ChatGPT's citations to Reddit fell 86% on a single day, 14 August, with no announcement from OpenAI and no explanation since. A source that was the most-cited domain on the platform in April is now a rounding error.

Promptwatch reports Reddit averaged 3.83% of all ChatGPT citations between 18 July and 7 August, then 0.52% between 14 and 17 August, a relative fall of 86.4%. The break is a single day, 14 August. The same dataset shows Google moving gradually rather than sharply over a comparable window: AI Overviews down 11.3% and AI Mode down 30.5% in Reddit citation share, neither with a step change. Promptwatch describes its collection as taken from the interfaces of ChatGPT, Gemini, Perplexity, Claude and AI Overviews, counts only responses that carried citations, and states an overall base of more than 26 billion data points without breaking out a sample for this specific measurement.

Two cautions, both important. First, Promptwatch says plainly that "a data-collection issue cannot be ruled out, so treat the size of the drop as provisional." That caveat is theirs, not mine. Second, the "query fanout change" explanation circulating in coverage (Crypto Briefing, 18 August, https://cryptobriefing.com/reddit-chatgpt-search-citations-drop/) is the author's inference from timing. It is not sourced to OpenAI, OpenAI has not commented, and no primary confirmation exists. Treat the cause as unknown and the effect as measured but provisional.

Google logged the August 2026 spam update on its ranking updates history at 09:27 Pacific on 18 August. Google Search Central's own wording: "This is a normal spam update, and it will roll out for all languages and locations. The rollout may take a few days to complete." No new policy categories were announced. For scale, the March update took roughly 19 hours and the June update took 2 days and 1 hour (Search Engine Land, 18 August, https://searchengineland.com/google-releases-august-2026-spam-update-485369).

The part that matters is older and is not in the announcement. On 15 May 2026 Google updated its documentation to state that its spam policies apply to generative AI responses in Search, covering AI Overviews and AI Mode, and applying the existing ruleset: cloaking, scaled content abuse, link spam, site reputation abuse, doorway pages, expired domain abuse, and inauthentic mentions (PPC Land, 18 August, https://ppc.land/googles-third-spam-update-of-2026-hits-every-language-and-region/, and PPC Land's May coverage, https://ppc.land/google-spam-policies-now-officially-cover-ai-overviews-and-ai-mode-in-search/). Google has said nothing about whether this particular update enforces against that behaviour.

Alex Galinos tracked 51,200 AI Overview click events across 1,661 unique cited snippets over nine months, September 2025 to June 2026. Method: a custom GA4 dimension capturing the `#:~:text=` fragment Google appends to the destination URL when a user clicks a cited snippet inside an AI Overview. Comparing those tracked events against GA4's channel attribution, an average 22.4% were attributed to Direct rather than Organic, roughly 11,468 sessions, with a monthly range of 16.8% in April 2026 to 29.3% in May 2026. In the same data AI Overviews drove 7.53% of total organic sessions across the period, peaking at 16 to 17% in February and March 2026 and sitting at 2 to 4% by the time of writing.

Sample caveat, and it is a big one: this is one brand, in transportation. Not a panel, not a survey. The author also notes the `#:~:text=` fragment is not exclusive to AI Overviews, since featured snippets and People Also Ask use it too, and that comparing event-level to session-level data is imperfect. Read it as one carefully-instrumented case, not a rate.