A large marketing agency published a study of nearly 600,000 AI sessions in July 2026. It set out to answer one question: “What content do we actually need to get found in AI search?”
The headline findings:
- Nearly one-third (31%) of all AI traffic goes to homepages
- More than 90% of AI referrals land on consideration and decision-stage content
- 98% of AI referral traffic comes from ChatGPT
How does any of that tell you what content to create?
It doesn’t.
What a study needs to show which content gets found in AI search
To find out what content gets a business found in AI search, a study has to do three things.
- Compare pages that get found with pages that don’t
- Examine the content itself - what the pages say and how - not only what type of page they are
- Measure being found - named, cited, or linked in an AI answer
This study does none of the three.
How the study counted AI sessions
590,868 AI sessions, from businesses in "15+ industries." None of them named. No time frame given.
-
They took the 2,500 URLs with the most AI referral traffic.What that means Every page was already winning. No page without AI traffic was included.
-
They counted visits that came from a link in an AI assistant's answer.What that means Clicks only. Not whether a business was named or cited in the answer.
-
They reported every result as a share of AI sessions.What that means No share of total traffic. You cannot tell how big AI is next to Google.
-
They compared AI traffic with organic search traffic.What that means Clicks from Google's AI Overviews and AI Mode count as organic. The comparison has AI on both sides.
-
They labeled each page by type, funnel stage, and intent.What that means How pages were labeled, and by whom, is not published.
What the study found, and what each finding measures
| Finding | What it measures | Does it tell you what content to create? |
|---|---|---|
| 31.28% of AI sessions landed on homepages | Where AI clicks landed, by page type | No |
| 92% landed on consideration (36.7%) and decision (55.3%) pages | Where clicks landed, sorted by the agency’s own funnel labels | No |
| 68.9% landed on transactional content | Where clicks landed, sorted by the agency’s own intent labels | No |
| 97.51% of AI referrals came from ChatGPT | Which tool sent the clicks | No |
| Pages with citations averaged 26% more AI traffic | A difference between two groups of winning pages | No. Both groups were already winning |
| “AI Lift” - blog articles +7.26 points, product pages +7.13, homepages -12.67 | A page type’s share of AI traffic minus its share of organic traffic | No |
Every finding describes where clicks from AI landed, on websites that were already getting them. None of them examines what the content says.
The total traffic the study leaves out
590,868 AI sessions sounds like a lot.
Out of how many?
The study never says how much total traffic these websites got. Every number in it is a share of AI sessions - 31% to homepages, 92% to decision and consideration pages, 98% from ChatGPT.
If those websites had 30 million sessions in the same period, AI was about 2% of their traffic. If they had 3 million, it was about 20%. Those are two very different pictures, and the study doesn’t say which one it is.
That isn’t what they set out to measure. Without it, though, there is no way to know how much of your content plan AI traffic deserves.
Across my clients’ websites, roughly 1-2% of total traffic comes from AI assistants.
An example client website, August 2026. Share of all sessions.
On this website, AI assistants sent 1.19% of all traffic. Every percentage in the study describes how a slice like that 1.19% is divided between page types. None of them says anything about the other 98.81%.
And that 1.19% is ChatGPT, Claude, Perplexity, and the other assistants. Clicks from Google’s AI Overviews and AI Mode are inside the 68.76% from organic search. Google does not separate them.
What holds up in the study
- The sample is large. 590,868 sessions is enough to describe where AI clicks land on these websites.
- They state their own limits. “This study measures AI referral traffic, not AI citations.” And “The dataset focuses on high-performing AI pages.”
- Some of the advice is sound. “Businesses don’t need a completely different content strategy for AI.” Clear homepages, strong service pages, and specific answers are good practice.
The problem is that none of that advice comes from the data. It was good SEO before this study, and the study doesn’t test it.
Where the study’s conclusions go beyond its data
“This reflects how people increasingly use AI”
A breakdown of landing pages does not show how people use AI. It shows where the clicks landed.
On the same websites, homepages get 44% of organic search traffic - more than the 31% from AI. If homepage traffic reflected a new way of using AI, organic search would not send a bigger share of its traffic to homepages than AI does.
“AI traffic magnets”
Product and service pages are most of the pages on a business website. Traffic landing on them is expected. Calling them magnets would take traffic per page, compared with how many pages of each type exist.
The study also records where people landed, not what they asked. Without the question, there is no way to tell whether AI preferred these pages or whether people asked the kind of question these pages answer. AI and Google send people to the page that fits what they asked.
“AI rewards comprehensive websites”
They write that “the highest-performing websites weren’t relying on one standout page.” What makes a website highest-performing is never defined. If it means the websites with the most AI traffic, the ones with more pages had more chances to land URLs in the top 2,500. That is the selection rule at work, not a reward.
Covering the questions your customers ask does help. It did before AI.
“More bottom-of-funnel than you think”
More than what? The study doesn’t compare AI with organic search on the same websites.
It also counts only people who clicked. Research that happened inside ChatGPT or Google’s AI, without a click, is not in the data. AI answers most informational questions on the spot, so the clicks that are left go to pages where people act.
“Invest where AI Lift is highest”
AI Lift compares a page type’s share of AI traffic with its share of organic traffic. Both columns add up to 100%. When homepages take a smaller share of AI traffic, other page types have to take a larger one.
The traits of winning pages
Citations. Expert credentials, on more than two-thirds of the pages. Original research, on 52.6%.
These describe pages that already won. Pages with the same traits that got no AI traffic were never included.
Limitations of the study
- Winners only. The 2,500 URLs with the most AI traffic, and no comparison group.
- Clicks, not visibility. Referral sessions, not whether a business was named or cited.
- Nothing outside the website. The study didn’t consider the authority these businesses have across the internet - reviews, mentions, links, reputation. AI answers draw on all of it. A page can win because of the business behind it, not its content.
- No total. Every number is a share of AI sessions. The study never says how much of these websites’ traffic came from AI.
- Google’s AI is on the organic side. Google counts clicks from AI Overviews and AI Mode with regular search results, and analytics shows them as organic search.
- No time frame, no named industries, no published labeling rules.
- Unnamed websites. The same page sells AI SEO services and reports 617,125 AI sessions driven for the agency’s clients.
What it would take to learn which content gets found in AI search
The question was “What content do we actually need to get found in AI search?”
It is a much harder question than this study makes it seem. To answer it, a study would need:
- One industry, or a few defined ones. Every industry is different. Fifteen unnamed industries mixed together hide that.
- A random sample that counts everything. Every page and every traffic source, not only the pages already winning AI traffic. 10,000 pages or a million, the size is less important than counting all of it.
- Pages that don’t get found, next to pages that do.
- Every other channel, for the same pages. In ecommerce, most clicks from AI land on product pages. Most clicks from Google and from social land there too. Without that comparison, it says nothing about AI.
- Everything outside the website. Reviews, mentions, links, and reputation across the internet, for every business in the sample. AI draws on all of it.
- A test. Change the content on some pages, leave others alone, and measure the difference.
That is a very large study. This one sits on a page that sells AI SEO services.
Even a study built that way would show tendencies across many websites. It still would not tell you what content your website needs.
That question gets answered one website at a time.
What the study means for your content
Their answer is a list of takeaways. “Build a content ecosystem that establishes authority in your niche.” “Prioritize content that helps customers make decisions.”
Neither is wrong. Neither tells you what to create, and neither comes from their data.
The general advice has been around for a long time. Answer the questions your customers ask. Show what makes your business different. Show your expertise.
However, it can be easier said than done. Which questions? What makes your business different? How do you show expertise on a page?
Without knowing exactly what to do, you could write the wrong content, or random content that doesn’t help Google or AI find or understand your business at all.
And none of that accounts for the content you already have. Some of it is helping. Some of it may need to be reworked or removed.
General advice has a cost. Pages get added that aren’t needed. Authority gets shown in ways that don’t fit the business.
What content you need depends on your industry, your customers, what you already have, what you are missing, and what is said about your business everywhere else.
If you are not sure, the most accurate way to find out is to have someone evaluate what you have and build a strategy for your website.
A study of someone else’s winning pages can’t tell you what your website needs.
Sources
- AI Features and Your Website developers.google.com
When you want to rank on top of Google and show up in AI answers, this is the work.
SEO + AI Search Optimization
