okaihub

What Makers of Thousands of AI Tools Have in Common

Patterns across 20,947 AI products in our directory: naming, category clustering, and the small minority that publish a changelog.

Published 2026-11-10 · 3 min read

Our directory holds 20,947 AI and creative products, each pointing at a live product website. Looking across all of them rather than at any one produces a picture of how this category presents itself — and where almost everyone makes the same choices.

They name their products the same way

61.3% of the 20,947 names are a single word — 12,833 products. Another 20.5% are two words. The average is 1.77 words.

The reason is domain availability rather than branding preference. Short invented words are close to the only remaining pool of unregistered domains, which produces thousands of names that mean nothing until you visit the site.

Name patternShare of 20,947
One word61.3% (12,833)
Two words20.5% (4,300)
Three or more words18.1% (3,795)
No countable word — symbols or numerals only0.1% (19)
Ends in ".ai" or "AI"15.3%
Contains "AI" as a standalone word20.3%
Mentions GPT, LLM or ChatGPT1.7%

That last row is the interesting one. Naming a product after a model was common a few years ago and has almost disappeared — at 1.7% it now dates a product faster than it explains it.

They describe output rather than jobs

Counting the most frequent words of three letters or more across all 20,947 names, the top entries are dominated by artefacts rather than problems:

WordOccurrencesWordOccurrences
generator387online99
image284pro98
video265gpt88
free186text81
app115maker76
studio109assistant73
your100music66

"Image generator", "video maker", "music studio" all say what comes out. Almost none say who it is for or which problem it removes. This is why so many products in this category are indistinguishable at the name level, and why the differentiation work has to happen in the first sentence of the description instead — where most products also fail to do it.

The 186 names containing "free" are a separate problem: a pricing claim embedded in the brand, which cannot be walked back later without a rename.

They cluster into a few categories

Across 30 categories, the five largest hold 11,930 products — 57% of the corpus:

CategoryProductsShare
AI Agents & Infrastructure3,21415.3%
Business, Finance & Legal2,40311.5%
AI Media Generation2,24110.7%
Developer Tools2,21810.6%
Marketing & Advertising1,8548.8%
The smallest of the 29 substantive categories holds 33 products. Category counts cover the whole corpus.

The practical consequence is that most products are competing in a category with hundreds or thousands of near-neighbours, and the categories are broad enough to contain products that do not actually compete. AI agents alone spans frameworks developers build on, infrastructure services, and finished end-user products.

They mostly do not publish a changelog

This is the most actionable pattern in the data. Of 20,946 products where we could check:

SignalProductsShare
Publishes a blog9,53745.5%
Maintains a changelog2,55112.2%
Presence only — a blog with one post from 2023 counts the same as one updated weekly.

A blog has become close to a default expectation; a changelog has not. Yet the changelog is the one that answers the question a buyer of small software actually has, which is whether anyone is still maintaining this. Publishing one puts a product ahead of roughly seven in eight of its peers, and it costs a line per release.

What the patterns suggest

  • Naming is a constraint, not a choice. If you have a one-word invented name, you are one of 12,833. The differentiator has to live in the description.
  • Describe the job, not the artefact. "Turns an hour of podcast into five captioned shorts" survives the swap test; "AI video generator" does not.
  • Do not put pricing in the name. 186 products have, and it is a decision they cannot reverse cheaply.
  • Assume your category is crowded and be specific anyway. Narrowing is the only response that works, and it costs a rewrite rather than a rebuild.
  • Publish a changelog. It is the rarest of the cheap trust signals and the one that answers the question buyers actually ask.

Across 20,947 products the patterns are remarkably uniform: a one-word name, a description of the output, a category shared with thousands of others, and no changelog. Being different on any one of those is easier than it should be.

More in Content & building in public