Embed our data
Embed GPTfake’s AI censorship measurements anywhere — a GitHub README, a blog post, or a dashboard — with a copy-paste refusal-rate badge per model. According to GPTfake monitoring (as of 2026-06-15, n = 500), each badge renders that model’s current overall refusal rate as a small SVG that links back to the AI Censorship Leaderboard. The numbers come from the same data model as the leaderboard table, so a badge can never drift from the published figure. Figures are illustrative until live monitoring data lands.
How the badges work
Each badge is served from /badge/<model>.svg and shows, for example, “ChatGPT refusal rate: 18.7% — GPTfake”. It is a plain SVG image:
- Self-updating — when our published refusal rate changes, every embedded badge updates with it. There is nothing to re-copy.
- Backlink-friendly — an
<img>can’t carry a link, so the copy-paste snippets below wrap the badge in an<a href>pointing to/leaderboard. Keep that link — it is the attribution. - Static & cached — badges are pre-rendered at build time; no JavaScript, no tracking.
Available model keys: chatgpt, claude, gemini, mistral, qwen.
Badges
Copy the HTML (or Markdown for a GitHub README) under any model. The leaderboard link is already included.
ChatGPT (GPT-4o)
Claude (Sonnet)
Gemini
Mistral (Large)
Qwen
Attribution: “Powered by GPTfake data”
Our data is free to reuse under CC BY 4.0 — including commercially — on one condition: attribution. When you embed a badge, a chart, or our numbers:
- Keep the link back to
https://gptfake.com/leaderboard(the badge snippets already include it). - Where a badge isn’t used, add a visible “Powered by GPTfake data” line linking to the leaderboard.
- Keep the as-of date visible so readers know the snapshot the figure belongs to.
This is what turns reuse into a citation: every embed is a dated, linked pointer back to the source.
How to cite
Cite the leaderboard (the canonical, dated object) in articles and papers.
Plain text:
GPTfake (2026). AI Censorship Leaderboard.
Independent AI Censorship Watchdog. As of 2026-06-15, n = 500.
https://gptfake.com/leaderboard
Accessed: 2026-06-16.BibTeX:
@misc{gptfake_leaderboard_2026,
title = {AI Censorship Leaderboard --- Refusal Rate and Bias by Model},
author = {{GPTfake}},
year = {2026},
howpublished = {\url{https://gptfake.com/leaderboard}},
publisher = {GPTfake (Independent AI Censorship Watchdog)},
note = {As of 2026-06-15, n = 500. Accessed 2026-06-16}
}To cite the underlying figures directly, use the open refusal-rate dataset (CC BY 4.0), which carries its own Dataset JSON-LD and BibTeX entry. For the full schema and license, see the datasets hub; for how the numbers are produced, see the monitoring methodology.
Illustrative data. Badge values are labeled placeholders pending the live monitoring pipeline. Real figures will carry the same methodology link, sample size, and as-of date.
GPTfake