Jump to content

Data Platform/Data Lake/Data Issues/2026-06-10 Nov 2025 spike in bot traffic

From Wikitech

Nov 2025 spike in bot traffic

Status Resolved
Severity High
Business data steward Kate Zimmerman
Technical data steward Andreas Hoelzl
Incident coordinator Guilherme Gonçalves
Incident response team Marcel Ruiz Forns, Hamid Ghani, Omari Sefu, Andreas Hoelzl
Date detected December 2025
Date resolved April 4, 2026: deployment of new bot detection heuristics

May 15, 2026: backfill completed

Start of issue November 2025 (large increases observed)
Phabricator ticket T413027 (December 2025 detection by Movement Insights)

T413949 (January 2026 detection by community members)

T421735 (Backfill plan)

Summary

As Wikipedia traffic comes in, we classify it as coming from either humans (“users”) or bots (“spider” or “automated”). Some bots attempt to bypass rate limits or other restrictions with increasing sophistication.

In December 2025, we observed an unusually large increase in our measurement of user pageviews for November 2025 (T413027). After investigation, it became clear that this increase was due to non-human (bot) traffic. The bot activity was predominately geolocated to 4 countries (Vietnam, United States, Philippines, and India) and employed spoofed user agents that appeared as desktop users in our metrics.

At its peak, the bot activity accounted for ~10% of global pageviews, with the most visible impact to smaller language wikis. As of February 2026, the bot activity had decreased to ~3% of pageviews. The bot activity continued to distort our metrics until April 2026, when we deployed new bot detection heuristics.

While we have been able to isolate the anomalous traffic and build controls to classify it as automated for the impacted period and going forward, some of the raw traffic data necessary for a full repair had already been deleted as part of our Privacy Policy. Due to this, we were not able to make corrections to the November 2025 data.

Impact on metrics

All Wikipedias
Metric Month Pre-correction % of automated traffic (now corrected)
User pageviews Dec/2025 15.13B 9.07%
Jan/2026 16,09B 10.67%
Feb/2026 13.50B 3.10%
Mar/2026 (first 10 days) 4,76B 2.86%
Unique devices Dec/2025 1.79B 6.21%
Jan/2026 1,79B 5.69%
Feb/2026 1,79B 4.25%
Mar/2026 1,88B 5.85%

Note: Post-correction metrics better reflect user behavior. Previous metrics were skewed by non-human activity.

Findings

Signals indicating bot traffic

Large amounts of traffic that diverged from standard browsing behavior.

Root cause

A new pattern for bot traffic emerged that was not being detected by our existing heuristics.

Affected datasets and services

The full list of datasets and services is available in T421735.

  • actor datasets (webrequest & pageview)
  • browser_general_daily
  • browser_metrics_weekly
  • clickstream_monthly
  • interlanguage_daily
  • referrer_daily
  • pageviews datasets (hourly & daily; per project, per article, top articles, top articles per country)
  • projectview_hourly
  • projectview_geo
  • unique_devices datasets (hourly & daily; per domain & per project family)

Resolution & decisions

In addition to correcting the data going forward, we decided to backfill historical data with corrections where possible.

New heuristic: We developed a new heuristic to identify bot traffic based on the signals we identified in our investigation. This heuristic was deployed on April 4, 2026.

Backfill to correct historical data: The backfill of publicly-available pageview and unique device data going back to December 1, 2025 started on April 28, 2026 and completed on May 15, 2026.