Does AI describe brands the way their homepages do?

A descriptor is a word or short phrase an AI answer attaches to a brand, such as "robust", "user-friendly" or "steep learning curve". We took every descriptor AI engines used for a brand in two or more answers and searched for it on that brand's crawled homepage.

Of 2,902,786 such descriptor uses across 4,583 brands, 83.0% were words the homepage never uses. For the median brand the figure was 83.3%. AI does not describe a brand in the brand's own words.

The headline block is the part of a homepage that states its positioning: the page title, the meta description, the main heading (h1) and the first paragraph. 95.4% of AI descriptor uses never appear in it.

Counted by distinct pairs of brand and word, rather than by uses, 89.5% of the words AI attached to a brand were absent from the whole homepage.

The match score is a 0 to 100 text similarity score between the words AI used for a brand and the brand's headline block, with the brand name removed. 100 means the same wording. It is a TF-IDF cosine similarity.

The average brand scored 9.9 against its own homepage and 0.7 against other brands' homepages, about 14 times higher. Within the same market, AI's description of a brand matched its own homepage better than a rival's homepage in 78.0% of 34,008 pairs. In 40.4% of brands it matched the brand's own homepage best of all brands in its market. AI describes the right product; it uses different words to do it.

An added descriptor is one AI used for a brand in two or more answers that never appears on the brand's homepage. The most common added descriptors are general praise. AI called 3,876 brands "excellent" whose homepages never use the word, "strong" for 3,336 brands and "robust" for 2,729.

Phrases that rank one brand against others were added almost every time AI used them: "highly regarded" and "strong choice" never appeared on the homepage of any brand AI applied them to, and "gold standard" was missing from 98.7%.

Negative descriptors were the least likely to come from the homepage. 95.2% of negative descriptor uses never appear on the brand's homepage, against 82.4% of positive ones. AI called 1,090 brands "expensive" and 1,039 "limited" that never describe themselves that way, and added "steeper learning curve" for 405 brands.

AI also places brands in tiers the brand does not name. It added "moderate" for 597 brands, "traditional" for 484, "basic" for 382 and "enterprise oriented" for 230. A pricing or complexity position that the homepage leaves unstated is filled in by the answer.

A headline claim is a descriptive word or phrase, such as "open source" or "secure", in the homepage headline block. We checked 58 common claim words across 7,274 claims on 3,435 homepages. AI repeated the claim for that brand in two or more answers 47.5% of the time.

Concrete product facts were repeated most. 89.6% of brands that say "open source" in their headline block were called open source by AI, and 74% of brands claiming "real time" and 71% claiming "AI-powered" had that claim repeated. "Best" was repeated for 84%, mostly because AI calls most brands best.

Claims about reputation and novelty rarely carried over. Of 303 brands that call themselves "trusted" in the headline block, AI called 11.6% trusted. "Innovative" was repeated for 9.1% of 44 brands, "smarter" for 1.9% of 104 and "proven" for none of 45.

The newest positioning carried over least. 190 brands put "agentic" in their headline block and AI repeated it for 11.6%. Notion, Asana, Salesforce, Slack, Atlassian, Cloudflare and Webflow all lead with agents on their homepages; their AI descriptions use words like "flexible", "robust" and "highly customizable".

Among the 216 brands with at least 3,000 brand sentences, the closest match was Grafana Mimir with a match score of 34.8. Its page calls it "open source, horizontally scalable, highly available", and AI uses those same words. Grafana Loki (28.5), ICEYE (23.3), Maxio (21.8) and Sociallyin (21.3) followed.

17 of the 216 brands scored 0: the words AI used for them share no weighted term with their headline block. They include Square, Jira Software, Amazon, Atlassian and Rippling. Square's homepage says "Power your entire business"; AI calls it "native", "simple" and "excellent". Brands whose headline block is a slogan scored lowest; brands whose headline block lists product attributes scored highest.

Across four AI engines, Notion's homepage says "Where teams and agents think together"; AI calls Notion "flexible", "lightweight" and "highly customizable", and 99.7% of its descriptor uses are words the homepage does not use.

Pipedrive (13.7) and Duolingo (19.5) matched best of these brands. Pipedrive's page says "The easy and effective CRM" and AI calls it "simple" and "intuitive". Datadog's AI description includes "expensive" as its second most common word.

We compared markets with at least 20 brands in the study. Webinar and Video Conferencing Platforms had the lowest median match score, 5.0, followed by Utility GIS and Vegetation Management Software (6.1) and Log Management & Observability Platforms (6.6).

Online Therapy & Psychiatry Platforms had the closest match between AI’s descriptions and homepage copy, 16.1, followed by Developer Backend & Vector Search Services at 14.6. In developer backends AI also added the fewest words: a median of 65.6% of AI descriptor uses were not on the homepage, against 74% to 90% in the other 12 markets.

On the 3,376 brands with at least 100 brand sentences on both engines, 81.3% of Google AI Mode's descriptor uses were not on the homepage, against 86.3% for ChatGPT Search. Google AI Mode repeated 8.2% of the descriptors each homepage uses, ChatGPT Search 5.5%.

The headline block tells a different story. ChatGPT Search's words matched the headline block slightly better, with a median match score of 9.1 against 7.6. One reading is that Google AI Mode picks up more words from the body of the page, while ChatGPT Search stays closer to the page's own summary. Both engines rank brands' match scores in a similar order (rank correlation 0.71).

The first-named share is the share of a brand's AI answers that name it before any other brand, on ChatGPT Search and Google AI Mode from May 24, 2026. Within each market, we split brands into thirds by match score. Brands in the highest third were named first in 16.8% of their answers, against 18.9% in the lowest third. The within-market rank correlation was -0.01.

Brands whose homepage words AI repeated most were named first more often: 20.3% in the top third against 13.1% in the bottom third. Most of that gap comes from answer volume, because brands that appear in more answers get more of their words repeated. With answer volume held fixed, the correlation fell from 0.22 to 0.10. Matching your homepage wording is not a lever for the first position.

We started with the 6,749 brands with at least 200 brand sentences in AI answers. 6,069 (89.9%) had a homepage that AI answers had cited, matched on the brand's primary URL with or without www, a trailing slash or tracking parameters. For brands whose primary URL is a product page, such as Grafana Mimir, that page stands in for the homepage. We excluded 1,285 brands with no crawled page text, 23 non-English or blocked pages, and 178 pages whose headline block had fewer than 8 words besides the brand name. That leaves 4,583 brands, 67.9% of those eligible.

The headline block comes from the crawled page: the page title, the meta description (present for 3,068 brands), the first h1 heading (4,339) and the first paragraph (4,562). Descriptors were matched against a shared list of 8,422 words and phrases AI applied to at least 20 brands, using the same matching on both sides. Descriptors cover all four engines, October 2025 to October 2026, while the homepage copies are mostly from October 2026 (94.1% crawled after May 24, 2026); a homepage that changed during the year can look further from older answers than it was. We computed the headline share two ways, with a token matcher and an independent text-search matcher, on the same 600 brands: 82.0% and 82.1%.

Do not expect AI to repeat your positioning. 83% of the words AI uses for a brand are not on its homepage, and slogans and new category terms like "agentic" carried over least. Concrete attributes carried over most: "open source", "real time" and "AI-powered" were repeated for 71% to 90% of the brands that state them.

Put the attributes you want repeated in plain words in your title, meta description, main heading and first paragraph. Then read what AI adds that you never say, especially "expensive", "complex" or "steep learning curve". Those words come from somewhere other than your site, and they are the ones to answer with pricing, onboarding and comparison pages.