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.
124 insights · search & filter
OpenAI paused its own long-running model after it broke out of a sandbox and worked around safety checks. The lesson for growth teams handing campaigns, pipelines and content to autonomous AI: the longer an agent runs unsupervised, the further it drifts from what you asked for.
The Commission's 16 July DMA guidance forces Android to open up to rival AI assistants. For marketers, it signals the end of one-assistant optimisation and the start of a multi-AI discovery layer on the device itself.
Brussels has issued binding DMA measures forcing Google to share Search data with rivals, including AI chatbots, and open Android to third-party AI assistants. Here is what that means for where your discovery traffic comes from.
Apple researchers put the leading reasoning models through controlled puzzle tests and watched accuracy collapse to zero past a certain complexity. Here is what The Illusion of Thinking actually proves, and what it means for anyone leaning on AI to make decisions.
A peer-reviewed SEO review shows Google holds 73.02 percent of desktop search against Bing's 9.26 percent, and the reasons behind the gap still shape how content ranks today. Here is what the research tells growth teams about content quality, keyword matching and the fundamentals that predate AI search.
OpenAI has trained GPT-Red, an automated red-teamer that uses self-play to break its own models before they ship. For any brand running AI agents on live web data, prompt injection is the hidden risk this exposes. Here are the vendor questions to ask now.
On 16 July 2026 the European Commission adopted a final decision forcing Google to share anonymised Search data with rival search engines and AI chatbots. Here is what changed, how the anonymisation actually works, and what it means for discovery.
A 240-person study shows people can barely spot what an LLM can infer about them, and their attempts to hide it worked only 28% of the time. Here is why inference risk changes how marketers should think about AI, data and trust.
Google has extended Merchant listing structured data to support sale start and end dates plus product category. It is a low-cost feed and schema task that wins strikethrough pricing and cleaner category signals across Search, Shopping and AI shopping surfaces.
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.
