Context Rot
Context rot is the drop in an AI tool's accuracy as the text it's given to work from gets longer. In a July 2025 study, Chroma, a company that makes database software for AI apps, tested 18 AI models on simple tasks and found their results slipped as the input grew, often in uneven and surprising ways.
Source: Chroma, Context Rot: How Increasing Input Tokens Impacts LLM Performance (Hong, Troynikov and Huber, July 2025), checked October 29, 2026
What Context Rot means for your business
In Chroma's tests, the further the answer's wording sat from the question's wording, the faster results fell as the text grew. And one near-miss passage, on topic but wrong, cut accuracy on its own, while four of them cut it further.
Those were lab tests, not ChatGPT or Google answering a buyer. Even so, we'd answer in the words your buyers use, not your trade's shorthand, and we'd clear out the almost-right versions, like an old price list sitting next to the new one, so there's only one answer to find.
If your site has two or three pages that almost answer the same question, trimming them down to one clear answer is the cleanup we'd do first.
Last reviewed October 29, 2026. All terms
How AI answers it today
On October 29, 2026, Google's AI Overview cited Chroma's study next to vendor blogs, LinkedIn and YouTube Shorts, framing it around coding tools and AI agents. ChatGPT used a long coding chat as its example and cited nothing. Neither one connected it to web search.
Our run through ChatGPT and Google's AI Overview, October 29, 2026. A sample, not what every buyer sees.
Related terms in answers and how content gets used
Find out where your brand stands.
Tell us what you're trying to figure out. We'll get back to you within one business day with a time to talk.