DEV's AI-disclosure labels after seven weeks: 1 post in 25, and a quarter of "AI-Assisted" was one burst

Description and use

After DEV introduced AI‑disclosure labels, a study counted all public posts and found only 3.5 % were labeled, with a 1,912‑post burst inflating AI‑Assisted rates. Labels correlate with richer metadata, not higher reactions.

Problem

DEV introduced AI-disclosure labels but no usage data existed; the study aimed to quantify label adoption and uncover anomalies.

Outcome

The study found that only ~3.5% of posts carried a label after removing a single burst, with AI‑Assisted at 2.0% and Fully Autonomous at 1.2%. The burst inflated the AI‑Assisted share to 2.6% and created a misleading weekly spike. Labelled posts had higher reaction rates, but this correlated with richer metadata rather than the label itself.

Workflow

• 1. Scraped DEV’s /articles/latest endpoint (1,000 posts/page) for 418 pages in 43.5 min.
• 2. Parsed each post’s label field from the API response.
• 3. Aggregated counts per label, per week, and per tag.
• 4. Detected a burst of 1,912 AI‑Assisted posts with identical tags and timestamps.
• 5. Re‑computed weekly shares excluding the burst to reveal true trend.
• 6. Analyzed reactions and cover‑image usage per label.
• 7. Compiled author‑level statistics on label usage.
• 8. Documented methodology and limitations in a GitHub repo.

Evidence

The snapshot covers every post in the /articles/latest listing from 12 August (two weeks before the labels) to 14 October, 19:01 UTC: 414,144 posts, 304,071 of them published after the labels launched.

Source

About 1 post in 25 carries a label

Source

The tag brussels had 1,912 posts, every one of them AI‑Assisted.

Source

Without the burst, the share of labelled posts since launch is 3.5% rather than 4.1%, and AI‑Assisted is 2.0% rather than 2.6%.

Source