How the AI Visibility Index Is Built | Composition Methodology

1. What the Index Measures

The AI Visibility Index ranks consumer brands by how well they appear in the answers generated by AI engines like ChatGPT, Gemini, and Perplexity.

Each brand in the Index is scored on a 0-100 scale by the same 35-signal deterministic scoring engine that powers GenSight.AI's brand audit product. The methodology for that engine — what each signal measures, how the score is composed — is documented separately at /methodology →

This page documents the other methodology question: how the brands are selected. Why these 50 and not other 50? How is the list composed, banded, and refreshed?

The two methodology questions

How is each brand scored?

35 signals across 4 pillars, deterministically weighted, sums to 100. Audit engine methodology →
How are the 50 brands selected?

Tranco-sourced, AI-classified, editorially reviewed. This page.

The Pipeline

2. How Composition Works

The Index is composed in three stages: source, classify, select.

Stage 1 · Source

Tranco

The starting point for every market's Index is the Tranco list, an academic traffic-rank list maintained by researchers at KU Leuven, imec-DistriNet, and Stanford. Tranco aggregates daily rankings from Cisco Umbrella, Majestic, Farsight, and Cloudflare Radar, deduplicates, and publishes a stable list of the most-visited domains on the web. We use it because its methodology is fully public, its data is independently verifiable, and its license permits redistribution in derivative work — three properties most commercial traffic-rank providers don't offer simultaneously.

We use a custom Tranco list configured to exclude non-commercial sources (parked domains, content-farm aggregators, technical infrastructure). The list is filtered to domains relevant to the market in question.

Stage 2 · Classify

What's a brand, and where does it belong

The filtered Tranco pool contains tens of thousands of domains. Most aren't consumer brands at all — they're B2B tools, content aggregators, government services, technical infrastructure. The classification stage reduces the pool to consumer-direct brands and assigns each to one of the Index's category buckets (10 categories per market).

Classification is performed by Google's Gemini language model with a structured, deterministic prompt. Each domain receives a category label or a non-Index rejection (B2B, aggregator, government, non-commercial, etc.). Brands the model cannot confidently classify are routed to editorial review.

Stage 3 · Select

Tranco rank within category, with banding

Within each category, the surviving candidates are sorted by Tranco rank. The mechanical default is straightforward:

Leaders

2 per category

The top 2 brands by Tranco rank within the category. The most-visited consumer destinations.

Challengers

2 per category

Brands ranked 3-20 by Tranco. Strong category players sitting below the very top.

Emerging

1 per category

A fast-growing brand ranked 21-50. Newer or less-established than the Leaders, but with momentum.

This produces 5 brands per category × 10 categories = 50 brands per market in the Index.

Editorial Integrity

3. Where Editorial Discretion Lives

The picking step allows up to one editorial swap per category in the Challenger band. We use this to move past brands that are technically high-ranked but methodologically unsuitable — a brand in administration, a brand that's functionally a duplicate of the Leader, a brand whose UK identity is borderline.

Leaders are never swapped. The top 2 by Tranco rank within category are always the Leaders. The Emerging band has more discretion than Challengers, because the rank-21+ pool is harder to default mechanically into a single best pick.

We're explicit about this because the alternative — pretending the list is purely algorithmic — would be dishonest. The Index is methodology-driven, not algorithm-only. The methodology produces the list; editorial discretion smooths the edges within published, bounded rules.

We don't publish the per-refresh swap log. Editorial decisions are recorded internally for our own audit trail.

Where editorial decisions can intervene

✓ Stage 2 misclassifications: a brand can be moved between categories during review.
✓ Stage 3 ambiguous classifications: reviewers make accept/reject calls on borderline cultural-perception cases.
✓ One Challenger swap per category is permitted in the picking step.
✕ Leaders are never swapped. The top 2 by Tranco rank are always the Leaders.
✕ Brands cannot be removed in response to objections about their score.

Two clocks, deliberately decoupled

Annual reshuffle

Every January

Composition is fully re-run. New Tranco snapshot, new classifications, new picks. The brands in the Index for a given year are fixed in January and don't change for 12 months.

Quarterly refresh

Apr / Jul / Oct / Jan

The 50 brands stay the same; their scores are re-audited. Detail pages update in place. URLs are unchanged.

4. How the Index Refreshes

The Index runs on two cadences, deliberately decoupled. Composition reshuffles annually; scores refresh quarterly. The 50 brands in the Index are stable for a year; the scores against those 50 brands are kept current.

A brand can only be removed from the Index mid-year if it ceases trading or is acquired and its identity is retired. It cannot be removed because of an unfavourable score. (See Eligibility & objections below.)

URL stability matters. Every brand in the Index has a permanent URL — /ai-visibility-index/{market}/{brand-slug} — which never changes across refreshes, even if the brand's name changes. SEO authority depends on this.

Eligibility & Objections

5. Eligibility & Objections

Brands cannot be removed from the Index because they object to their score. The Index reflects how AI engines perceived each brand at the most recent quarterly refresh. A score is not an opinion about the brand; it's a measurement of the brand's footprint in AI-generated text.

If a brand believes its score reflects a methodological error — an audit that hit a technical failure, a misclassified category, an outdated cached source — we'll re-audit and publish the updated score. Email contact@gensight.ai. We'll respond within five working days.

We don't accept payment in exchange for inclusion, exclusion, or score adjustment.

Per-Market · United Kingdom

6. The UK Index

The methodology choices specific to the UK Index — categories, the cultural-perception classifier, refresh status.

What "British" Means in the UK Index

We include brands that a UK consumer would perceive as British, regardless of who legally owns them. This is a methodological choice, and it's worth being explicit about.

An earlier version of this methodology used Companies House to verify that each candidate brand had a UK legal entity in good standing. We replaced that approach because legal incorporation doesn't predict how AI engines perceive a brand. Cadbury is owned by Mondelez International (American), but in AI-generated text about UK confectionery and high-street retail, it is unmistakably British. Dyson is now headquartered in Singapore, but every AI engine identifies it as a British brand because that's how the open web describes it. Conversely, Companies House would happily verify Dell UK Ltd or Samsung Electronics UK Ltd as legitimate UK entities, despite both brands being unambiguously American and Korean in consumer perception.

AI engines don't read company filings. They read the open web. Cultural perception is the methodologically honest signal for an Index that measures AI visibility.

We classify cultural perception by asking Gemini, for each candidate, whether a typical UK consumer would perceive the brand as British. Brands the model is confident about are auto-included or auto-rejected; ambiguous cases are reviewed by GenSight.AI's editors. The genuinely-borderline calls — where reasonable people might disagree about whether a brand reads as British — are a smaller subset of the editorial pile, but they exist, and we make case-by-case decisions based on how the brand is described in the open web.

This is a soft signal, not a deterministic one. We accept that some edge cases are subjective and that two reviewers might disagree on a small number of borderline brands. We document our specific calls in the internal record so they can be reproduced and challenged.

Current refresh status

Live

Current

Q2-2026 refresh

Next score refresh

July 2026

Next composition reshuffle

January 2027

Per-Market · United States

7. The US Index

The US Index is in development. This section will be filled in when it ships.

Honest Limitations

8. Honest Limitations

Methodologies should be honest about what they don't yet do well. For the current UK Index:

01

Tranco rank is a proxy for consumer attention, not a direct measure.

A brand can be high-traffic in raw terms but niche in cultural footprint, or low-traffic but iconic. We use Tranco because it's the best openly-published proxy available; we don't claim it's perfect.
02

The cultural-perception classifier is a language model's judgement.

It reflects the open web as Gemini was trained on it. New brands, recently-acquired brands, or brands undergoing repositioning may be misclassified. Editorial review catches most cases; not all.
03

The Emerging band is editorially-led.

The mechanical default (rank 21 within category) is a weak proxy for "fast-growing brand." Editorial discretion is heavier in this band than in Leaders or Challengers.
04

Tranco rank shifts over the year.

The brands in the current Index reflect their Tranco rank at the time of the most recent composition reshuffle. If a brand's rank moves significantly between reshuffles, that doesn't change its Index inclusion until the next annual reshuffle.

Questions

For methodology questions: contact@gensight.ai

For the underlying scoring engine methodology — what each of the 35 signals measures and how the score is composed — see the audit engine methodology →

View the UK Index → Read the audit engine methodology

How the
Index Is Built.