Target prompt: the question your brand needs to be cited in
A target prompt is the natural-language question a person asks an AI assistant, and in which a brand wants to appear as a cited source. It is the unit of measurement for AI Visibility: what you track is not a position in a list of links, it is presence inside the answer.
What is a target prompt?
A target prompt is the natural-language question a person types into an AI assistant and in which a brand wants to be cited. It replaces the keyword as the unit of tracking: what you measure is presence inside the answer, not position in a list of links.
In practice a target prompt is a full sentence, written the way someone would actually type it: "which Brazilian asset manager has the best fixed income research?", "who runs AI visibility diagnostics for companies?", "is it worth publishing the monthly letter in HTML as well as PDF?". Each of those produces a single synthesized answer with a handful of cited sources. That short list is where the competition happens.
The target prompt is the first thing a diagnostic defines, before any decision about content. Without it there is no baseline, and without a baseline no one can claim a brand improved: you would need a prior record of the questions it was absent from. In an AI Visibility program run in-house or by an agency, the set of target prompts is what ties production, measurement and optimization to the same target.
What is the difference between a keyword and a target prompt?
A keyword is short, ambiguous and built to rank on a results page. A target prompt is long, specific and carries context: role, sector, constraint, intent. Keywords measure position; target prompts measure citation. Both coexist: the keyword sets the title, the prompt sets the H2.
| Layer | Keyword | Target prompt |
|---|---|---|
| Shape | 2 to 5 words | a full sentence, typically 8 to 25 words |
| Where it lives on the page | <title> and H1 | an H2 in question form, with the anchor answer right below |
| What it measures | position on the results page | citation inside the generated answer |
| Known volume | yes, from search tools | no, generative engines do not publish prompt volume |
| How it is tracked | rank tracker | running the prompt on each engine on a schedule and logging cited sources |
Source: ReBo AI Visibility method, September 2026.
Confusing the two layers is expensive in both directions. Treating a prompt like a keyword leads to one page per question, which does not scale: twenty long questions become twenty thin pages. Treating a keyword like a prompt leads to pages that rank and are never cited, because they never answer the question in the form it was asked.
The fix is to use the right layer for each. Short keyword in the <title> and the H1, where it still pays; long tail as H2 questions, each followed by an anchor answer of roughly 45 words. That short, self-contained, verifiable passage is what a model copies when it decides to cite someone.
How many target prompts should a company track?
In the ReBo method, 15 to 30 in the first baseline, split across three blocks: category, problem and brand. Fewer than that is not a sample; many more raises the cost of measurement without changing a decision, because source selection varies between runs.
The three blocks exist because they measure different things. Category covers questions where no one is named yet ("who does this?"): that is where new clients come from. Problem covers the pain in the buyer's own words ("my PDF letter never shows up in ChatGPT, what do I do?"): that is most of the long tail. Brand covers questions that already carry the company name: it measures what the model says to someone who has already heard of you, and it is the block where a wrong answer hurts most.
Every prompt has to run on more than one engine, because cited sources barely overlap between them. A Profound study of 100,000 distinct prompts, published in July 2025, measured 11.0% shared sources between ChatGPT and Perplexity, with a range of 6% to 16.4% across the engine pairs tested. Tracking one engine measures one market.
The practical cadence is monthly, always against the baseline, and the result feeds the citation rate and share of voice. It is Phase 5 of the ReBo method, and it is what makes it possible to state, with evidence, where a company sits on the AI Visibility maturity framework.
Related entries
The target prompt is what you measure; the citation rate is how you measure it. The aggregate result is the brand's AI Visibility. Before measuring anything, check that an AI crawler can read the site at all. Without that, no prompt stands a chance.
Frequently asked questions
How do you choose a target prompt?
Start from what a buyer asks before they know the company exists. The sources are the sales team, the questions that keep coming up in calls, the site's own search box and Search Console queries. A prompt invented in a meeting room measures the imagination of whoever wrote it, not the market.
Does a target prompt always return the same answer?
No. Source selection in generative engines is not deterministic: the same question, asked again, can cite different sources. That is why measurement uses repeated runs and a citation rate, never a single screenshot.
Do target prompts change over time?
They do, but the core set has to stay stable for the time series to mean anything. The practice is a quarterly review: new questions the market started asking come in, questions no real person asks go out.
Do you need a paid tool to track target prompts?
Not to start. A set of 15 to 30 prompts run manually once a month, with cited sources logged in a spreadsheet, already produces a valid time series. A paid tool solves collection scale and frequency, not the decision of what to measure.