Insights · Analysis
Analysis of how ChatGPT, Claude, Gemini, Perplexity, Copilot, Grok, DeepSeek, Meta AI, Google AI Overviews, and AI Mode are rewriting discovery, and what to do about it.
135 insights · search & filter
75.5% of every AI model ever announced is already dead, and the median model on sale today is 60 days old. What 261 releases mean for anyone buying one.
ChatGPT's share of assistant traffic fell from 76.4% to 52.7% in eleven months. Claude grew 5.8-fold. What the numbers mean if you are choosing between them.
68.0% of US Google searches now end without a click. AI Overviews cut click-through by 46.7%, and only 1% of people click a source inside the summary.
The open web now gets 0.57 clicks per search across a 12-query basket, and 8 of the 12 return under one. What answer engine optimization is really working against.
Queries an assistant can finish fell 64.2% from peak. Everything else fell 4.5%. A 14.2x gap, and the one question that sorts your keywords into the right pile.
Reddit is cited 149x more often than Stack Overflow. Wikipedia gets 8.11 citations per page, YouTube 1.87. What generative engine optimization actually rewards.
Only 11.3% of the times G2 gets quoted by an AI assistant come from ChatGPT. Six assistants quote wildly different sources, so tracking one gives you the wrong number.
Open weights hit 60% of model releases in 2024 and have fallen every year since, to 38.3%. And 85.9% of Chinese releases are open against 34.0% of everyone else's.
The median AI model on sale today holds a million tokens, 488x the 2022 figure. The search box still takes about eight words. That gap explains a lot.
Primary research · August 2026
An audit across 10 categories and 60 buying questions. I recorded what ChatGPT reads, throws away and links to when a buyer asks it which software to buy, and what that decides.
Free download
35 pages · PDF · 536 KB. Enter your details and it downloads straight away.
