What is a prompt in AI visibility?
In AI visibility, a prompt is a question tracked across AI engines, written the way a buyer would ask it. "What's the best AI visibility tool for a small team?" is a prompt. Each one is asked on every engine, every time a tool samples, which is why prompts are what AI visibility tools count, price and report on.
What makes a good prompt for AI visibility tracking?
A good prompt is a full question a buyer would type before they know your name, tied to one buying decision. It is not a keyword, and it does not name you.
Shape matters more than wording. In our sample, questions asking for alternatives to a named competitor named the brand in 14.9% of answers, questions with a price or team-size constraint in 15.1%, and plain "what's the best" questions in 13.9%, while head-to-head "A vs B" questions named it in 2.9% (the post). Cover the shapes a buyer moves through, and keep head-to-head questions in their own group.
How is a prompt different from a search keyword?
A keyword is a few words typed into a search box; a prompt is a whole question, often with the buyer's situation in it, and the engine rewrites it before it searches.
That rewrite matters. In 2,502 ChatGPT answers we collected, ChatGPT added the current year to the searches behind 55% of them when the question had no year, and a brand that appeared in its very first search was named in 88.6% of answers, against 7.0% for a brand in no search (the post). The prompt is the input you control; the query fan-out is what the engine actually looks up.
What do prompts look like in a real example?
The public Tesla example (Oct 4, 2026) tracked 25 buyer prompts in six topics, such as Powerwall home battery and Supercharger network, plus 5 prompts that name Tesla, each asked once on four engines.
One buyer prompt read "What is the best home battery for whole-house backup power during outages?" ChatGPT, Perplexity and Claude named the Powerwall 3; Google AI Overviews did not name Tesla. Across all 25 buyer prompts, 53 of 100 answers named Tesla, and 19 of 20 answers to the prompts that named it did.
How many prompts should you track?
Start with five or six topics and up to three prompts each, written in the shapes buyers use, then add more once you have acted on the first set.
Reliability comes from answers, not from the prompt count alone: 25 prompts on four engines is 100 answers a run, and repeated runs narrow the interval on each figure. A hundred prompts nobody reads are worth less than twenty that someone acts on. See prompt coverage for building the set.
How does Proofsource choose prompts?
Proofsource reads your site and proposes buyer prompts grouped by topic, ranked ahead when a demand signal backs them, and you confirm, edit or replace every one before the first scan. Prompts that name you are kept in their own set. See how questions are chosen and the Prompts feature page.
Common questions about Prompt
Is a prompt the same as a query?
In AI visibility, a prompt is the question you track; a query is a search the engine runs while answering it. One prompt can turn into several queries the engine writes itself.
Should prompts include my brand name?
Not the ones you count for visibility. A prompt that names you will mention you almost every time. Track named prompts separately to check accuracy.
Why do AI visibility tools price by prompt?
Because each prompt is asked on every engine on every run, so the cost of a tool grows with prompts times engines times runs.
How often should prompts be rewritten?
Rarely. Keep the set fixed so readings compare over time, and start new prompts as their own series. Changing wording mid-quarter resets the history.
Related terms
- Buyer question
A buyer question is a question about your category that does not name your brand, written the way someone choosing a product would ask it.
- Prompt coverage
Prompt coverage is how much of your set of tracked buyer questions AI engines answer with your brand named, broken down by question, topic and engine.
- Prompt volume
Prompt volume is an estimate of how often people ask AI assistants a given question or about a given topic.
- Query fan-out
Query fan-out is when an AI engine splits one question into several related searches and builds its answer from all the results.
Worked example: Tesla, public AI answers, Oct 4, 2026: Proofsource's first scan of Tesla, 25 buyer questions and 5 brand questions, one answer per question on each of four engines. A small sample from one brand, shown as an example, not a benchmark.
See where AI leaves you out, and what to fix.
See which buyer questions name your brand, who gets named instead, and what to fix. 200 answers. 25 Prompts/Question. Top AI engines. 2 days. Free. No card.
Google AI OverviewsGoogle AI Mode
