, date and device, with no query dimension. Bing Webmaster Tools' AI Performance report, introduced as a public preview in February 2026, counts citations of your pages in Copilot, Bing's AI summaries and select partner integrations, and shows a sample of the grounding queries behind them.
Both reports cover only your own site, so neither shows the answers that left you out or the rivals they named instead. Prompt tracking samples the answers themselves, including in engines neither report names, such as ChatGPT, Perplexity, Claude and Grok.
An example: seven engines, seven source lists
| Engine | Answers recorded | Third-party citations | Most-cited site (citations) |
|---|---|---|---|
| ChatGPT | 552 | 2,598 | otterly.ai (123) |
| Perplexity | 552 | 11,131 | llmpulse.ai and reddit.com (300 each) |
| Gemini | 552 | 1,910 | llmpulse.ai (111) |
| Claude | 552 | 4,377 | aiclicks.io (171) |
| Grok | 552 | 2,481 | rankability.com (75) |
| Google AI Overviews | 523 | 4,117 | youtube.com (371) |
| Google AI Mode | 248 | 1,814 | youtube.com (159) |
The same tracked questions drew about 20 third-party citations per answer from Perplexity and about 3.5 from Gemini, and no single site led more than two engines. A reading from one engine says little about the next; what seven AI engines cite goes deeper.
Common misconceptions
- "One check shows where you rank." SparkToro and Gumshoe.ai, an AI tracking startup, had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. They put the chance of ChatGPT or Google's AI giving the same list of brands in any two of 100 responses at under 1 in 100. Their conclusion: list position is close to random, but the share of runs in which a brand appears is probably a reasonable measure. City of Hope appeared in 69 of 71 ChatGPT answers about West Coast cancer care, yet was named first in only 25.
- "Tracked prompts are what customers type." They are a sample. In the same study, 142 survey respondents wrote their own prompts for one need, headphones for a traveling family member, and barely two looked alike; the answers still drew on a consistent set of brands.
- "A tracker sees inside the engine." It sees outputs only. Google says no third-party tool has access to its internal ranking or AI systems.