State of AI Visibility 2026

Finding 01

ChatGPT leads 59% of brands. 5 engines lead zero.

Across 75 brands, 44 have ChatGPT as their top citing engine — a 59% concentration that exceeds even ChatGPT's consumer-AI market share. Meanwhile 5 of the 9 engines we track (Grok, DeepSeek, Llama, Google AI Overviews, Microsoft Copilot in this cohort) lead zero brands. That is not a measurement artifact — it is the cohort. Most brands have invested in ChatGPT visibility without yet building the Wikipedia, schema, and Bing-grounded foundations the other engines reward.

Finding 02

Consumer brands outscore B2B by ~30 points, consistently across all three cities.

The pattern repeats in every regional cohort. OpenAI (AI platform, score 96) leads the cross-cohort ranking; Edge Group (Defence + tech, score 28) anchors the bottom. Across all 75brands, mean consumer-category score sits roughly 30 points above mean B2B-category score. The structural cause: consumer brands accumulate web mentions through reviews, social discussion, and category-specific PR that B2B brands don't generate at the same scale. Wikipedia coverage is also markedly stronger for consumer brands. Both feed every engine's training corpus.

Implication for B2B operators: assume you start ~30 points behind a consumer-category equivalent and that the gap closes only through deliberate Wikipedia, Wikidata, G2/Capterra, and podcast-appearance work. Standard "publish blog posts on our domain" content rarely closes it.

Finding 03

Schema adoption is uneven across cohort. llms.txt adoption is near-zero.

Spot-check of the 75 brand homepages: roughly 60% emit at least one JSON-LD block, but only ~25% emit complete Organization + Product/Service + FAQPage triplets. Of the 25 brands per regional cohort, fewer than 3 publish an llms.txt file. The brands at the top of each regional ranking (OpenAI et al.) are disproportionately the ones with complete schema — the correlation is not subtle.

Implication: schema is one of the cheapest visibility moves available. Brands without complete Organization + sameAs to Wikidata leave compound AI-engine recognition on the table. The schema generator produces clean output in <5 minutes per page.

Finding 04

Regional variation is real but smaller than category variation.

The three regional cohorts (New York, San Francisco, Dubai) differ by mean score by under 8 points. The variance within any single cohort, by category, is 40+ points. AI engines do not optimise their citations by city; they optimise by entity strength. Regional brand-strategy framing has marketing value but limited GEO leverage. Category strategy carries more weight.

Finding 05

Negative legacy events still dominate AI citations years later.

Three brands in the cohort carry visible AI-citation drag from incidents three+ years old (MoviePass: 2019 collapse, Bird: micromobility category contraction, WeWork-era touchpoints in adjacent listings). The training-corpus weighting of authoritative news coverage means these incidents define the brand for AI engines even after the brand itself has moved on. Standard reputation-management tactics don't fully resolve this — the news coverage is the corpus.

Implication: brands recovering from a public incident need 2-3 years of new, dense, authoritative coverage before the original event stops dominating AI responses. Wikipedia article updates with verifiable post-incident milestones accelerate this faster than blog content.