New research from Profound measured how often Google AI Mode cites Google itself. The number moved 8.4x in eleven weeks.
It explains a lot of what we have been watching across ChatGPT, AI Overviews, Gemini and Perplexity all year.
It also lines up with what our own tracking keeps showing inside Fulcru (fulcru.app): the pages AI quotes are almost never the pages brands spend their money on.
Let's break it all down.
Quick context. Profound is an AI search analytics company. Their research team, with analyst Davis McCain, published this on 14 July 2026. Named analyst, stated window, stated sample. That matters, because most numbers in this category have none of the three.
1. What they actually measured
Between 15 April and 30 June 2026, Profound tracked citation share inside Google AI Mode across more than 32 million instances of google.com/searchviewer.
In that window, citations pointing back to google.com grew 8.4x.
That makes google.com the number two most cited domain in AI Mode. Not a publisher. Not Reddit. Google.
2. The part everyone skips: which Google pages
The growth is not general search results. It comes almost entirely from two surfaces:
↳ Google Business Profiles
↳ Product Knowledge Panels
Both are structured cards Google hosts and controls, showing hours, photos, reviews, specs and prices.
So the citation resolves inside Google. The user reads your hours, your reviews and your price without ever reaching your site.
Your Business Profile is now your landing page, whether you treat it as one or not.
3. A second study, different method, same curve
SE Ranking ran a separate analysis published in February 2026: 68,313 keywords, 1,321,398 AI Mode citations.
↳ June 2025: google.com appeared in 5.7% of AI Mode answers
↳ February 2026: 17.42%
↳ Google was the most cited domain in 19 of the 20 niches tracked
The only exception was careers, where Indeed was cited about three times more often.
Two vendors, two methods, one direction. That is the definition of a trend rather than a headline.
4. Where it bites hardest
The shift is not spread evenly. It concentrates where Google's own cards hold the richest data.
↳ Travel: 53.18% of citations go to Google
↳ Entertainment and hobbies: 48.74%
↳ Real estate: 30.54%
↳ Insurance: 6.48%
↳ Finance: 5.13%
Profound names the same verticals from the other side: hospitality, home services, restaurants, real estate and healthcare.
If you sell locally, this is already your reality. If you sell finance or insurance, you still have room.
5. Why this is structural, not a phase
Ask why a model would prefer a Google card to your page.
↳ The card has uniform fields the model can quote with confidence
↳ It is always fresh, because Google owns the database
↳ It never returns a 403 to a crawler
↳ Your page might be paywalled, badly marked up, or blocking the bot outright
A model optimising for a reliable answer takes the reliable source. Structural advantages compound, which is exactly what the curve from 5.7% to 17.42% to 8.4x shows.
6. What our own data says about the rest of the answer
Google's cards take the top of the answer. The rest still goes somewhere, and it is not your homepage.
We track 601 sources feeding answers in our own category. The biggest are reddit.com, youtube.com, semrush.com, ahrefs.com and linkedin.com.
Not one of them is a company homepage.
We published our own number two days ago, unedited: 34 buyer questions, three engines, zero mentions of us. The tool that measures it is ours, which is exactly why we published it.
https://x.com/stfvsl/status/2082191782876475609
7. What to actually do this week
If you sell locally. Fix the card before you touch the site. Complete every field, replace stale photos, answer the reviews. The card is what the model quotes and what the buyer reads.
If you sell products. Product structured data is now a visibility lever, not a nice to have. The Knowledge Panel quotes your feed.
Everyone else. Count first. Run the buyer questions in your category across every engine and write down who gets named. You cannot fix a gap you have not measured.
The free report at fulcru.app does that in about two minutes, no card, no account.
If you want your agent to run the whole loop instead:
npx skills add gsmmediaro/fulcru
That is the free skill. It finds the questions you lose, writes the page that closes the widest gap, and re-runs the same questions afterwards so you see whether the number moved. Repo: https://github.com/gsmmediaro/fulcru
The shape of the change
The old model was: index, rank, click.
The new model is: crawl, quote, decide.
Google just took the quote step in house for local and product queries. Everywhere it has not, the quote comes from Reddit, YouTube and LinkedIn.
Either way, the page you are proudest of is not in the room.
If you want the prompt set we use to measure this per category, reply "COUNT" and I will send it over.
Top comments (0)