June 21, 2026 · AI Visibility
Ask ChatGPT for “the best CRM for small agencies” or “a reliable plumber in Austin,” and it will happily name a few companies. For the brands it names, that is free, high-intent exposure. For everyone else, it is an invisible loss, because none of it shows up in Google Analytics. So how do you actually track it?
Why screenshots alone don’t work
You can open ChatGPT and ask once, but a single answer tells you almost nothing. Model answers vary by phrasing, by location, by account, and over time. To get a signal you can trust, you need to ask many variations, repeatedly, and record the results consistently.
A repeatable method
- List the prompts that matter. Start from real buying questions: “best [category] for [audience],” “[service] near me,” “alternatives to [competitor],” “is [your brand] any good?”
- Decide what counts. Define a mention (named at all) versus a recommendation (in the shortlist) versus a citation (your site linked as a source).
- Run them on a schedule. Weekly is a good baseline; daily for fast-moving categories.
- Log the details. For each run: were you mentioned, in what position, with what sentiment, and which sources did the model cite?
- Track share of voice. Compare your mention rate with named competitors, so you have a benchmark, not just a number.
What to do with the data
The pattern you are looking for is simple: prompts where a competitor is recommended and you are not. Each one is a content brief. Models tend to cite pages that are easy to quote, so the fix is usually some mix of:
- A clear comparison or “best X for Y” page that includes you on the merits.
- Original data, or a strong, specific answer to the exact question.
- Reviews and third-party mentions that reinforce the model’s confidence.
- Structured data (FAQ, How-To) that maps to the prompt.
Doing it without living in a spreadsheet
You can run this by hand, and it is worth doing once to feel how it works. But it gets old fast across dozens of prompts and several engines. Hypado automates the whole loop: it runs your prompt set across ChatGPT, Gemini, Perplexity and more, tracks mentions, citations and share of voice over time, and points you at the exact pages to create. You can even connect your own ChatGPT or Claude account, so the assistant answers using your real data and drafts the fixes for you. See your first report free.