For the last year, whenever a credit union CMO asked me “are we showing up in AI search,” the honest answer was: I can sample it, but I cannot prove it. I could run 50 prompts through ChatGPT and Gemini and Google’s AI Mode, log which sites got cited, and hand over a spreadsheet. Useful. Not authoritative.
On August 31, Google changed that, at least for its own AI features. The Generative AI performance report in Search Console is now live for every property worldwide. Log in, open Performance, and there is a new report that counts how many times links to your pages appeared inside AI Overviews and AI Mode. You can split it by page, country, device, and date.
If you run marketing for a credit union or community bank, you should open it this week. Here is what I would look at, and what I would be careful about.
What it actually shows
Impressions. That is the whole metric. An impression means a link to your page was shown inside an AI answer. It is the closest thing to a citation count Google has ever given site owners, and it is free.
Group it by page and the report becomes interesting fast. My expectation, based on what I see across financial sites, is that the pages getting AI impressions are rarely the pages marketing cares about most. Rate tables and product landing pages barely register. Pages that explain something in plain language, like how a HELOC draw period works or what a share certificate actually is, are the ones that get pulled into answers. If your site has ten product pages and two explainer pages, do not be surprised if the two explainers carry most of the report.
What it hides
Three things, and they matter.
No clicks. You can see that a page was shown 400 times in AI Overviews last month. You cannot see whether anyone clicked through. Given what we know about click behavior inside AI answers, most did not. Similarweb’s July numbers on where generative AI referral traffic actually lands are a reminder of how concentrated that traffic is: Reddit took 59 percent and Wikipedia 41 percent of the referrals they measured. Everyone else was a rounding error.
No queries. The report will not tell you what someone asked to trigger your page. So you know the HELOC explainer got shown, but not whether the question was “best HELOC rates in Austin” or “can I use a HELOC to pay off a car.” Those are different members with different intent, and the report treats them the same.
No ChatGPT, no Perplexity, no Gemini app. This is Google Search only. AI Mode and AI Overviews. For a lot of your members, the first AI question about a loan happens in ChatGPT, and that stays invisible here.
The opt-out switch
The same rollout added a control that lets you exclude your site from Google’s AI features entirely. Google says it is not a ranking factor for regular search. I would still leave it alone. Opting out means your competitors’ explainer pages, or Bankrate’s, get cited instead of yours. For a regulated institution, the risk of being absent from the answer is bigger than the risk of being quoted imperfectly.
What to do with it
Pull the report for the last 28 days and sort by page. Note which pages get any AI impressions at all. Then compare that list against your traditional search performance for the same pages. The gap between the two lists is your working GEO priority list.
Second, take the pages that are already getting AI impressions and make them better answers. Clearer first paragraph. Direct sentences that can be lifted. Updated dates. Structured data that matches the content. These are the pages Google has already decided to trust; make them worth trusting.
Third, set a baseline now. This report is new, and every credit union that starts tracking it in September will have a year of data by next fall. The ones who do not will be guessing.
One caution. Google notes the report only shows data for properties above a minimum impression threshold, and the newest data is preliminary. If your report is empty, that is a finding too. It means your site is not currently feeding Google’s AI answers, and that is the conversation to have with whoever owns your content.
I have spent 20 years reading marketing reports that measure what is easy rather than what matters. This one measures something that matters. It just does not measure all of it.