Google has two AI answers. They don't agree.

I took the summer off to sort my head (and my chemo) out. I'm back. Known & Cited has changed. And this newsletter has had a bit of a makeover too. Monthly. Two-minute read. One thing worth knowing, one update from me. That's it. Short and sweet.

One thing worth knowing

Google has two AI answers. And they don't agree.

That box at the top of your search results? That's AI Overview. Ask it a follow-up and it quietly drops you into AI Mode, Google's full-blown chatbot. No app. No sign-up. No idea it's even happened.

You'd assume they're the same thing. They're not. One study ran the same 400 searches through both, on the same day. Where both gave sources, nearly a third didn't share a single one. (It's one run, in the US, by a GEO agency. So maybe take with a pinch of salt. But they showed their working and even corrected their own maths, which is more than most.)

Why does that matter? Because 51% of US adults say they never use an AI chatbot. (Pew Research, 5,119 people, out last week.)

Lovely. Panic over. Everyone can go back to Google...

Except Google is AI now. So a big chunk of the "I never use AI" crowd use it every single day. They just call it Google.

My take. "Are we in Google's AI answer?" is now two questions. You can't treat them as one. And the people seeing those answers mostly don't think they're using AI at all, which is exactly why it matters.

From me

Earlier this year I sold AI visibility scores and big juicy reports. Loads of data, charts, tiers, colours, lots and lots of pages. They looked awesome, and for the right client they were great. But I've decided to stop, because:

1. AI visibility scores are a made-up number based on a made-up formula, which is so specific to a particular brand that it's not really comparable to anyone else anyway.
2. Even if you want to track that score over time, by running exactly the same research, the chances are that the LLMs will have changed so much that getting a comparable score is unrealistic anyway.
3. There is so much data available when you do proper GEO research that a key part of delivering to the client needs to be NOT sharing it all. It's too much. It makes everyone's head hurt.
4. Generally I'm talking to PR and comms people. They (generally) don't need scores. They need maybe 3–4 points, based on tailored research, that they can use to improve what they offer to clients.

Known & Cited is just me, offering GEO and AI search for PR and comms people. No dashboards, no scores, no BS. Just working out what AI says about you, who it's quoting, and what's actually worth doing about it. (Here's how I work.)

And I'm keeping it small on purpose. One new project a month, ideally, with brands that actually want to work with me. If that's you, and you're thinking about AI search for the end of this year or next, drop me a message.

Be known. Be cited.

Russ

Sources: Underneath, AI Mode vs AI Overviews (v1.1, updated 8 Oct 2026): 400 US searches, single run on 26 September 2026; 30.5% of the 236 searches where both cited sources shared no URL. · Pew Research Center, 9 Oct 2026: 5,119 US adults, fieldwork February 2026. · Google, 27 Jan 2026.