Mentioned

Mentioned Research · August 19, 2026 · July study follow-up

Reddit citations in ChatGPT dropped 93%. Google and the API didn’t move.

In our July study, ChatGPT cited Reddit in 6.0% of 1,273 buyer queries. We re-ran the exact same queries on August 19 through three paths at once. The real ChatGPT interface now cites Reddit in 0.4% of answers — a 15× collapse. The same queries through OpenAI’s search API still cite Reddit at the July rate, and Google still ranks the same threads. Whatever changed in mid-August lives in ChatGPT’s citation layer, not in the model or the index.

6.0% → 0.4%

of ChatGPT answers cited Reddit, same 1,273 queries, ~3 weeks apart

0 / 637

informational queries got a Reddit citation from ChatGPT on August 19

5.3%

the same queries through OpenAI’s search API — unchanged from 6.0%

66.2%

of the same searches still show Reddit in Google’s top 30 (69.3% before)

Same queries, three surfaces, one mover

Each bar pair is the same 1,273-query panel measured in the July study and again on August 19. For the two ChatGPT rows the value is the share of answers containing at least one reddit.com citation; for Google it is the share of searches with a Reddit result in the top 30. If the model or Reddit’s standing in search had changed, all three would move. One moved.

ChatGPT, the product

Collected from the real ChatGPT Search interface. 5 of 1,273 answers cited Reddit — and one of those five queries literally contains the word “Reddit”.

ChatGPT, the API

gpt-5-search-api, the same measurement the July study used: 5.3% vs 6.0%. Anyone monitoring “AI visibility” through the API cannot see this change at all.

Google, the control

Fresh depth-30 SERP pull for all 1,300 keywords: Reddit on 66.2% of pages, top 3 on 48.4% — within two points of July. The threads didn’t lose standing.

Informational queries went to zero

The collapse is not uniform noise — it is near-total everywhere. Of 637 informational queries, not a single answer cited Reddit. The five survivors are edge cases: a query naming Reddit, hyper-local recommendations, and product-review lookups.

A week of live buyer prompts says the same thing

Independently of the query panel, Mentioned collects every AI answer for the prompts our customers track — 453 ChatGPT runs across 368 real buyer prompts between August 13 and 19. Reddit took 7 of the 2,089 citations ChatGPT emitted that week (0.34%), with zero on five of the seven days. The telling detail from the raw payloads: ChatGPT still retrieved Reddit threads 35 times while answering. It reads Reddit and then cites something else.

DayChatGPT runsCitationsRedditReddit share
Aug 13202700.00%
Aug 148438100.00%
Aug 152741,39260.43%
Aug 16104312.33%
Aug 17197400.00%
Aug 18259300.00%
Aug 19217900.00%

The same prompts, the same days, other engines: Perplexity cited Reddit in 7.4% of its citations, Google AI Mode 3.7%, Gemini 2.2%. ChatGPT sits 10–20× below every other engine on identical prompts.

Method and limits

The query panel is the same 1,273 unique US-English buyer queries (26 industries, commercial / informational / transactional intents) used in the July 2026 study. The July ChatGPT baseline was collected July 31 – August 1 via gpt-5-search-api with web search. On August 19 we measured three paths: the ChatGPT Search interface (via DataForSEO’s llm_scraper, US location — the same transport our product uses for daily monitoring), gpt-5-search-api again (identical to July), and Google organic results at depth 30 (DataForSEO, US). “Cited Reddit” means a reddit.com URL appeared in the answer’s structured source list; retrieval without citation is counted separately.

Limits: the panel is one location, one language, one day per wave. The product-run series (Aug 13–19) covers six customer workspaces with B2B-skewed prompts and begins the day our collection went live, so it cannot show its own before/after — the panel rerun carries that comparison. External daily trackers date the product-side break to August 14; our data is consistent with that date but does not pin it independently. Consumer-surface scraping reflects what a logged-out US user sees and can differ under personalization. Correlation across surfaces isolates where the change happened; it does not reveal why OpenAI changed source selection.

What this means for Reddit visibility work

The channel that collapsed was never the reliable one — our July study already found no relationship between Google rank and ChatGPT citations. The channels that still pay are unchanged: half of buyer searches put a Reddit thread in Google’s top 3, and Google’s AI surfaces and Perplexity keep citing those threads. Mentioned measures presence per engine, so a shift like this shows up in your scoreboard instead of your assumptions.

See your share of voice