How to Build the Prompt Set You Track, Before You Buy a Tracking Tool

Every AI visibility tool reports on the questions you feed it, so the question list is the measurement instrument and it deserves building first.

An AI visibility tool monitors a list of prompts you supply. Feed it a weak list and it will report precisely on the wrong thing, month after month, with charts. The list is the instrument.

The method below is Subscribe PR’s own framework. No published study prescribes how to construct a prompt set, and this article reports no results from running it. Treat it as a procedure to adapt, not a finding.

Step one: write the questions, not the keywords

Start from the questions a buyer asks in the weeks before hiring you, phrased the way they would type them. Pull them from sales call notes, intake forms and the first three emails of any new inquiry.

Keep the awkward ones. A prompt like “is it worth paying a lawyer for a small claim or can I do it myself” sits closer to real usage than a tidy service-name phrase.

Step two: size the set deliberately

This framework uses 30 to 50 prompts per client. That range is a judgement call rather than a researched optimum, chosen so a person can read every answer by hand in a sitting.

Split the set into three groups: questions about the problem, questions comparing options, and questions naming you or a competitor. The first group is where you are least likely to appear and most likely to learn something.

Step three: choose engines against their behaviour

Engines differ in what they consult. Goodie, a vendor that sells answer engine optimization software, measured 31 million citations across 11 AI surfaces from October 2025 to July 2026 and found news share ranging from 23% on Grok to 5.1% on Claude, with AI Overviews at 19.5%, Gemini at 13.5%, ChatGPT at 11.9%, AI Mode at 10.6% and DeepSeek at 7.9%.

Semrush, working from over 1 billion lines of US clickstream data and a 200 million user panel, found ChatGPT enabling web search on 34.5% of queries in February 2026, down from 46% in late 2024. Run each prompt on at least three engines so one product’s behaviour does not stand in for the whole market.

Step four: record the domains, not only your presence

For every answer, log each cited domain alongside whether your brand appeared. The domain list becomes your media plan and it is the more useful half of the data.

Pew Research Center, reviewing 68,879 Google searches from 900 US adults in March 2025, found that 88% of AI summaries cited three or more sources and only 1% cited one, with a median summary length of 67 words. Expect several domains per answer and a short answer that holds few brand names.

Step five: note the page, not just the site

Similarweb, measuring about 600,000 citation events in January and February 2026, found roughly a third of ChatGPT citations coming from pages three folders deep in a site’s URL structure.

Capture full URLs in your log. A pattern of deep, narrow pages tells you what to publish, and a pattern of homepages tells you something different.

Step six: fix the cadence and freeze the wording

Re-run the same prompts on the same schedule with the wording unchanged. Any edit to a prompt breaks the comparison with every prior run, so keep a dated version of the set and add new prompts as a second cohort.

Engine behaviour moves underneath you. OpenAI’s documentation notes that robots.txt changes take about 24 hours to register, and the Semrush figures above show a product’s search rate shifting across a single year. A fixed set is what lets you tell platform change apart from your own progress.

Then buy the tool

With a finished set, you can evaluate vendors on whether they cover your engines, store full source URLs and let you lock prompt wording across runs. Those three questions are answerable in a demo.

You can read how this framework fits a wider program at subscribepr.com. The free AI visibility audit at subscribepr.com/audit runs a starter set against your category so you have a baseline before any subscription.

Subscribe PR is an answer engine optimization and digital PR agency in Phoenix.

Add a Comment