When we talk about AI in Google Search, there are a few layers worth separating. The broad AI most people know is the machine learning Google has used for years. That’s the tech working behind the scenes to decide which results show up and in what order, what we often refer to as ‘the algorithm’.

Then came AI Overviews: the big summary boxes that started rolling out in 2024 in the UK – where Google does the writing for you instead of just showing links. There was a lot of conversation in the SEO world around how Google had gone from aggregation to contextual interpretation. Now, as of Summer 2025, there’s AI Mode: a toggle in the top left of browsers. It feels like Google and ChatGPT had a baby, except of course Google wouldn’t use GPT it would use its own AI, called Gemini.

On the face of it, AI Mode looks like progress. You don’t just get Google’s version of the truth, you get links too. The reality’s … less clean. Some of those “sources” are credible publishers, but it’s been found that some are directories or paid listicles that only admit they’re ads in the small print. It raises the obvious question: how much of what’s shown in AI Mode is genuine, and how much is pay-to-play with better branding?

There’s also a second headache for those who care about website traffic and the attributions associated with it. And the pain stems directly from frustration that Google Search Console doesn’t split out clicks from AI Overviews or AI Mode, so we can’t see the direct impact. What we do see however, is impressions staying steady while clicks slip. (Frequently described as ‘the crocodile’ with the double line graph showing the separation of the numbers)

Google insists overall organic traffic hasn’t changed much year on year … Plenty of site owners and digital marketers looking at their own analytics would disagree.

The final cherry on top is the question of accuracy. AI Mode has already been caught serving outdated or flat-out wrong information, which says a lot about how stale these summaries can be. Small slip-ups or teething problems they may be, but it’s enough to chip away at trust from people who know better, and instil a sense of concern for the users who don’t.

So where does that leave us? At best, with a tool that still blurs the lines between organic and paid, while giving us no real data on impact. For all the noise, there’s no evidence of a magic solution here – just another layer we need to watch closely.