How to Check Whether ChatGPT Recommends Your Brand
A practical method for testing AI brand recommendations without confusing one answer with an official ranking.
Published 5 August 2026 · Evidence-led guideStart with brand-neutral buyer questions
Questions that include the target brand force it into the answer and reveal little about natural discovery. Use recommendation, comparison, trust, problem-solving and decision questions instead.
Keep the exact wording, language and market fixed. That makes later measurements comparable.
Test more than one AI system
OpenAI, Gemini and Claude can use different search results, sources and model behavior. A brand may appear in one system and be absent in another.
Display provider failures separately. A timeout is not evidence that a brand has zero visibility.
Separate mentions, position and citations
A mention shows that the brand name appeared. Position shows where it appeared in a recommendation. An official citation shows whether the answer used the brand's own domain as evidence.
These signals should remain separate because a brand can be mentioned while competitors or publishers receive the citations.
Repeat important prompts
Generative answers vary. A single run is useful for discovery, while repeat runs reveal an observed range and whether a result is stable.
After improving a page, rerun the same prompt set. Changing the prompts at the same time weakens the before-and-after comparison.
FAQ
Is this the same as an official ChatGPT ranking?
No. It is a standardized benchmark for documented prompts, language, market and measurement time.
How many prompts should a first audit use?
A focused set of 10–20 commercially relevant prompts is usually more useful than hundreds of vague questions.
Should a failed provider call count as zero?
No. It should be marked unavailable and excluded from the score.