AI Slop Index

Nearly two in five new English web pages are now written by AI.

That is 37.6% (95% range 35.1% to 40.2%) of new English web pages in the latest archive snapshot, Oct. 2026.

This page is a short summary of the Slop Index, a study run by DropSlop. DropSlop makes the AI-writing checker at dropslop.ai, so this is the company’s own study, summarised on its own page. Across everything published since ChatGPT was released on Nov. 30, 2022, the figure is about one in four: 25.7% (95% range 24.8% to 26.7%).

Read the full study

The share, snapshot by snapshot

The first snapshot made wholly after ChatGPT, June 2023, found 5.6%. The latest, Oct. 2026, found 37.6%. The share passed about one in ten by Oct. 2023 and about one in five by Oct. 2024, and has stayed above both lines since. The 16 snapshots from before ChatGPT run from -2.2% to 2.2% around zero. That spread is the noise in this measure.

Share of new English web pages written by AI, by archive snapshotLine chart of the estimated share of new English web pages written by AI, by archive snapshot, from Jan. 2019 (0.0%) to Oct. 2026 (37.6%).-10%0%10%20%30%40%50%ChatGPT releasedJan. 2019Oct. 2026
Each point is one archive snapshot of the web. The shaded band is the 95% range. The dashed line is the release of ChatGPT.
Show the numbers as a table
Estimated share written by AI, by archive snapshot
SnapshotShare95% rangePages scored
Jan. 20190.0%-2.1% to 0.5%1,995
April 20190.0%-2.4% to 0.0%1,946
July 20190.0%-2.9% to -0.5%2,023
Oct. 20190.0%-3.4% to -1.1%1,995
Jan. 20200.0%-3.1% to -0.7%1,945
June 20200.0%-2.0% to 0.4%1,985
July 20200.0%-2.6% to -0.3%1,975
Nov. 20200.0%-1.5% to 1.2%2,002
Jan. 20210.9%-0.5% to 2.3%1,930
April 20211.0%-0.4% to 2.4%2,008
Aug. 20211.6%0.2% to 3.0%1,994
Oct. 20210.8%-0.5% to 2.3%1,982
Jan. 20220.4%-1.0% to 1.7%1,970
May 20222.2%0.7% to 3.8%1,893
Aug. 20221.9%0.4% to 3.3%2,046
Dec. 20221.3%-0.1% to 2.7%2,007
Feb. 20232.9%1.3% to 4.5%1,951
June 20235.6%3.9% to 7.4%2,012
Oct. 202313.0%11.1% to 14.9%1,989
Dec. 202313.7%11.7% to 15.7%1,966
March 202416.1%13.9% to 18.2%1,877
April 202416.1%14.1% to 18.2%1,973
July 202419.7%17.5% to 22.2%1,974
Oct. 202422.1%19.7% to 24.4%2,021
Jan. 202524.1%21.9% to 26.7%2,008
Feb. 202522.6%20.1% to 24.9%1,973
March 202522.5%20.2% to 24.8%1,931
April 202522.6%20.1% to 25.0%1,976
May 202523.3%20.9% to 25.7%1,993
June 202525.2%23.0% to 27.6%2,066
July 202526.1%23.7% to 28.5%1,994
Aug. 202524.8%22.4% to 27.3%2,007
Sept. 202526.9%24.3% to 29.4%2,010
Oct. 202527.1%24.6% to 29.6%1,970
Nov. 202527.2%24.8% to 29.5%2,099
Dec. 202530.4%28.0% to 33.0%2,097
Jan. 202631.6%29.0% to 34.4%2,097
Feb. 202630.3%27.9% to 32.9%2,139
March 202632.3%29.7% to 34.9%2,149
April 202634.6%32.1% to 37.3%2,217
May 202633.8%31.3% to 36.3%2,136
June 202635.6%33.1% to 38.3%2,174
July 202636.2%33.5% to 38.9%2,191
Aug. 202637.8%35.3% to 40.6%2,225
Sept. 202636.9%34.4% to 39.5%2,217
Oct. 202637.6%35.1% to 40.2%2,217

Which kinds of page

The share of new pages written by AI since ChatGPT, by the kind of page each says it is. The highest share is on the page type “Blog post” (46.4%) and the lowest on “Listing or category page” (14.3%). Each kind is compared with its own pages from before ChatGPT.

  1. Blog post46.4% (95% range 43.3% to 49.5%)
  2. Article42.4% (95% range 40.2% to 44.7%)
  3. General web page34.4% (95% range 32.8% to 36.1%)
  4. News article20.4% (95% range 18.5% to 22.4%)
  5. No declared page type17.9% (95% range 17.1% to 18.9%)
  6. Shop page17.8% (95% range 15.4% to 19.8%)
  7. Other declared type15.9% (95% range 13.9% to 18.0%)
  8. Listing or category page14.3% (95% range 11.9% to 16.7%)

How it was measured

The study covers new English web pages of 150 words or more, archived at 46 crawl points from 2019 to 2026. We scored 61,566 pages published since ChatGPT and 31,779 published before it with DropSlop’s detector. On pages from before ChatGPT, when chatbots were not yet in wide use, the detector flagged about one in fourteen. On AI-written pages made for the study it caught about 19 in 20. “Written by AI” takes out the first figure and puts back the misses.

The result is an estimate. It is not a count.

Limits

The data

The files behind this page are published by DropSlop. They hold counts, shares and ranges, not the texts that were scored.

DropSlop asks that findings are used with a link back to the study. Read the full study, with the method and every figure.

Check a paragraph or a whole site on DropSlop