In Google's AI Mode, the source getting cited most is Google itself.

There is a finding from this month that reframes how you should think about local and commercial visibility, and it is not a subtle one. When SE Ranking studied how AI Mode assembles its answers, they found that Google has become the second most-cited source inside its own AI surface, and it earned that position by citing its own properties. Business Profiles, Product Knowledge Panels and other Google-owned entities are being pulled into answers at a rate that dwarfs most of the open web, so the query gets answered and increasingly the material it draws on came from inside Google's own ecosystem rather than from anyone's website.

For years the mental model of local SEO was that your Google Business Profile fed the map pack and your website earned the organic click. That model still runs, but a second job just landed on top of it, because your GBP is now feeding the AI answer directly and the accuracy of what Google says about your business is downstream of what you have put on the properties Google controls.

The study that surfaced the inward turn

The data comes from an analysis of 50,032 commercial keywords in the US, run at the end of June and broken down in SE Ranking's AI Search digest. The headline was that Google-owned entities surged as AI Mode citations, growing roughly 8.4 times in about two months on Authoround's tracking, which is not a drift so much as a step change. Google Business Profiles and Product Knowledge Panels showed up prominently, and the effect was strongest in the categories where Google already holds rich structured data, so hospitality, home services, restaurants, real estate and healthcare.

Two related numbers from the same study sharpen the picture. AI Mode sold ads on around 30% of commercial queries, so the surface is being monetised aggressively, and yet only 11.53% of advertiser domains were cited as sources. Paying to appear and being used as a source are close to unrelated, which is worth holding onto because it undercuts the instinct to solve AI visibility with ad budget.

Why the answers keep resolving inside Google

The self-citation is one expression of a wider pattern, which is that Google's AI surfaces are becoming self-contained. The clearest new example arrived on July 14, the 25th anniversary of Google Images, when Google rebuilt the image search homepage into a gallery layout and announced that AI Overviews can now generate images inside the results using its Nano Banana model, so an informational query with a visual component can be answered with a picture Google made on the spot rather than a click to a stock library or a photographer. Put the pieces together and the direction is consistent. When the answer is text, Google writes it from pages it crawled and from its own entities, and when the answer needs an image, Google can now draw one itself, so the query resolves without the user leaving and the external website is edged out of more and more of these journeys.

Your Business Profile is now a citation asset

The practical consequence is that GBP hygiene has stopped being a tidy-up task and become a visibility strategy. The hours, the categories, the photos, the reviews and the product data on your profile are the raw material Google is feeding into AI Mode answers about your business, so an incomplete or stale profile is not just a weaker map listing, it is a weaker and less accurate presence inside the AI answer.

The pattern we see most often in local audits is a profile that was filled out once at setup and never revisited, with three photos, a primary category that undersells half the services, and product fields left blank. When a profile is that thin, the model has less to work with and leans on a competitor's richer entity or describes the business in vague terms that could apply to anyone.

Fixing it is unglamorous and it compounds. Keeping the profile complete and current, responding to reviews, uploading real photographs rather than leaving Google to pick them, and getting the product and service attributes right are now the work that determines how you show up in the answer, not only in the pack. It is exactly the sort of task that gets deprioritised because it never seemed to move rankings, and it moves something more valuable now.

When your Knowledge Panel is wrong, the AI answer is wrong

Sitting behind the Business Profile is the Knowledge Panel, and its accuracy has become load-bearing in a way it never was. If Gemini is drawing on your Knowledge Panel to describe you, then an error in the panel becomes an error in the answer, repeated to everyone who searches your name or your category, so a wrong founding date, an outdated leadership name, a service you no longer offer or a category you were mis-slotted into now propagates into a confident AI sentence about your business.

The same logic runs through the rest of Google's structured data, since product information in Merchant Center feeds AI shopping recommendations inside AI Mode, which means the quality of your product feed shapes whether the model recommends you and what it says when it does. The uncomfortable part is that much of this data lives on surfaces you maintain but do not fully control the presentation of, so the fix is diligence rather than cleverness. Audit what Google currently believes about you, correct it at the source, and keep checking it, because the cost of a wrong entity is no longer a wrong panel in the corner of the page, it is a wrong answer at the centre of it.

Why buying ads will not buy you the citation

It is worth staying with the 11.53% figure, because it kills a tempting shortcut. When only about one in nine advertiser domains gets cited as a source, spending more on Google Ads is not a reliable path into the AI answer, since the ad and the citation are decided by different systems with different criteria, and the citation is earned through relevance, structure and entity strength rather than bought through the auction. Teams that try to spend their way to AI visibility get impressions and clicks on the ad unit while staying absent from the answer that is actually shaping the user's decision, which leaves the durable work exactly where it has always been, being genuinely authoritative and accurately represented across the entities Google trusts.

What to fix first if you run anything local

This is also not purely a Google problem, which is easy to forget. The same clean entity data, consistent name, address and category, accurate hours and a well-maintained profile, is what feeds Bing Places and, through Bing's index, the answers ChatGPT grounds in it. Getting your entity right pays off across more than one engine, so it is a better use of effort than optimising for a single surface.

The first move is to find out what Google currently says about you inside AI Mode, because you cannot fix a representation you have not looked at. Search your own category and your own brand, read the answer, note where a competitor's richer profile is winning the citation, and check every fact Google is asserting against what is actually true, then work the properties you can influence until the source material is complete and correct. If your map presence and your AI presence have drifted apart, closing that gap is what our Local SEO work does, starting with the profile, the entity and the citations that now decide how you show up.