tonywangcn's picture
Upload README.md with huggingface_hub
e16f606 verified
|
Raw
History Blame Contribute Delete
3.11 kB
metadata
license: cc-by-4.0
pretty_name: The State of AI Browser Extensions 2026  Open Dataset
tags:
  - chrome-web-store
  - ai
  - browser-extensions

The State of AI Browser Extensions 2026 — Open Dataset

Aggregate data behind the study The State of AI Browser Extensions 2026: The Labs' Own Extensions Are the Worst-Rated.

A July 2026 crawl of 39,241 AI-labeled Chrome Web Store extensions, measuring install counts, star ratings, host permissions, and declared data collection — with a first-party AI-lab vs third-party aggregator comparison.

Headline findings

  • The AI labs' own first-party extensions are the worst-rated in the category. Anthropic's Claude extension is the 2nd most-installed AI extension (11M) but carries a 2.71-star rating — the lowest of any AI extension over ~1M users.
  • Zero distribution overlap between the two groups: every first-party lab extension (Claude 2.71, ChatGPT 3.33 / 3.74, Perplexity 3.65–3.83) rates below every major third-party aggregator (Sider 4.92, Monica 4.89, Merlin 4.79, MaxAI 4.75, HARPA 4.69) — a 1.36-star gap between group averages.
  • Only three major labs ship a verified first-party browser extension: OpenAI, Anthropic, and Perplexity. Google (Gemini), Microsoft (Copilot), xAI (Grok), Mistral, and DeepSeek ship none.
  • Popularity buys permissions: broad host access rises from 28% of all AI extensions to 63% of the 100k+ tier.

Files

File Description
data/summary.json Universe sizing, rating-contrast rollups, privacy counts, and the list of labs with no first-party extension.
data/top-extensions.csv The most-installed AI extensions with users, rating, rating count, and permission flags.
data/first-party-vs-third-party.csv First-party AI-lab extensions vs third-party AI aggregators, with users and ratings.

Method & caveats

Data was pulled through Crawlora's Chrome Web Store dataset endpoints. User counts are Chrome's bucketed install values (e.g. "11,000,000"), not exact. rating is the store's mean star score; rating_count is the number of reviews and is far smaller than installs (e.g. Claude: 1,306 ratings against 11M installs), so lab scores reflect a small, self-selected sample. The q="AI" universe is a keyword match that includes writing, PDF, translation, and scraping tools; the clean apples-to-apples comparison is the lab-vs-aggregator cut. First-party publishers were verified by developer email domain and known extension IDs; impostor/spoofed developer names were excluded.

This dataset is aggregate-only — no individual reviews, reviewer names, or review text.

License

Creative Commons Attribution 4.0 International (CC BY 4.0). Attribute to Crawlora and link to the study.

Citation

See CITATION.cff.