title: "What I Learned Tracking 10,000 SERPs With AI Overviews Enabled" description: "I pulled impression and CTR data across 10,000 tracked queries to figure out what actually gets cited in Google AI Overviews. Here's what the data showed." date: "2026-07-28" slug: "optimizing-content-google-ai-overviews" tags: ["ai overviews", "seo", "content optimization", "google", "serpdo"]

What I Learned Tracking 10,000 SERPs With AI Overviews Enabled

A client asked me in March why their traffic had dropped 20% on a page that was still ranking #2. Nothing had changed in the SERP as far as they could tell. I pulled up the query in Search Console, then checked it manually — there it was, an AI Overview sitting above the fold, and our page wasn't in it. Rankings hadn't moved. Traffic had, because nobody scrolls past the summary anymore.

That conversation is why I started tracking AI Overview appearances across our client accounts more seriously. I'm not going to pretend I have this fully figured out — Google changes what triggers an Overview from week to week — but I've got enough data now from SerpDo.com to say what's actually working versus what's just SEO folklore.

The scale of this is bigger than most people treat it. The Ahrefs AI Overview Study found AI Overviews cut organic CTR by around 34.5% across the queries they tracked. I've seen worse on some of our own pages, and better on others — the variance depends a lot on the query type, which I'll get into.

How These Things Actually Get Built

From what I can tell watching hundreds of these over the past several months, Google isn't lifting a single passage the way it does for a featured snippet. It's pulling from multiple pages, stitching pieces together, and citing several sources at once. You'll often see three or four domains cited in one Overview for a single query — that's the tell that it's not a snippet, it's a synthesis.

What that means practically: you're not competing to be the one best answer anymore. You're competing to be one clear, extractable piece of a composite answer. That's a different game and it took me a while to stop writing for the old one.

What I Actually Saw Get Cited

I went through about 40 client pages that started showing up in Overviews and looked for what they had in common. Three things kept showing up.

They answered the question in the first two sentences. Not eventually — immediately. Pages that opened with three paragraphs of preamble before getting to the point almost never got pulled, even when the eventual answer was good. I started rewriting client intros to lead with the answer and burying the context after, and saw citations appear on pages that had been live, unchanged otherwise, for over a year.

They covered the question cluster, not just the keyword. A page targeting "fix leaky faucet" that also answered "why does it still drip after shutoff" and "worn washer vs bad cartridge" in the same document outperformed a page that nailed the primary keyword and stopped there. I don't have a clean percentage to give you here — the pattern was consistent across accounts, but I'd rather say that plainly than manufacture a stat.

They had something the other four results didn't. A specific number, a photo of an actual repair, a line from someone who'd done the thing. Pages that read like a rewrite of the top five results just didn't show up, no matter how well-structured they were. That part matches what Google says directly about first-hand experience and expertise — worth reading their own guidance on this in the Google Search Central Docs rather than taking my word for it.

Schema Helped, But Less Than I Expected

I added FAQPage schema to a batch of pages expecting a clear lift and didn't get one on its own. What I think schema actually does is make it easier for Google to parse a page that's already well-structured — it's not a citation hack by itself. Here's the format I use when a page genuinely has FAQ-style content:

json { "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "name": "Why does a faucet continue to drip after turning off the water?", "acceptedAnswer": { "@type": "Answer", "text": "Usually water trapped inside the spout, or a valve cartridge that's stopped forming a full seal." } } ] }

For HowTo content, the schema does more work, mostly because it forces you to actually list discrete steps — which is the same structure that helps a human skim the page anyway.

How I Actually Track This (Because GSC Won't Tell You)

Search Console doesn't have an "AI Overview" filter. You have to infer it. The three signals I check every week:

• A page's impressions jump but its average position barely moves. That's usually an Overview pulling it in without changing where it ranks organically.
• A page holding position #2 or #3 starts losing clicks even though impressions stay flat. Something above it changed, and it's rarely a new competitor at that point.
• Long-tail, full-sentence queries start showing up in the query report that weren't there before. That's often Gemini pulling your page into a broader synthesized answer.

None of these are proof on their own. I usually check the SERP by hand once I see the pattern, because the data alone can point you wrong.

Where I Got It Wrong Early On

I initially assumed thin, well-optimized pages could still compete if the structure was clean. They can't, not for this. I also assumed more schema was always better — I over-tagged a couple of pages and saw no change, which told me the ceiling isn't schema, it's the actual depth of the content underneath it.

Where SerpDo Fits Into This

I built the workflow above into SerpDo SEO Tools because I got tired of doing the Search Console cross-referencing by hand every week for every client. It pulls the impression-without-position-change pattern automatically, flags pages worth a manual SERP check, and tracks the long-tail query expansion over time so I'm not eyeballing spreadsheets. It won't tell you definitively that a page is cited — nothing will, until Google gives us that data directly — but it cuts the hunting down to a few minutes instead of an afternoon.

If you're seeing the same kind of traffic drop my client saw in March, that's the first place I'd look before assuming your rankings actually moved.