The pages ChatGPT and Google’s AI cite tend to answer the question in their first lines, show they’ve been updated in the past year, and include prices, tables and first-hand experience. Pages that ranked in Google’s top 10 for the same question but never came up in the AI’s searches had those things less often. I’ll call those the passed-over pages.
At TJ Digital, we do SEO and GEO, and we track the questions our clients’ customers ask AI in Peec AI. As of October 2026, we work with about 80 businesses, and this study covers the 78 client businesses we track in Peec AI. It’s built on 193,507 answers from ChatGPT, Google AI Mode and Google AI Overviews, collected every week from August 27 to October 21, 2026.
Which pages do ChatGPT and Google’s AI cite most?
Product pages get cited more than any other kind of page. They took 28% of all citations in our data, followed by homepages (12%), category pages (10%) and profile pages (9%). Articles of every kind added up to about a quarter of citations (25%), led by how-to guides at 8.5%.
So most of what AI cites is the pages that describe what a business sells. Comparison pages and “alternatives to” pages were rare, at 2.5% and 0.2% of citations.
The mix depends on the engine. ChatGPT cites product pages and profile pages more than Google’s AI does, and articles get only 17% of its citations. AI Overviews gives articles 28% of its citations, and it cites forum-style discussion pages far more often than ChatGPT does.
The narrower question: once an engine finds a page, does it cite it? While it writes an answer, an engine searches, pulls up pages, and links only some of them as sources. AI Overviews and AI Mode cite 85% to 92% of the pages they pull up. ChatGPT cites about 27%, and it favors product pages and homepages (about 31% cited) over comparisons, alternatives and other articles (17% to 20%).
Which websites get cited in AI answers?
Business websites get most AI citations. Company websites, including competitors’ sites, took 67% of the citations in our data, and no single platform came close. That share ranged from 52% for consumer products and e-commerce to 87% for B2B industrial and distribution companies.
The platforms that get the rest change with the business model. Real estate questions leaned on Zillow and Realtor.com, legal questions on Super Lawyers and Justia, home services questions on Yelp, and consumer product questions on YouTube.
These figures come from the questions we track for our clients. Before you put time into a platform, check what AI cites when your own customers ask their questions.
A small share of sites gets most of the citations. The top 10% of domains got 81% of all citations, yet more than half of the cited URLs (55%) were cited only once in the whole study.
How often do AI citations change?
AI citations change a lot from week to week. When the same engine answered the same question a week later, only 24-28% of the pages it had cited were cited again. (When I give a range like that, it runs from the lowest of the three engines to the highest.) After seven weeks, the share dropped to 11-16%.
Even the same question asked twice on the same day gave only 23-25% of the same cited pages. Just 2-3% of cited pages showed up in at least three of every four weekly checks.
The three engines also rarely cite the same pages for the same question. AI Mode and AI Overviews shared 12% of their cited pages, and ChatGPT shared about 5% with either one.
So one AI answer, or a screenshot of one, tells you very little about where you stand. You find out by tracking a set of questions across several engines for weeks, which is what we do for our clients. If you’re choosing a tool for that, I compared the options in my guide to the best AI rank tracking tool.
How often do AI answers cite Reddit and YouTube?
Reddit and YouTube show up mostly in Google’s AI and hardly at all in ChatGPT. Reddit appeared in 16% of AI Overview answers, 7% of AI Mode answers and 1% of ChatGPT answers. YouTube appeared in 23% of AI Overview answers, 9% of AI Mode answers and essentially none of ChatGPT’s.
ChatGPT leans toward official sources instead. Government, university and nonprofit sites took 17% of its citations, compared with 7% in AI Overviews.
Once a Reddit thread gets cited, it tends to keep getting cited. A cited thread had a 47-50% chance of being cited for the same question the next week, about twice the rate for pages in general. About 4 in 10 cited threads were cited in more than one week, and they took about three-quarters of all Reddit citations.
The threads AI cites are usually small and old. The typical one had 6 upvotes and 18 comments, and about half were more than two years old. You don’t need a popular thread to be cited.
Cited YouTube videos were modest too. The typical one had about 2,500 views on a channel with about 12,000 subscribers, and nearly all were captioned. We only had details for 591 videos, so treat this as a rough picture.
I explain how we use Reddit in Reddit for SEO and AI search.
What do cited pages have that other pages don’t?
Cited pages were fresher, showed more first-hand experience and were easier to pull answers from than the pages AI passed over. The biggest single difference was the year in the title: pages with it had more than twice the odds of being cited.
We found this by downloading and measuring 21,874 cited pages and 4,507 passed-over pages, for 434 questions. The comparison accounts for each site’s authority (its Ahrefs Domain Rating) and the type of page, so a difference doesn’t come only from big sites or from product pages.
I give the results as higher or lower odds of being cited. When I say pages with a table had 69% higher odds, I’m comparing them with similar pages that don’t have a table. In the chart, a number above 1 means the trait was more common on cited pages, and below 1 means it was less common.
Freshness made the biggest difference. Along with the year in the title, being published or updated in the past 12 months gave a page 88% higher odds. If you only change one thing, keep your important pages updated and put the year in the title where it makes sense. I’ve written more about putting the year in your page titles and keeping content fresh for ChatGPT.
Pages that show real experience got cited more. Pages describing first-hand experience had 45% higher odds. A “best of” list that explains how the picks were made had 51% higher odds, though fewer pages did this, so that number is less certain. For a business, that’s a good reason to write about what you’ve actually done for customers.
Pages that are easy to pull information from got cited more. A table made the biggest difference (69% higher odds), then an FAQ section (26%) and a list (22%). Longer pages and pages with more H2 sections also did a little better.
This is where the engines differed most. In ChatGPT, length made no difference, and tables mattered even more (88% higher odds). In AI Overviews and AI Mode, length and the number of H2 sections mattered more than they did in ChatGPT.
Pages that use the question’s own words got cited more. AI tools turn a question into a few narrower searches before they answer. Pages whose opening closely matched the question did better, and so did pages whose headings matched those narrower searches, with the opening making the bigger difference.
Pages that show prices had 28% higher odds of being cited. If you’ve been keeping prices off your pages, this is one more reason to reconsider. Schema markup was a little more common on cited pages too (13% higher odds), but that’s an association, and I wouldn’t put schema work ahead of the content itself.
A few traits were less common on cited pages. Pages with lots of images, harder reading levels and long sentences were slightly less likely to be cited. Pages with a call to action had 12% lower odds, and pages with an author byline had 28% lower odds.
I can’t tell you why the byline and call-to-action results came out that way. Because they’re associations, I wouldn’t remove either from your pages based on this alone.
Does domain authority matter for AI citations?
Higher domain authority goes with more AI citations once you compare pages that rank at the same position in Google. We needed a separate check to see it, because Google already picks its top 10 partly on authority, homepages and titles that match the search.
That check used only pages in Google’s top 10. Higher Domain Rating went with more citations in every engine, most clearly in AI Mode. Homepages got cited more, and in AI Overviews, so did pages whose title matched the question. The content results held up too, and the table has the figures for each engine.
Does content quality affect AI citations?
Pages that answer the question directly get cited more. We had AI raters score 195 pairs of pages blind, meaning they didn’t know which page in each pair had been cited. The cited page opened with a direct answer 20% of the time, compared with 5% for the passed-over page.
Many passed-over pages never answered the question at all. Half of them didn’t answer it anywhere in the part of the page the raters read, compared with 23% of cited pages.
Cited pages were also more complete. They scored 2.3 for completeness on a 1-5 scale, compared with 1.7. They were at least twice as likely to include a summary at the top, a “best for” breakdown or a ranked list.
Being promotional didn’t count against a page. About three-quarters of the pages in both groups were a business pitching its own products or services, and the raters scored them almost the same for promotional tone (3.8 and 3.7).
On the narrower question of which pages get cited among those the same answer pulled up, an earlier round of ratings showed the same pattern: the cited page opened with a direct answer 23% of the time, compared with 14%.
Do you need to rank in Google to get cited by AI?
You don’t need a Google top-10 ranking to get cited by AI, but ranking well helps, especially in AI Overviews. Over the eight weeks, AI Overviews cited 24% of the pages in Google’s top 10 for the question, AI Mode cited 17% and ChatGPT 7%.
Very few of the pages AI cites come from Google’s top 10, though. Only 2-5% of cited pages were in it. That’s because each engine cited 24 to 45 different pages per question over the eight weeks, and the top 10 holds only 10.
Our clients’ own sites showed up in 19% of AI Overview answers to questions that don’t name a brand, and in 11-13% of ChatGPT and AI Mode answers. That’s a share of answers. Each answer cites about five pages, most of them from other sites, so as a share of all citations, a client’s own site got about 5% on average.
Our clients’ Search Console data shows that ranking near the top of Google helps. When a client’s page ranked in Google’s top 3 for a question, AI Overviews cited it 13% of the time, compared with 8% for positions 4 to 10. AI Mode and ChatGPT showed the same pattern.
AI also cites client pages that Google barely shows. About a third of the client pages AI cited ranked in Google only for searches that include the brand’s name. Some had no Google impressions at all in 90 days: 5% to 17% depending on the engine, with ChatGPT at 17%.
Where do AI answers get the brands they recommend?
AI answers usually name brands that their cited pages mention. When an answer named a brand, a cited page mentioned that brand 85% of the time in AI Overviews, 66% in AI Mode and 62% in ChatGPT. Some mentions had no supporting source at all: 31% in AI Mode, 25% in ChatGPT and 13% in AI Overviews.
So if you want AI to name your brand, the pages it already cites for your customers’ questions are the place to start. Being first on one of those pages doesn’t mean being first in the answer, though. When a cited page ranked options, the AI’s order followed the page’s order only loosely.
What does ChatGPT search for when it answers?
ChatGPT ran 1.8 searches per answer on average, and each search was about 11 words long. When a question didn’t name a city, state or region, 30% of its answers added a location to at least one of those searches. If you serve a specific area, that’s a reason to say clearly on your pages where you work.
How do you get your pages cited by AI?
Based on this data, these are the changes I’d make first. The full how-to is in my guide on how to get cited in AI search.
- Answer the question in the first lines of the page. It was the clearest quality difference between cited and passed-over pages, in both rounds of ratings.
- Keep pages current, and put the year in the title where it fits. These two traits had the biggest odds in our main comparison.
- Show prices. Pages with prices had 28% higher odds of being cited.
- Use tables, lists and FAQ sections where the content calls for them. Write headings in the words people search. I cover this in how to structure content for AI search.
- Show your experience. Describe what you’ve done first-hand, and when you recommend options, say how you chose them.
- Say where you work. ChatGPT often adds a location to its searches even when the question doesn’t name one.
- Make the pages about what you sell answer buyers’ questions. Product pages, category pages and homepages are the pages AI cites most.
- Get your brand named on the pages AI already cites for your questions. Most brands AI names are mentioned on a page it cites.
- Check which platforms AI cites in your industry before investing in one. Reddit, YouTube, Yelp and Zillow matter a lot for some business models and very little for others.
- Track a set of questions over weeks. Citations change too much to judge from a single answer.
How we ran this AI citation study
- Businesses: the 78 client businesses we track in Peec AI (as of October 2026), across 9 business models.
- Questions: 4,167 buyer questions that don’t name a brand.
- Answers: 193,507 (63,033 from AI Mode, 62,931 from AI Overviews and 67,543 from ChatGPT), with 943,673 citations to 209,859 different URLs on 95,007 domains.
- Timing: August 27 to October 21, 2026, in 8 weekly snapshots, plus repeat runs of the same questions the next day.
- Pages measured: 69,401 pages downloaded and measured, and 866 pages rated blind by AI raters in two rounds.
- Google data: Google’s top 10 from Ahrefs for 371 keywords standing in for 451 questions, and Search Console data for 67 clients.
What are the limits of this study?
- These are associations. The study shows which traits cited pages have, and it can’t prove that any trait causes a citation.
- Keywords stand in for questions. People ask AI full questions, and Google rankings come from keywords. Of the 371 keywords, 174 were close to the question and 197 were broader.
- The Google data is one snapshot. The Ahrefs results come from one crawl location on one date.
- Passed-over pages are Google’s top 10. That’s the comparison group throughout, which is why we ran the like-for-like check.
- The YouTube data is partial. We had video details for 591 videos.
- Some figures depend on what we track. A client’s share of citations depends on the questions we track for that client in Peec AI.
If you have questions about this study or what it means for your own pages, reach out through our contact page or email me at tj@tjrobertson.com.