Last updated: 12th September 2026
Entity Confidence publishes research into how AI-powered search engines respond to questions about businesses. This page explains how that research is produced, what it can reasonably be taken to show, and where its limits lie. It is written so that anyone who encounters one of our findings can understand exactly what was done and challenge it if they wish.
This page covers our published research. The assessment methodology behind our client reports is described within those reports.
What we test
Engines. We use one or more of ChatGPT, Claude, Gemini and Perplexity.
Questions. We issue natural questions of the kind a member of the public would ask when looking for a service, rather than technical queries or brand-name searches. [INSERT: description of the question set and how it is constructed]
Sample. When businesses are selected for study the base information is drawn from publicly available sources such as business directory listings or regulatory publications.
Each study states its own engines, question set, and sample. Where these differ from the general description above, the study takes precedence.
How we run queries
Queries are issued through the standard public interfaces of each engine. We record the question exactly as issued, the engine, the date and time, and the location settings in use. We capture the response in full rather than summarising it.
We do not attempt to influence results, and we have no commercial relationship with any of the engines we test other than the payment of any required usage fees.
What we record
For every finding we publish, we retain a record containing the question as issued, the engine and version, the timestamp, the location settings, and the verbatim response. These records are kept for a minimum of two years.
If you want to know the basis of a specific finding, ask us. We will tell you what was asked, when, and what came back.
How we report findings
We do not name businesses in connection with performance findings. Businesses appearing in our research are described by characteristics — size, setting, region, ownership structure — rather than by name. We name a business only where that business has given written permission, or where the reference is a neutral, already-public fact such as regulatory registration.
This is a deliberate position. Results from AI-powered search engines shift over time, and we do not think it is fair to attach a business’s name to a snapshot it cannot control and may not have seen.
What these findings can and cannot show
They are dated observations, not fixed properties. A result reflects what a particular engine returned to a particular question at a particular moment, from a particular location.
They vary. AI-powered search engines are not deterministic. The same question, asked twice, can produce different answers. Our own testing measured a [INSERT: verdict-flip rate] rate of change in outcome across repeated queries. We regard this as a central feature of the subject rather than a caveat, and we design our studies around it by aggregating across many queries rather than relying on single results.
They are not a ranking. We do not publish league tables, and a finding about one business should not be read as a comparison with another unless the study explicitly makes that comparison.
They are not a judgement about the business. We observe what engines return. We do not infer anything about the quality of a business, the competence of its owners, or the effort they have put into anything.
Dates and re-testing
Every published finding carries its date of capture. We do not present older findings as current. Where a study is re-run, the new figures are published as a new dated finding rather than as an amendment to the original.
Questions and challenges
If you believe something we have published is wrong, or you want to see the basis for a finding, please see our corrections and right of reply page or write to information@entityconfidence.ai. We respond to every query and we do not amend or remove published material quietly.
Entity Confidence Ltd
information@entityconfidence.ai entityconfidence.ai
Be the business AI chooses