How the index is measured.

What a sector is

A sector is a buyer market: a thing people ask AI engines to recommend, a typical buyer, and a region. “CRM software for a small business in the UK” is a sector. “App development agencies in London” is another. We cover B2B software, services, consumer brands and marketplaces, in the UK, the US, Europe and worldwide, and add sectors continuously.

Each sector has ten unbranded buyer questions. They are written to sound like a real buyer, not a marketer: a recommendation, a comparison, an alternatives question, a budget question, a reviews question. No question names a brand except the alternatives question, which names one well-known incumbent and asks what else exists.

How the engines are asked

1. Every question is sent to every connected engine: ChatGPT (OpenAI Responses API with web search), Claude (Anthropic API with web search), Perplexity (Sonar) and Gemini (Google Search grounding). Live web search is on for all of them, so answers reflect what the engines read today rather than their training data.
2. Each question is a fresh conversation with no system prompt, no persona and no prior turns.
3. The full answer text and every citation URL the engine returns are stored.
4. A separate model pass extracts the companies, products or brands the answer recommends or names as options, in the order they first appear. AI platforms themselves, generic terms and publishers cited only as sources are excluded.
5. Brand names are normalised to a single key so that spelling and punctuation variants count as one brand. Where two keys turn out to be the same brand, they are merged and the period is re-aggregated.

How the score is calculated

For each brand in a sector and period, across all answers from all engines:

Rank is the order of brands by score, then by raw mentions, then alphabetically. The same calculation is repeated per engine, which is where the per-engine ranks on each page come from. Average position is the mean of the brand’s position in the answers that named it.

The weights are a judgement. Being named at all matters most, because an unnamed brand is not considered. Being early matters, because AI answers are short and the first name frames the rest. We publish the weights so the number can be argued with.

Change over time

Each sector is re-measured every month with the same questions. The change column on a sector page is the brand’s rank movement against the previous period. A brand marked new was not named in the previous period at all. Brand pages keep the full history.

What it does not tell you

Engines are not deterministic. The same question can produce a different list on a different day, which is why the index aggregates across many questions and engines rather than reporting single answers, and why a one-place move between periods is noise rather than signal. Treat the top three as robust, the top ten as indicative, and movement of three places or more as real.

The questions are in English and the regions are coarse. A sector labelled “UK” tells you about English-language answers for a UK buyer; it does not tell you about regional variation within the UK or about answers in other languages. The engines’ own behaviour also changes: model updates, search index changes and product changes all move rankings in ways that have nothing to do with any brand.

The index measures recommendation, not accuracy. A brand named first is not necessarily the best choice; it is the one the engines are most likely to give. That is exactly why it is worth measuring.

Who runs it

The measurements are produced by GEOffice, Wall & Fifth’s own AI-visibility tool, and published here as open research. Wall & Fifth is a UK digital product studio that helps businesses get cited by AI engines by rebuilding the pages and entity signals the engines rely on. The index is how we show the problem is measurable, and it is the baseline we use when we work with a client.

Back to the index