I was on the sofa the other evening with a coffee and the iPad, half-working the way you do, when I came across two things Google had quietly shipped. Neither made much noise. Both tell you something about where search is heading, and what's actually worth doing about it.
First, the reassuring bit. There's been a lot of noise this year about "AI SEO", AEO and GEO, and a small industry has grown up around it, much of it promising to get you cited by ChatGPT or pinned to the top of Google's AI answers. In June, Google updated its own guidance and was about as plain as it gets: optimising for AI search is still SEO. No separate discipline, no secret playbook. The work that earns you rankings, clear, fast, genuinely useful pages with sensible structured data, is the same work that gets you pulled into AI answers, because those answers are built on the same index. So if someone is selling you a special "AI optimisation" package, the honest starting point is that you probably already have most of what it's selling.
That's the headline, and it's true. But two of the things I came across complicate it slightly, in a way worth understanding.
The first is in Chrome. Lighthouse, the tool a lot of us use to check the quality of a site, has quietly added a whole new category called agentic browsing audits, sitting alongside the usual performance and SEO checks. One of them looks for an llms.txt file, a short markdown summary of your site aimed at AI agents rather than people. Now, Google Search has been dismissive of llms.txt for ranking, and rightly so, it won't move you up the results. But here's the thing. Google Search says don't bother, while Google's own browser tooling has started checking for it. Two different teams, two different jobs. One cares about ranking your pages; the other cares about whether an AI agent can read and find its way around your site. The check is gentle for now, it doesn't even count against you if the file is missing, but its arrival tells you which way the wind is blowing.
The second is OKF, the Open Knowledge Format, which Google Cloud published in June. It's a tidy standard for handing curated knowledge to AI agents, plain markdown files with a bit of structure. The important thing is what it isn't. It's not an SEO trick and it won't touch your rankings. It's for the agents a business runs against its own data, not for Google's crawler. We're interested in it for our own automation work, not as something we'd ever sell you as a visibility service.
So what do you actually do with all this? Less than the noise suggests.
Keep doing the fundamentals. A fast, crawlable, genuinely useful site with honest structured data is still the thing that wins, in normal search and in AI answers. That hasn't changed, and Google has just confirmed it.
Treat llms.txt as cheap housekeeping, not a growth lever. It costs next to nothing to add a tidy markdown summary of your site, so we do, the way you'd keep a clean sitemap. Just don't let anyone sell it to you as a ranking trick.
And ignore anyone promising guaranteed AI citations. Nobody can see inside those systems, Google has said as much itself, and the moment a model updates, the trick stops working. You get mentioned because your business genuinely deserves to be.
None of this is dramatic, which is rather the point. The reason I poke at these things on a quiet evening is so that when a client asks "do we need to do something about AI?", the answer is grounded in what's actually happening, not in whatever's being sold that week. Thirty years in, the fundamentals still do the heavy lifting. The edges are worth watching. The trick is knowing which is which.