Prompt tracking

How do you build a prompt tracking panel?#

Build a prompt tracking panel by collecting 25–50 questions your buyers really ask, written without your brand name, then running them on the same engines, on the same schedule and under the same conditions every time.

1. Source prompts from buyers, not keyword tools. Sales-call notes, support tickets, demo-form answers and the question-shaped queries in Search Console are the best raw material. The AI question generator can fill gaps.
2. Write two or three phrasings of your key questions. People phrase the same need very differently: in SparkToro’s 2026 study, prompts volunteers wrote for the same intent had a semantic similarity of just 0.081. Several phrasings stop you tracking one lucky wording.
3. Pick engines by where your buyers ask. ChatGPT, Google AI Overviews and AI Mode, Perplexity, Gemini and Claude search different indexes, so track each one separately rather than as a blended score.
4. Freeze the conditions. Use a clean session with memory and personalization off, and keep the location constant. OpenAI says ChatGPT infers location and may use memories when it rewrites a query.
5. Run every prompt more than once per cycle. Answers vary from run to run, and two or three runs per prompt smooth out the noise.

What should you record for each prompt?#

For every answer, record whether the engine searched the web, which brands it named, which pages it linked, whether your brand was named or cited, and whether what it said about you was accurate.

date,engine,prompt_id,run,searched,brands_named,your_brand,cited_urls,accurate
2026-09-21,chatgpt,cat-03,1,yes,"Loopcraft; Plannora; Taskwell",named+cited,"plannora.io/pricing; stackreview.co/best-pm",yes
2026-09-21,gemini,cat-03,1,yes,"Loopcraft; Taskwell",absent,"loopcraft.ai/features",-

• Searched or not. An answer written from training data won’t change because you published a page this month; one built from a live search can. Semrush put ChatGPT’s search rate at 34.5% of prompts in February 2026.
• Named vs cited. Keep mentions and links in separate columns. In Semrush’s June 2026 study, Gemini named brands in 83.7% of their appearances but linked them in only 21.4%. See AI citation.
• The cited URL. Which of your pages earned the citation shows what to build more of, and the rival’s cited page shows what to beat.
• Accuracy. A wrong price or a retired feature is an AI hallucination worth fixing at the page the engine is reading.

How often should you run prompt tracking?#

Run prompt tracking weekly for the engines that matter most and monthly for the rest, with each prompt run more than once per cycle — AI answers vary so much between runs that one check a week can’t separate a trend from chance.

SparkToro and Gumshoe found less than a 1-in-100 chance that ChatGPT or Google’s AI returns the same list of brands twice for one prompt, and Attrifast found about half of the domains Claude cites change between runs. What stays stable is frequency — how often a brand appears across many runs — so repetition is part of the method, not an extra.

Prompt tracking vs rank tracking: what's the difference?#

Rank tracking records where your page sits for a keyword on a results page; prompt tracking records whether an AI answer names or cites your brand for a question — so it measures presence across repeated runs instead of one stable position.

The two overlap less than you’d expect. Ahrefs found only about 8% of ChatGPT’s citations rank in Google’s or Bing’s top 10 for the original prompt, because the engine searched narrower sub-queries than the one the user typed. A page can hold position one and still miss the answer.

Google’s own caveat applies to every tracker: “No third-party tool has access to our internal ranking or AI systems.” Prompt tracking samples what users see. It can’t read the engine’s scoring, which is why the sample has to be large and consistent.

Common mistakes with prompt tracking#

The most common prompt tracking mistakes are putting your brand name in the prompts, treating one run as a result, checking from a personalized account and counting only links — each one makes the numbers look better or worse than buyers actually experience.

Related terms#

• MeasurementAI share of voiceThe percentage of AI-generated answers, across a fixed set of relevant prompts, that mention or cite a brand compared with its competitors — the AI-search version of the classic measure of a brand’s share of the market conversation.Read the entry
• AI search & GEOAI visibilityA measure of how often, how prominently and how favorably a brand appears in answers from AI systems such as ChatGPT, Perplexity, Gemini and Google AI Overviews for the questions its buyers ask — the AI-search counterpart to search rankings.Read the entry
• AI search & GEOAI citationA link or source reference that an AI answer engine attaches to its response to show where a claim came from, and it differs from a brand mention in that it points to a specific page that can be clicked and can send traffic.Read the entry
• MeasurementAI referral trafficThe visits a website receives from links inside AI assistants and answer engines such as ChatGPT, Perplexity, Gemini, Claude and Copilot, identified in analytics by referrer domains like chatgpt.com and perplexity.ai.Read the entry
• How LLMs answerQuery fan-outAn AI search technique in which the engine rewrites one user question into several narrower sub-queries, runs them in parallel and builds its answer from the combined results — which is why a page can be cited for a prompt it doesn’t rank for.Read the entry
• GEOAI search & GEOGenerative engine optimizationThe practice of making content easy for AI answer engines such as ChatGPT, Perplexity and Google AI Overviews to retrieve, quote and cite — where SEO competes for a ranked link, GEO competes to be a named source inside the generated answer.Read the entry