How we measure AI visibility

Each branch tracks questions from a shared library, written the way sellers, landlords and buyers actually ask, grouped by what they want:

• Selling
• Valuation
• Letting
• Landlords
• Reviews and comparison

Questions are filled in with the branch's town and the neighbourhoods it covers, for example “best letting agent in Southsea”, and balanced across these groups so no single type dominates. Agents can edit, switch off or add questions within their plan.

We ask ChatGPT, Perplexity, Gemini, Claude through their official APIs, with each assistant's live web search turned on, so answers reflect what they find on the web today rather than only what they learned in training. We use current models, kept up to date as the providers release new ones, and set the location to the UK and the branch's town where the assistant supports it.

Every question starts from a clean slate: no account history, no previous conversation, no personal settings. That keeps results comparable from week to week and between agencies, and is why they can differ from what you see in your own logged-in app.

AI answers vary from one go to the next. On paid plans we ask every question 3 times on each assistant, every week, and report how often you're named across all of those answers. One answer is a snapshot; a rate across many is a measurement.

When a figure rests on only a few answers, we mark it Low sample rather than present it as settled.

Each answer is stored word for word. A language model then reads it and lists every estate or letting agency it names, in order, with how the answer describes each one and the websites it cites. Portals, review sites and solicitors are left out.

We match those names to your branch and its competitors. Matching ignores punctuation, generic words like “Estate Agents” and “Ltd”, and town or neighbourhood suffixes, so “Hunters Southsea” and “Hunters” count as one agency. A website that clearly belongs to an agency also counts as a match.

When a name only partly matches, such as a sub-brand, we flag it for review instead of counting it. An answer that couldn't be read at all is left out of every figure, never counted as “not named”.

A cited website counts for an agency only when the cited page is about that agency, for example its profile page on a directory. Appearing in the same answer isn't enough, because a well-known agency appears alongside almost every common source.

Agencies' own websites are labelled as agency sites and never treated as gaps: a competitor's website will never mention you. Gaps are third-party sites, such as directories and review platforms, that AI reads for other agencies but where it found nothing about you.

Visibility, Position, Sentiment and Share of voice are each counted from stored answers using one shared set of formulas, so every screen and report agrees. Plain-English definitions with examples are in What the numbers mean. Every number in the app links to the answers behind it.

Fixes come from rules applied to your own results, website and Google profile. A rule only produces a fix when we have the evidence for it, and each fix shows that evidence. When you mark a fix done, we check it again on the next scan.

We report what AI assistants say and what to fix. We can't promise how any assistant will answer.

AI assistants change how they answer, which sources they read and which models they run. We measure what they say, consistently, and show you the evidence. We don't control what they say.