Research
AI Visibility Research and Methods
Understand how a check is collected, what its results mean, and which findings are ready to use.
Available now: measurement method
Our methodology explains controlled buyer questions, named model versions, application-provided search evidence, saved answers, and the limits of comparing checks over time. API checks do not reproduce a person's conversation in a consumer AI app.
Available now: evidence standards
Our editorial policy explains how we label examples, review claims, disclose our commercial interest, and correct errors. A product illustration explains a workflow; it is not a customer outcome.
In progress: buyer-question pilot
The pilot has a target of 300 observations. It is not a completed study and has no published findings. Its study page shows collection status and remains excluded from search indexing until the publication requirements are met.
Customer evidence
The case-study page describes the evidence required for a customer result. It does not currently establish a measured customer uplift. Published outcomes require permission, original observations, dates, an explained intervention, and a comparable recheck.
Reading a change over time
Compare the same questions, model versions, search settings, and brand-matching rules. Show completed and planned checks together. A later improvement can be consistent with a useful change without proving that a page edit caused it.
How to evaluate an AI visibility report
- Read the exact buyer question and collection dates.
- Check whether the evidence comes from an API benchmark or a consumer application.
- Keep failed and unavailable checks outside recommendation-rate denominators.
- Inspect the saved answer and source references behind each conclusion.
Sources
Conclusion
Start with the method and the evidence behind a claim. We publish study findings only when the underlying observations support them.
Frequently Asked Questions
Is the 300-observation study complete?
No. It is a planned pilot without published findings. Use the methodology and evidence policy now; do not quote the target sample size as collected data.
Do API results reproduce ChatGPT or other consumer apps?
No. Our benchmark uses named model APIs and application-provided search evidence. Consumer apps can use different retrieval, instructions, personalization, and settings.
Does a missing answer mean the brand was not recommended?
No. A failed or unavailable check has no valid recommendation result. It must remain separate from a completed answer that omits the brand.
Who publishes this research?
In The Answer publishes these methods and sells AI visibility software. That commercial interest applies to our research and comparisons. Questions and corrections can be sent to support@intheanswer.ai.