We introduced Google AI Overviews as the fifth platform worth testing in an earlier post, and made the case that it's arguably the highest volume AI recommendation surface a small business will ever appear in. Explaining why it matters is one thing. Actually testing it properly is another, and most business owners who try skip half the steps that make the test meaningful. Here's the process we use, in order.
Step 1: write down the real questions first, not guesses
Before opening a browser, list six to eight questions phrased the way an actual customer would ask them, not the way you'd write a Google Ads keyword. "Best golf club near Wilmslow for a corporate day" beats "golf club Wilmslow." "Reliable emergency plumber open weekends in Leeds" beats "plumber Leeds." The more naturally phrased the question, the more representative the test, because Overviews responds to the same conversational query patterns that chat assistants do.
Step 2: search each query in a signed out or incognito browser
Google personalises results based on your account, your location history, and your search history. If you're signed in on your own device, testing "best golf club near me" will often surface your own club just because Google knows you work there, which tells you nothing useful. Use an incognito or private browsing window, and if possible clear location permissions or test from a device that isn't tied to your business's own Google account.
Step 3: record whether an Overview appears at all
Not every query triggers an AI Overview. Some still return a traditional results page. That's a useful data point on its own: if none of your category's core queries trigger an Overview, the platform matters less for you right now than it might for a competitor category where Overviews appear consistently. If an Overview does appear, that's the point where the real test starts.
Step 4: read who's cited, not just who's named
An Overview typically summarises an answer and links out to two or three source pages beneath it. Note three things for each query: whether your business is named in the summary text itself, whether your website is one of the cited sources, and which specific page of your site is cited if it is. A page being cited that isn't the page you'd have chosen (an old blog post instead of your current pricing page, for example) is itself a useful and fixable finding.
Step 5: check what a competitor is doing differently
If a competitor is cited and you're not, open their cited page and compare it against your equivalent page. In our experience the gap is rarely dramatic: it's usually that the competitor answers the specific question in the first two sentences of a page, in plain text, while the equivalent content on the losing site is buried under introductory paragraphs, locked behind a booking widget, or simply missing. That comparison is more useful than any abstract SEO advice, because it's a direct, in-category example of what's working right now.
| What to record per query | Why it matters |
|---|---|
| Did an Overview appear? | Tells you whether the platform is even active for this query category |
| Is your business named in the summary? | The clearest sign of Overview visibility |
| Is your site cited as a source? | Shows whether your content is trusted enough to link to, even without being named |
| Which page is cited? | Confirms whether the right page is being surfaced, not an outdated or irrelevant one |
| What does a cited competitor's page do differently? | Gives you a concrete, in-category benchmark rather than generic advice |
Step 6: repeat monthly, not once
Overviews content changes as Google's index updates and as competitors change their own pages. A single test tells you where you stand today. Testing your core queries monthly, alongside the same discipline you'd apply to rank tracking, tells you whether changes you've made are actually moving the needle, and catches a competitor's improvement before it costs you real visibility.
Where to start
Pick your six most important customer queries, run them signed out this week, and record the five data points in the table above for each. It takes under half an hour and gives you a genuinely current answer, rather than an assumption carried over from however your SEO looked a year ago.