Track ChatGPT mentions

Build a stable set of buyer questions

Start with 20 to 30 questions a buyer could ask before choosing a product. Split them into category discovery, problem, comparison, alternative and trust questions. Keep the wording stable between runs. Otherwise a prompt change can look like a visibility change.

• Category: “What are the best tools for [job]?”
• Problem: “How can a [buyer] solve [pain]?”
• Comparison: “Which is better for [constraint], A or B?”
• Trust: “Which providers have proof for [outcome]?”

Record more than a brand mention

For every ChatGPT answer, save the date, exact question, model or surface, brands named, order of appearance, cited domains, cited URLs and a short accuracy note. A mention without a citation or with the wrong positioning should not count as a clean win.

Separate four different failures

Treat absence, wrong description, missing citation and competitor displacement as separate findings. They require different fixes: category clarity, factual correction, source-worthy evidence or stronger comparative coverage. Combining them into one visibility score hides the next action.

Connect each finding to one source

Map each weak answer to the page that should resolve it. A category question belongs to a clear product or category page. A comparison needs explicit criteria. A trust question needs named proof, methodology or a credible third-party source.

• Rewrite vague product descriptions with audience, job and constraint
• Add verifiable customer, methodology or benchmark evidence
• Create a fair comparison that states who each option suits
• Link related pages so the strongest source is easy to discover

Review changes on a useful cadence

Run the set weekly while pages are changing and monthly when the site is stable. Compare several runs before calling a movement real because model updates and answer variation can affect a single observation. Keep screenshots for context, but make decisions from the structured history.