Write natural buying questions
Use the questions a customer would actually ask. Include a known-product check, an unbranded category search, a problem-led request, and a competitive choice.
Practical guide
See what AI agents think of your product.
Ask natural buying questions, let each AI client research independently, and preserve the evidence behind the answer. The result shows whether your product was found, understood, trusted, and chosen.
ChatGPT · ClaudeGemini · Grok
The short answer
Test the customer's decision, then inspect where it changed.
A generic website audit shows what an agent notices when it is told to inspect your page. A product-perception test starts with the customer's actual question. That gives the agent room to miss your product, misunderstand it, distrust a claim, or choose a stronger alternative.
Run the same questions across the AI clients your customers use. Preserve each answer independently before a separate analysis turns the traces into page and positioning work.
See the complete methodologyRun each question in a fresh conversation
Give every AI client the same situation. Let it decide what to search and read without telling it which claims, pages, or competitors to inspect.
Preserve the complete research record
Keep the answer, cited sources, reported searches, surfaced pages, alternatives, and final decision. Label any search or visit detail the provider does not expose.
Compare the decision paths
Look for where the product disappeared, what the client misunderstood, which evidence changed trust, and why an alternative fit the request better.
Change the relevant page and repeat
Rerun the matched situations after the release. Compare only compatible completed paths and read every change alongside its current evidence.
What the paths reveal
Turn a vague AI answer into specific product work.
The product never entered the option set
Inspect the language the agent searched, the products and sources it found, and the category or problem association your public pages need to establish.
The product entered with the wrong story
Compare the agent's description with the intended positioning. Strengthen the page, examples, and internal links that explain the product's job and fit.
A claim lacked enough support
See the exact concern, the evidence the agent used, and the proof, policy, pricing detail, or company information that would make the decision easier to support.
A competitor fit the request better
Study the winning alternative, the constraint it satisfied, and the page or product distinction that needs to become clearer in the next release.
Common questions
Keep the test natural and the evidence useful.
What does it mean for an AI agent to think something about a product?
It means the agent formed a product representation and decision from its instructions, web research, public sources, and the customer request. The useful evidence is what it found, how it described the product, what raised confidence or concern, and what it recommended.
Should I ask the agent to audit my website?
Use a natural customer request when you want to understand real recommendation behavior. An audit prompt tells the agent what to inspect and produces a different task. You can use an audit later to investigate a specific issue found in the natural run.
Why test several AI clients?
Each client can search differently, rely on different sources, and weigh evidence in a different way. A product can be discovered by one client, missed by another, and recommended for different reasons across the panel.
How often should I repeat the test?
Run a baseline, then repeat after a meaningful website, product, pricing, or positioning release. Recurring AI visibility monitoring is useful when you also need continuous mention and citation trends.
Your first major-client panel is free