Marketing and product leadership
How to measure whether ChatGPT, Gemini, Perplexity and Claude recommend you
By the MarqIQ operations team · Updated
The short answer
Pick the questions your buyers ask before they know your name, run them on a schedule against each assistant, and record four things per answer: whether your brand is named, how it is described, which competitors are named beside it, and which sources are cited. Track the mention rate over time against a fixed question set. MarqIQ’s AI Answer Visibility does this across ChatGPT, Gemini, Perplexity and Claude.
Start with the questions, and leave your name out of them
The most common mistake is tracking prompts that contain the brand. "Is Acme any good?" will always mention Acme. It tells you about sentiment and nothing about discovery. The questions that matter are the ones a buyer asks in week one: "what tools write lead source into HubSpot", "how do I prove analyst relations ROI". If you are absent from those answers, you are absent from the shortlist.
Twenty to forty questions is plenty. Write them from sales call notes and search queries, not from a keyword tool. Keep the set stable, because a mention rate is only comparable month to month if the denominator is the same set of questions.
Score four things
- Mention rate: the share of answers that name you, per assistant and overall.
- Share of voice: your mentions against named competitors. It is meaningless until you have entered competitors, and a tool that shows you 100% without them is flattering you.
- Sentiment: how the answer characterises you where it does mention you. Worth reading in full, since a score hides the sentence that matters.
- Citations: the domains the assistant quotes. This is the actionable part. If every answer in your category cites the same three review sites and one community thread, that is your publishing plan.
Be strict about the denominator
Assistant APIs fail, rate-limit and time out. If eight questions were asked and two came back, a tool that reports "100% coverage" on the two is reporting nonsense. Count against the questions asked, show how many engines actually answered, and refuse to score a run that mostly failed. We rebuilt our own scoring around this after auditing it, and we would hold any vendor to the same standard.
One more caveat. Answers from an API are not identical to what a logged-in person sees in the consumer app, which may browse the web, personalise and vary by region. Treat the score as a consistent index of direction, not a census of what every buyer reads.
What actually moves the number
Assistants repeat what is stated clearly, consistently and in more than one place. In practice that means: pages that answer a specific question in the first two sentences, product facts that match everywhere they appear, presence on the third-party sites the engines already cite for your category, and an open door for their crawlers. Blocking GPTBot and then wondering why ChatGPT has not heard of you is more common than it should be.
MarqIQ’s recommendations are tied to the citation list for your own tracked questions, so the advice is "this domain is quoted in six of your missed answers" and not a generic checklist.
Quick answers
How do I check if ChatGPT mentions my brand?
Run a fixed set of buyer questions that do not contain your brand name against the assistant on a schedule, and record whether each answer names you, which competitors it names and which sources it cites. Track the mention rate over time.
What is AI answer visibility?
It is how often, and how favourably, AI assistants such as ChatGPT, Gemini, Perplexity and Claude name your brand when buyers ask category questions. It is also called answer engine optimisation (AEO) or generative engine optimisation (GEO).
Which MarqIQ plans include AI Answer Visibility?
AI Answer Visibility is included from the Business plan. Recommendations and the citation list behind them are part of Enterprise.
Read next
Check your own numbers first
Connect one tool and run the audit. It shows how much of your lead source data is missing, and the report is yours to keep either way.