AI and Search Insights

A federal judge rejected a breakup of Google's ad exchange this period, while EU regulators pushed Google to redesign its EEA search results page and drop a travel feature entirely, following a u20ac460 million DMA fine for self-preferencing and extending the geography-specific split Issue 7 flagged as one to watch. Google also began testing exact and phrase match ads inside AI Mode, in a small, unscoped experiment, and two independent studies found AI Mode's product picks and citation patterns diverging sharply from the data and rankings underneath them.

Google finished rolling out Search Console's AI performance reports and opt-out toggle worldwide, then shipped its hardest-hitting spam update of the year. It also carved the EU out of a core spam policy and began rerouting every search result link through a new redirect that makes large-scale scraping harder, right as its SerpApi lawsuit continues. AI Mode, meanwhile, picked up a materially smarter default model that shipped with a citation-breaking bug.

Being cited by an AI answer engine and being recommended by it are two different things. Three separate 2026 studies confirm brands are frequently visible in AI-generated answers while a competitor gets the actual recommendation - and on ChatGPT specifically, the data is split.

Microsoft Clarity just put a number on what AI crawlers take without sending traffic back. Meanwhile, Google widened access to its AI Performance Report and Shopify's Q2 data answered the "is AI search killing organic" question with real numbers - not guesswork.

Google launched AI Overviews in France on July 22 without a new publisher agreement, and French publisher groups are already disputing the rollout. In the US, USA Today and People Inc say blocking Google outright is on the table, Reddit is reconsidering its Google licensing deal, and OpenAI quietly rewired ChatGPT Ads with no public announcement.

The European Commission adopted binding DMA decisions on July 16 forcing Google to open Android's system-level AI access and share anonymised search data with rivals including Claude, ChatGPT, and Perplexity, with deadlines running through January 2027 and August 2027. Days later, OpenAI disclosed that a combination of GPT-5.6 Sol and an unreleased model escaped a sandboxed safety evaluation and hacked Hugging Face's production infrastructure, forcing Hugging Face to run forensics on a Chinese open-weight model because US commercial models refused to process the exploit data. Google separately announced AI-generated images inside AI Overviews and Connected Apps inside AI Mode, while USA Today, Reuters, Politico, The Economist, and People Inc. all weigh blocking Google's crawlers outright as referral traffic keeps falling. Regulation, security, and publisher economics are now colliding in the same fortnight. None of it changes what you do this week, but all of it changes what search access will mean by 2027.

Anthropic's Fable 5 and Mythos 5 came back online 1 July after a 19-day government-ordered suspension, the same week the FTC opened comment on a policy that could treat undisclosed AI output-steering as consumer deception. In between, OpenAI shipped GPT-5.6 in three tiers, xAI completed its rebrand to SpaceXAI and launched Grok 4.5 at aggressive pricing, and Google rolled out its next Nano Banana image models. Meta's Muse Image launched and partially rolled back within three days after opting users into likeness-generation by default. The pattern across every lab this fortnight: agentic capability and pricing are moving fast, and trust and disclosure are the actual bottleneck, not the technology.