How Mentsight will measure AI visibility

How websites are analysed

When a user adds a website, Mentsight will read a limited set of public pages to understand what the business does, who it serves, which services and locations it describes, and what language it uses. It will not edit the site, connect to a CMS, or treat the crawl as a complete website audit.

How prompts are generated

Mentsight will draft customer-style prompts from the observed business context. The planned prompt categories are:

• Discovery: broad questions about who can help.
• Services: questions about a specific service or need.
• Location: questions involving a city, region, or service area.
• Comparison: questions that put businesses or approaches side by side.
• Audience: questions about fit for a type of customer.

These categories cover different customer intents. They are not a claim that every possible search is represented.

Why users approve prompts

Prompt wording changes the answer. Human review helps remove questions a real customer would not ask, correct a wrong assumption, and add a useful question Mentsight missed. Only approved prompts will be sent for analysis.

How requests are run

Each approved prompt is sent to every assistant in the website's plan in five separate rounds. Mentsight keeps every response and citation returned, then reports the observed mention and citation rates. A finished report stays tied to its original assistants, rounds, answers, and citations.

How the visibility score works

The visibility score is a 0–100 summary of one report. It is not an industry rank, and it cannot be compared against a score from any other tool. It answers one question: across the questions you approved, how often did this assistant put you in its answer?

• Being named counts for 70%. How often an answer named your business at all.
• Having your own pages cited counts for 30%. How often an answer used a page on your domain as a source.
• Questions that name your business are left out of both halves. Asking an assistant about you by name guarantees it says your name, so those answers measure how you are described rather than whether you surface on your own.
• Position is deliberately not part of the score. Most answers are not ordered recommendations, so scoring position would reward the shape of an answer rather than the evidence in it.
• Answers that never came back shrink the basis rather than counting against you. A question the assistant failed to answer is not evidence that you were absent from it.

Because each question is asked five times, the score is a sample rather than a fixed figure, and every report states the range the reading could reasonably land in. Read that range, not the single number. Two reports have only moved when their ranges do not overlap.

Which AI assistants are asked

Mentsight asks the assistants a customer would actually use, and every answer, figure, and citation in a report names the assistant that produced it. A website owner chooses which of these to enable in settings, and a finished report keeps the assistants it was measured with.

• ChatGPT (OpenAI): searches the web as it answers, and the target market is set through the vendor's API.
• Gemini (Google): searches the web as it answers, and the target market is stated in the question itself.
• Claude (Anthropic): searches the web as it answers, and the target market is set through the vendor's API.

Because the market is applied differently by each vendor, the same question can be grounded slightly differently from one assistant to the next. That is a property of the assistants, not a measurement error.

Not measured: Microsoft Copilot, Perplexity, Grok, and AI answers shown inside a search results page. No public interface returns those answers in a form Mentsight can record honestly, and Mentsight does not scrape any assistant's web interface. A report says nothing about visibility in an assistant that was not asked.

What the evidence terms mean

Mentioned

The business name appeared somewhere in the answer. This does not imply a position or level of prominence.

Cited

One of the returned citation links points to the submitted website. External links remain separate.

Position

The business's order among explicitly recommended businesses. It is recorded only when the response actually presents an ordered recommendation.

Opportunity

An evidence-backed idea to investigate, based on observed answers and site context. It is not a guaranteed improvement.

How competitors are identified

Competitors are other businesses that appear in answers for the approved prompts, including answers where the submitted business is absent. They are not a complete market map or proof that every named business competes directly.

How opportunities are generated

Opportunities connect a recurring pattern to something worth checking. For example, an answer may name competitors for a service and location while the submitted site has no clear page covering both. Each opportunity should point back to the answer or page that suggested it, and a person should review it before making a change.

Why reports change

• The model or provider may change.
• The web index and cited pages may change.
• Prompt wording or the approved prompt set may change.
• AI output can vary between runs.
• The business website and competing public pages may change over time.

Because of that variation, one report is a sample rather than a fixed fact. Every assistant receives its own reading and 95% range over questions × completed rounds. Assistants are never pooled as extra samples. The combined figure is an unweighted assistant average whose range is no narrower than any assistant behind it. Two reports are only described as having moved when they used the same assistant set and their ranges do not overlap.

What Mentsight cannot determine

Mentsight cannot guarantee an AI ranking, provide a universal visibility score, attribute every result to one specific page or change, or reveal proprietary ranking algorithms. A citation or mention is evidence of one observed response, not a forecast. Read the AI and agent guide for the current product scope and limitations.