ChatGPT Slashes Reddit Citations by 86% | The Search Shake-Up That Happened in Just 4 Days
機械翻訳 / Machine-translated
A
AI Friends
3時間前・12分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
What You'll Learn from This Article
In August 2026, something strange happened in ChatGPT's search results. The rate at which the major American forum site Reddit was cited plummeted by 86% in just four days.
According to Promptwatch, a research firm that tracks AI search data, from July 18 through August 7, Reddit was cited in ChatGPT search results at an average rate of 3.83%. In other words, roughly 4 out of every 100 searches drew on information from Reddit.
But from August 14 onward, that figure cratered to just 0.52% — meaning Reddit was being used in fewer than 1 out of every 100 searches.
This dramatic shift came with no advance notice whatsoever from OpenAI, the company behind ChatGPT. Reddit has made no official statement either.
According to expert analysis, the change stems from a shift in how ChatGPT conducts searches.
On August 8, ChatGPT dramatically increased its use of the "site: operator" — a function that limits searches to a specific website. Usage jumped from 0.4% of all searches to 17% in one go, representing a roughly 46-fold increase.
The site: operator works like this: searching "site:nytimes.com AI" finds articles about AI exclusively within the New York Times website. As ChatGPT began relying heavily on this method, searches became concentrated on specific sites, making it harder for broad-content platforms like Reddit to get cited.
At the same time, the number of searches ChatGPT performs per response approximately doubled. In short, it is now conducting more searches but drilling deeper into narrower sources.
What makes this situation even more complex is the partnership between Reddit and OpenAI.
In May 2024, OpenAI and Reddit signed a partnership deal worth approximately $70 million per year (roughly ¥10.5 billion). Under this agreement, OpenAI gained access to Reddit's data API, enabling it to use real-time posts and discussions for AI training.
In other words, OpenAI is paying Reddit for the right to use its data — yet Reddit citations in actual search results have plummeted, creating a seemingly contradictory situation.
According to reports, Reddit began pushing in 2026 for dynamic, usage- and quality-based pricing rather than a simple flat fee. Some have suggested that this negotiation process may have led to some kind of policy change behind the scenes.
What is particularly striking is that this phenomenon appears to be specific to ChatGPT.
In Google's generative AI feature, AI Overviews, Reddit citation rates have declined gradually — from around 2.5% in early July to around 2.1% in August — but nothing like the sharp drop seen with ChatGPT.
This illustrates that each AI company determines its search results using its own algorithms and policies, and that the same website can be treated very differently depending on which AI is involved.
This incident has thrown into sharp relief the importance of a new concept: GEO, or Generative Engine Optimization.
Traditional SEO (Search Engine Optimization) aimed at ranking highly in Google search results. GEO, by contrast, is about optimizing content so that generative AI systems like ChatGPT and Gemini will cite it.
According to 2026 data, OpenAI has 900 million weekly ChatGPT users, Google has 900 million monthly Gemini app users, and Google's AI Mode has surpassed 1 billion monthly users. Exposure through generative AI has become a scale that can no longer be ignored.
Yet as this incident demonstrates, the world of GEO is extraordinarily volatile. Algorithms can change without warning, and citations can drop by 86% — as has actually happened.
There are several lessons that Japanese media and businesses should take away from this incident.
First, heavy reliance on generative AI carries significant risk. Even having a data partnership agreement with an AI company does not guarantee actual citations or exposure.
Second, spreading presence across multiple platforms is essential. Securing exposure not just on ChatGPT but also on Google Gemini, Perplexity, Microsoft Copilot, and other generative AI platforms is key to risk management.
Third, businesses need to regularly monitor how often their content is being cited by each generative AI. Search Engine Land has warned that "companies should begin tracking their own AI citation trends."
As of 2026, the adoption of AI search performance tracking tools is advancing in Japan as well. Having data on which AI is citing your articles and content — and how often — is becoming the new standard.
OpenAI has issued no official comment on this matter.
Why did it increase its use of the site: operator? Why have Reddit citations dropped? What happens next? None of these questions have been answered.
Promptwatch has also been cautious, noting that "the data shows what happened, but not why," and that the possibility of a data collection issue cannot be entirely ruled out.
That said, multiple independent analyses have confirmed the same trend, and it is widely believed that some significant change did indeed take place.
Content strategy in the generative AI era operates under entirely different rules from traditional SEO. Unpredictable, opaque, and rapidly changing — that is the reality this incident has laid bare.
This article is a cross-post from AI Friends.
