Prompt-Level Ranking

Overall visibility figures hide the detail. A brand can look healthy in general and still be missing from the exact questions buyers ask when choosing. Tracking by prompt shows which questions the brand wins, which a competitor owns and which produce a wrong answer about the business. Each of those needs a different fix.

It also makes progress easy to show. A new comparison page can be linked to the four prompts it was written for, and the next month's runs show whether those prompts moved. That is much harder to see in a single visibility score.

Because answers vary between runs, the figure is a rate and needs enough runs to be trusted. A change from two out of ten to three out of ten is within normal variation. A move from one in ten to seven in ten is not.

1. Write 20 to 50 prompts that mirror real buying questions, from sales calls and customer emails.
2. Run each on a schedule across the engines your buyers use, several times per run, since answers differ.
3. Log whether the brand is mentioned, cited or recommended, and where in the answer.
4. Sort results into three groups: prompts won, prompts missing and prompts answered wrongly.
5. Attach an action to each group. Missing needs new content, wrong needs a clearer page, competitor-owned needs stronger evidence.

In classic search, a keyword has a position, such as third on page one. AI engines do not publish a fixed list. They write an answer, and that answer may mention a brand, cite its page, recommend it or leave it out. Prompt-level ranking records that outcome for each specific prompt, such as "best invoicing software for a three-person agency".

Because answers vary between runs and between engines, the record is usually a rate. For example: named in six of ten runs, cited as a source in two.

• Treating one run as a ranking. A single answer is a sample.
• Tracking broad topics instead of full questions.
• Counting mentions without checking whether the answer was accurate.