UK Regulator Pushes Google to Give Publishers More Control Over AI Search

The UK's Competition and Markets Authority (CMA) is opening a consultation about new rules for Google Search, specifically asking whether Google should give publishers better control over how their content appears in AI features. The concern is straightforward: as Google adds AI-generated summaries to search results, publishers want a say in whether their articles get used for this purpose.

The CMA consultation focuses on requiring Google to let publishers opt out of AI Overviews — those AI-written summaries that now sit above traditional search results — and to prevent their content from training Google's Gemini models. The regulator's goal is to expand the choices available to both publishers and users.

Google's Response and New Controls

Google has responded positively to the CMA's concerns. The company released Google-Extended, a new tool that lets websites control whether their content can be used to train Gemini. Website owners can now use something called a robots.txt directive (a simple text file that tells search engines what to do with a site's content) to say "don't use this for Gemini training," separate from traditional search indexing.

This approach fits a pattern we have seen before in tech regulation. When the EU's data protection law, GDPR, was being rolled out in the mid-2010s, major companies introduced granular consent controls while regulators were still writing the rules. A company essentially says: "Here's a technical solution to the problem you're raising — and we're willing to deploy it."

Google's decision to introduce Google-Extended now suggests the company believes that publisher control over AI training data will become a regulatory priority not just in the UK, but globally. By building a technical system now, Google may be positioning itself to meet similar demands as they arise elsewhere.

Why the UK is Focused on This

The CMA's work on Google Search AI features sits inside a much larger investigation. In October 2023, UK regulator Ofcom flagged the supply of cloud infrastructure services — the servers and systems that power most of the internet's backbone — and referred it to the CMA for examination. The CMA released its initial findings on January 28, 2025.

This timeline matters because it shows how regulators think about AI. They are not treating AI features as isolated technologies — instead, they are looking at them as part of broader questions about which companies control the most powerful infrastructure and data. By requiring publishers to have opt-out controls for AI training, the CMA is trying to reshape who gets to decide how the internet's vast archive of content gets used.

The underlying concern is this: a small number of very large tech companies now control most of the data and computational power needed to train modern AI systems. If Google can freely use publishers' content to build AI models, and those models power search results, Google gains an advantage that competitors cannot match. Regulators want to level that playing field by giving publishers a choice.

The Technical Reality is More Complicated Than It Sounds

Here's where things get tricky. There are two different ways Google uses publisher content in AI features. AI Overviews pull content in real-time — they grab an article right now to summarize it for a search result. Model training is different — Google reads vast amounts of content once and "learns" patterns from it to build the AI system. Those patterns stay inside the model forever.

Google-Extended addresses the training question, letting publishers opt out of Gemini training data. But it does not fully solve the question of real-time content use for AI Overviews. Publishers must now juggle multiple decisions: allow traditional search indexing or not, allow content in AI summaries or not, allow training data use or not, and potentially different rules for different kinds of AI models.

This technical gap — that publishers can control training but not real-time summarization — suggests that more controls may need to be added to fully answer what the CMA is asking for.

What This Could Mean Broadly

The way the CMA is approaching this question — through existing competition and market power rules, rather than new AI-specific laws — is significant. It suggests that as AI becomes embedded in more services, regulators will treat it like any other feature that affects whether a market is genuinely competitive.

For publishers, this creates a genuine strategic problem. Appearing in AI-generated summaries can bring traffic, but it can also mean readers get the answer from Google itself and never click through to the original article. Now publishers have to make a choice about that tradeoff.

For other AI platform companies, the CMA's approach sends a message: if you build AI services at scale and have market power, expect regulators to ask how you structured publisher controls. Companies that build robust choice mechanisms early may face less regulatory friction later.

Over time, this pattern suggests AI governance will work through existing competition law frameworks rather than through new AI-specific regulation. That means tech companies should expect competition authorities to apply traditional questions about market dominance to AI-enabled features, with a focus on whether publishers and users actually have real alternatives and meaningful control.