What is a prompt set?

A prompt set (also called a prompt sample or tracking prompts) is the list of questions a tracker asks AI engines on your behalf. Each entry has three parts:

• The question, for example "best accounting software for a small agency".
• The engine it is sent to, such as ChatGPT, Gemini, Perplexity or Claude.
• The settings: whether web search is on, the location and language, and whether the session is logged out.

The set is a sample of what buyers might ask. It is not a log of what they actually typed. For the step-by-step method, see how to build a prompt set.

The tracker sends each prompt to each engine on a schedule. It records the answer, then checks for brand mentions, cited URLs and stated facts. Metrics are ratios over the answers that came back.

The prompt is not what the engine searches for. As of October 2026, OpenAI says ChatGPT search rewrites a query into one or more targeted queries before sending them to search providers. Google's Gemini API generates and runs its own search queries. Google also says AI Overviews and AI Mode may use different models and techniques. So your prompt goes into a process you cannot observe.

Retrieval is also optional. An engine may answer from what the model already knows, with no live search at all. A fair report says which answers were grounded and which were not.

Three choices decide what the numbers mean.

Size

With a handful of prompts, one answer can swing the score. More prompts reduce noise, but a well-built score should not fall just because the set grew. If yours does, the scoring formula is counting errors without dividing by answers. See whether more prompts should change your score and how many prompts you need.

Wording

A prompt that names your brand tests what the engine knows about you. An unbranded prompt tests whether you get recommended at all. Leading phrasing ("why is X the best...") inflates results.

Brand choice

Large brands dominate AI answers across engines. Comparing a mid-market company against category giants mostly measures brand size. Compare against competitors of similar scale.

What makes a set fair

• Covers the real buyer journey: discovery, comparison, pricing, problems.
• Uses neutral, unbranded wording for recommendation questions.
• Stays fixed between runs, so trends reflect the engines, not your edits.
• Runs logged out with memory off, since ChatGPT can use saved memories when rewriting a query.
• Reports each engine separately instead of blending them.

• "Our prompts are what customers type." They are a guess. The queries engines actually run stay hidden, except in Bing Webmaster Tools, which shows grounding queries for Copilot and Bing. Comparing grounding queries with your prompt set shows how far the two drift.
• "The tracker sees every citation." It sees answers to its own prompts. In Otterly's 2026 experiment, Microsoft's own report credited one site with 42,817 citations while Otterly's prompt-based tracker logged 3,934. The time windows may not match, but the gap shows a sample is a sample. More in why Bing's counts differ from your tracker.
• "A prompt set gives you a rank." No engine publishes position data for AI answers, so there is nothing to check a rank against. Report presence, citations and accuracy instead.

The free checker on this site runs up to three buyer questions. That makes it a snapshot, not a tracked prompt set.

• Prompt set template: categories and examples
• Writing your own prompts vs letting a tool generate them
• Are AI prompt-tracking tools accurate?
• Grounding query: the search an engine actually runs after rewriting a prompt.

How is a prompt set different from a keyword list? A keyword list targets search terms that a search engine matches against pages. A prompt set contains full questions sent to AI engines, which then rewrite them into their own searches or answer from memory. Keywords tell you what to rank for. Prompts tell you what an AI answer says about you for a specific question, under specific settings.

Should a prompt set change over time? Keep a stable core so trends compare like with like. If you edit prompts every month, a rising or falling score may reflect your edits rather than the engines. Add new prompts as a separate group when your products or market change, and report them apart from the core until they have enough history to compare.

Do I need separate prompt sets for different countries? Usually yes. Answers vary by location and language, and buyers in each market phrase questions differently. Translating one set word for word often produces questions nobody asks. Write each market's prompts in its own language, run them with matching location settings, and report each market on its own rather than averaging them together.

Can a prompt set tell me why I am missing from AI answers? Not by itself. It shows where you are absent, not the cause. The first things to check sit outside the set: whether your robots.txt lets search crawlers such as OAI-SearchBot and PerplexityBot in, whether your firewall returns errors to them, and whether key facts appear in the HTML before JavaScript runs.