What Is Google Choosing From Your Website for AI Search?

I've been paying quite a bit of attention recently to how AI is changing search.

Part of that is because it affects the websites we build, but I'm also interested in it from the point of view of my own website. I write these Field Notes about software development, websites and technology, and naturally I want to understand how people find them.

Until recently, understanding whether any of that content was appearing in Google's AI results involved quite a bit of guesswork.

Google Search Console now has a Generative AI performance report. It shows when pages from your website appear in AI Overviews and AI Mode, and Google rolled it out worldwide at the end of August.

At first glance it looks like another analytics report. There's a graph, some impressions and a table of pages.

But I think there's a more interesting question we can start asking with it.

What is Google actually choosing from our websites when it needs information for an AI response?

A real example from this website

A few days ago I wrote a Field Note about the release of Scala 3.9.

I could have written an article explaining that Scala 3.9 had been released and listing everything that had changed, but there wasn't much point. The Scala team had already published an excellent release announcement.

Instead, I wrote about the question I'd be asking if I was responsible for a Scala application in production: should we actually upgrade yet?

That article started ranking in Google remarkably quickly. When I checked a few days later, it was appearing second for "scala 3.9", directly underneath the official Scala website.

Now there's another thing I can watch.

Does Google also decide that article is useful when somebody asks a more conversational question in AI Mode, perhaps about whether it's safe to move a production application from Scala 3.3 to 3.9?

Search Console won't tell me the exact question somebody asked. That's one of the limitations of the new report.

But it can tell me whether that particular page is being surfaced in Google's AI features.

I'm genuinely interested to see what happens.

This could tell us something useful about our websites

Think about a completely different business.

Imagine a company running a handful of holiday cottages in Northumberland.

Its website probably has the pages you'd expect. Home, accommodation, prices, availability, things to do and contact details.

Those pages are important, but they're also fairly similar to thousands of other accommodation websites.

Now imagine the owner has also put together a really useful page about staying in Northumberland with a dog. It explains which cottages have enclosed gardens, which nearby beaches allow dogs throughout the year and which local pubs are dog friendly.

Someone planning a holiday might ask Google something like:

"Where can I stay near the Northumberland coast with two dogs and an enclosed garden?"

Suddenly that detailed page has something useful to contribute that a generic accommodation page doesn't.

If Search Console starts showing that page regularly appearing in Google's AI results, I'd want to know.

Not because we need to celebrate another impression, but because we've learnt something about the information people are finding useful.

The interesting bit isn't the total number

The new report doesn't tell us everything.

An impression means that a link to your website appeared within one of Google's generative AI features. It doesn't mean somebody visited the website, made an enquiry or bought anything.

Google also isn't currently showing the queries that produced those impressions.

That makes the big headline number considerably less interesting to me than the pages underneath it.

If I discovered that my homepage had 5,000 AI impressions, I'm not sure what I'd do with that information.

If I discovered that three Field Notes were responsible for most of them, I'd immediately want to compare those articles.

What were they about?

What questions did they answer?

Were they based on experience rather than information that could be found on hundreds of other websites?

And perhaps more importantly, what did I write that Google didn't use?

That starts to become useful.

Maybe the type of content that works will change

Traditional Google searches have often been short.

"scala 3.9"

"dog friendly cottages northumberland"

"accountant newcastle"

AI search makes it much easier to ask something more complicated.

Instead of searching for "legacy software modernisation", someone might explain that their business has a ten-year-old application, the original developers have left, releases are becoming difficult and they're trying to decide whether to rewrite it or improve it gradually.

That's a very different search.

A generic page saying "we provide software modernisation services" doesn't answer much of it.

A detailed article written by somebody who has dealt with older production systems might.

That doesn't mean every business suddenly needs to start producing hundreds of articles for AI. But it could make genuinely useful, specific knowledge on a website more valuable.

Google's own guidance around generative AI search points in a similar direction. It talks about original, non-commodity content and says the normal foundations of good SEO still apply. Google has also explicitly said that things such as llms.txt aren't required for Google Search and don't improve visibility or rankings there.

So there doesn't appear to be a magic AI file we can add to a website and consider the job done.

I'm going to watch what happens

That's probably what I like most about this new report.

It gives us something we can actually observe.

Over the next few months I'll be watching which Mugnai Digital pages appear in Google's AI results. I'm particularly interested in the Field Notes because they cover quite different subjects.

Does a technical Scala article appear more often than something about website design?

Do practical articles perform differently from broader opinion pieces?

Does an article continue appearing months after the news that prompted it has passed?

And does being visible there eventually translate into people actually discovering the website?

I don't know yet.

But that's much more interesting than somebody giving me a checklist of ten things I'm apparently supposed to do to optimise a website for AI.

For the first time, we've got a little window into what Google is choosing from our websites when its AI tries to answer somebody's question.

I'm going to be watching mine.