What words AI uses to describe brands

When an AI answer describes a brand, it almost always says something nice. Across every adjective and short positioning phrase AI attached to 67,013 brands in a single 30-day window, 85.3% were positive and only 4.7% were negative.

Being spoken well of by AI is not a competitive advantage, because nearly every brand is. The advantage is narrower and quieter, and it has nothing to do with how flattering the average adjective is.

The window produced 719,860 descriptor mentions across 67,013 brands, drawn from 277,963 distinct words and phrases. That is a large enough vocabulary to read a position from, if a brand has one.

But the volume hides a sameness problem. The overwhelming majority of those mentions are approval, repeated across thousands of brands that AI cannot otherwise tell apart.

In a full month of monitoring, just 12.6% of brands collected even one negative descriptor. Criticism from AI is rare enough that its absence tells a buyer nothing. The model defaults to approval, so approval carries almost no signal.

The words AI reaches for most are not descriptions, they are ratings. The single most common descriptor across all brands was recommended, applied to 5,671 different brands, followed by excellent (4,223 brands) and best (4,119).

These are verdicts. They tell a buyer that the model approves, and nothing about what the product is or who it is for.

The praise is common but spread thin. The top 10 descriptors account for only 7.6% of all descriptor mentions, the top 50 for 18.0%, and the top 100 for 23.5%. There is no handful of words that defines how AI talks about brands. The vocabulary is wide and shallow.

The descriptors that actually describe a brand do not sit at the top of the list, they sit in the tail. Of the 277,963 distinct descriptors AI used, 85.4% (237,411 of them) attached to exactly one brand.

The vocabulary of real description, the specific and unusual phrasings, is almost entirely single-brand. That is where the competition actually is.

Most brands are barely characterized at all. The median brand collected just three distinct descriptors in the window, with a mean of 8.8 and a 90th percentile of 15. AI has an opinion about thousands of brands, but it has a vocabulary for very few of them.

Take the share of a descriptor's total mentions held by its single most-associated brand. For generic praise, that share is tiny, because the word is spread across thousands of brands.

Best sits at 1.0%, popular at 1.0%, recommended at 1.3%, good at 1.3%, and excellent at 1.4%. No brand owns best. Being called best puts you in a crowd of four thousand.

For specific attributes the top brand holds a real chunk of the word. Fast sits at 11.0%, held by Linear. Customizable is 6.1% (Monday.com), lightweight 4.8% (Notion), open-source 4.8% (SigNoz), enterprise-grade 2.6%, and affordable 2.0%.

A specific word is 5 to 11 times more concentrated than a generic one. That concentration is the asset. It is the difference between being one of four thousand brands AI calls best and being the brand AI calls fast.

The line between a word you can own and one you cannot is the line between an attribute and a verdict. The most owned generic word, excellent at 1.4%, is still less concentrated than the least owned attribute on this list. A buyer who hears excellent learns nothing distinctive. A buyer who hears fast learns who you are.

You can see the divide in the descriptors well-known brands actually own. Linear's most-used word was fast, Notion's was flexible, ClickUp's was highly customizable, Stripe's was developer-friendly, Snowflake's was scalable, Mailchimp's was user-friendly, and Zapier's was simple. Each of these is a position, not a rating.

But plenty of equally prominent brands top out at praise. Asana, HubSpot, Slack, Datadog, and Vercel all had excellent as their single most-used descriptor, and Shopify's was best. Those companies are visible to AI, but AI has not yet attached a distinct idea to them.

Ignore the baseline approval. Positive sentiment from AI is the norm, shared by 85% of every brand's descriptors, so a positive-sentiment number is a constant, not your position.

Track which specific descriptors attach to you and how exclusively, not how nice the average adjective is. The question is not whether AI likes you, it is whether AI has assigned you a word it has not handed to your competitors.

To plant a specific word, the sources AI cites for your category need to say it about you, repeatedly and specifically, so the model has something concrete to repeat instead of falling back on excellent.

We read every adjective and short positioning phrase AI attached to a brand across ChatGPT Search and Google AI Mode, over the 30 days ending June 26, 2026, and labeled each one positive, neutral, or negative. That produced 719,860 descriptor mentions across 67,013 brands and 277,963 distinct words and phrases.

How concentrated the vocabulary is means the running share of all mentions held by the most common descriptors. Ownership of a single word is the share of that word's total mentions held by its most-associated brand. Figures describe one observed 30-day sample from the two engines in scope, not Perplexity, Gemini, or Copilot, and are point-in-time rather than a trend claim.