Your Best Pages on Google May Not Exist to AI
By Ava Nicewonder, Head of SEO
I came across a study last month that I keep thinking about. Not because it surprised me exactly, but because it put real numbers behind something I had been watching play out across client accounts for months.
The piece, by Adam Gnuse on "The SEO-GEO gap: How AI search traffic differs from organic traffic", analyzed GA4 data from 10 websites, covering 150,000 indexed pages across a single month in March 2026. The researchers isolated LLM referral sessions from ChatGPT, Claude, Perplexity, and Copilot, then compared them directly against traditional organic sessions. What they found is something every DTC brand should sit with.
Your top Google pages and your most-cited AI pages are almost certainly not the same content.
That is not a small nuance. That is a structural problem. If you are running a Shopify brand and treating your Google rankings as a proxy for AI visibility, you are working with an incomplete picture.
What the data actually says
The study's top 10 organic pages captured 55% of all organic sessions. Those same pages captured only 29% of LLM sessions.
Think about what that means for your brand. The pages you have spent the most time optimizing, the ones ranking at the top of Google, are generating less than half their proportional share of AI traffic. It is like a restaurant packed every lunch shift that sits empty every dinner service, even though the menu is the same. The crowd changed. The criteria changed.
LLM traffic is correlated with organic performance, but it is not simply organic performance re-labeled.
That line from the study is the one worth writing down. SEO and GEO (generative engine optimization) are two different evaluation systems running at the same time, judging the same pages by different rules.
And here is the number that really stops me: of the top 100 organic pages in the study, 49 had zero LLM traffic during the entire month-long window. Not low LLM traffic. Zero. Half of your best Google pages may be completely invisible to AI systems right now.
A note on your GA4 data: it is probably undercounting
Before going further, there is something important to understand about how AI traffic shows up in your analytics, because it changes how you read everything else.
GA4 does not have a native LLM channel. The study methodology notes this directly: LLM bot crawls (GPTBot, ClaudeBot, and others) are not recorded by GA4 at all because they make server-level HTTP requests before client-side JavaScript fires. The sessions the researchers could measure were only the human visits referred from AI platforms, and even those arrive inconsistently. Some show this as direct traffic. Some show as referrals. Some get lost in organic.
Think of it like a store with six entrances but only one door counter. The foot traffic is real. You are just not counting all of it.
What this means practically: if you set up proper GA4 channel groups to isolate LLM referrals, you will see more AI traffic than you expected. This means you should still assume the number you see is an undercount of actual AI-influenced visits.
What AI actually cites (and what it ignores)
This is where the study gets useful for your day-to-day content decisions.
The researchers categorized blog content by topic theme and compared LLM citation rates across types:
That 12% figure for how-to content is the one that should land hardest. How-to posts are the backbone of most content calendars, including ones I oversee. They perform well for SEO. They are relatively cheap to produce. And they get cited by AI at a rate that is, frankly, dismal.
The reason is almost embarrassingly simple: AI can write how-to content itself. Asking ChatGPT how to set up an abandoned cart flow is like asking an accountant what a receipt is. The model already knows. It does not need to pull from your site.
But ask that same model what the average abandoned cart recovery rate is for Shopify beauty brands in Q1 2026 and it is stuck. It has to find that information somewhere. That is where original, specific, data-forward content earns citations that generic guides never will.
The question that changes your content strategy
This shifts the framing for every piece of content you brief. The question is no longer just what does this person search for on Google. It is what would this person ask an AI, and can this page answer it in a way the model cannot produce on its own.
For most brands at the $250K to $1M stage, the majority of existing content was built to answer Google queries. Some of it will continue to perform well, but there is a growing category of questions your customers are now asking AI assistants before they ever open a search engine, and if your content is not structured to answer those questions, you are simply not in that conversation.
The finding that surprised me most
I expected blog content to dominate LLM citations by raw volume. It does, but that is a misleading metric.
When the researchers normalized for LLM sessions per 1,000 organic sessions, a fairer measure of relative performance, service and product pages outperformed blog content significantly:
The study also found that interactive tools showed the highest per-page LLM citation rates of any category. LLMs actively recommend specific tools by name when users ask about evaluations or comparisons. If your Shopify store has a quiz or product recommendation tool with a clear, searchable name, it is probably one of your strongest GEO assets right now, and you may not even know it.
For product detail pages, the implication is harder to ignore. Most PDPs are written to convert, not to answer. A PDP that answers the question, “does this protein powder have third-party testing?”, “will this wine pair with spicy food?”, or “how long does this hair oil last?” has a real shot at being cited in an AI response. A PDD that uses keywords based on search traffic might say “premium formula for superior results” although this is not the case when it comes to AI search.
That is a completely different brief than what most copywriters receive for product pages today.
What this means for your brand specifically
Based on what the study shows and what we are observing across client accounts, here is what to expect:
The engagement quality from AI traffic is higher
The study found something interesting about time-on-page. On average, LLM and organic sessions look nearly identical: 46.9 seconds for organic versus 47.1 for LLM. But that average hides what is actually happening.
On tool and demo pages, LLM users spent an average of 146 seconds versus 101 seconds for organic visitors. On homepages, LLM users spent 82 seconds versus 36 seconds for organic. Someone directed to your page by an AI after asking a specific question is not browsing. They are evaluating. That intent difference is significant.
Some of your pages may only be getting AI traffic
The study found that 14% of all LLM-receiving pages had zero organic clicks during the study window. Not low organic traffic. Zero. These are pages AI systems are recommending that Google is essentially ignoring. Their engagement quality was among the highest recorded in the study. Do not dismiss them as low-performers just because they do not show up in your organic rankings report.
The compounding effect
AI citations tend to compound. A page cited by one LLM tends to get referenced by others. Original data you publish now can drive AI referrals well into the future, the same way a well-optimized blog post can hold a Google ranking for years. The brands building for this now are creating an advantage that is genuinely difficult to close later.
How EcomIQ is addressing this
We have been tracking LLM referral traffic across client accounts for several months. Here is where we stand:
GA4 custom channel groups for AI traffic
The first move is making the traffic visible. We built GA4 custom channel group configurations that isolate referrals from ChatGPT, Claude, Perplexity, Copilot, and other LLM sources into a dedicated channel. This is now standard setup for new clients. If you want to understand the full scope of what we cover, take a look at what we cover at EcomIQ.
Answer capsule integration in content briefs
We updated our content brief template to require an answer capsule on every page we brief. An answer capsule is a two to four sentence direct answer to the page's core question, placed early, written in clean prose, with no internal links inside it. Think of these as a “Too Long; Didn’t Read TL;DR” section. Prior research cited in the study, covering 15 domains and nearly two million sessions, found this to be the single strongest structural predictor of LLM citations.
ZipTie.dev for AI citation monitoring
We use ZipTie.dev to monitor AI citation activity across accounts. The tool tracks which pages are being cited by specific LLMs and how citation rates change over time. This gives us a feedback loop that keyword ranking tools do not provide.
Product and collection page GEO audits
Based on the finding that product and service pages outperform blog content for LLM traffic, we are running structured GEO audits for PDP and collection page copy across our managed accounts. The audit evaluates whether each page directly answers a specific question, whether that question maps to what someone would actually ask an AI, and whether the copy is structured in a way that makes it easy for an LLM model to extract and cite.
Original data publishing
This is the highest-ceiling work and the most effort. We are helping clients identify internal data they have not published: seasonal performance trends, product usage patterns, customer behavior benchmarks. You are sitting on data an LLM cannot produce on its own. That is genuinely valuable.
Three things to do this month
You do not need a developer or a complete content overhaul. These are the highest-leverage starting points:
1. Make your AI traffic visible in GA4
Set up a custom channel group to isolate LLM referrals. Without this, you are flying blind. You cannot optimize a channel you cannot see.
2. Add an answer capsule to your top 10 organic pages
For each of your top 10 organic pages by traffic, write a two to four sentence direct answer to the page's core question. Place it early. No internal links inside it. Specific, not general. This is the single structural change that research shows most strongly predicts LLM citations.
3. Publish one piece of data you already have
Pick one data point from your own business that your audience would find genuinely useful. Your best-selling SKU by season. What your post-purchase surveys actually say. Your customer return rate after a first order. Write one specific, honest post around it. This is exactly the content this study proves will get cited.
FAQ: What DTC founders are asking about this
I rank well on Google. Does this actually matter for me?
Yes. The study is clear: high organic rankings do not translate proportionally into AI citations. Your SEO investment is not wasted. It is just not doing the work in the AI channel that you might assume.
Is my GA4 data on AI traffic accurate?
Not fully. GA4 does not have a native LLM channel, and AI-referred sessions arrive inconsistently across direct, referral, and organic buckets. The study's methodology explicitly notes that LLM bot crawls are not captured by GA4 at all. The traffic you measure after setting up proper channel groups is real, but it is still an undercount.
What is the difference between SEO and GEO?
SEO gets your pages to rank in Google and traditional search engines. GEO (generative engine optimization) is the practice of structuring content so AI systems like ChatGPT, Claude, and Perplexity cite and reference it in their responses. Google rewards authority, backlinks, and technical performance. AI systems appear to reward specificity, original data, and content that directly answers questions the model cannot answer from its own training data. Same pages, two different judges.
Does this apply to brands at my stage, or just big brands?
It applies at your stage, arguably more than it does for large brands. At $250K to $1M, you are competing on merit: the quality and specificity of your content. That is a level playing field where a focused approach can get you cited alongside brands ten times your size.
Do I need to rewrite all of my content?
No. Adding answer capsules to existing pages and making sure your top product and collection pages directly answer a specific question are the highest-leverage starting points. Full rewrites are not necessary in most cases.
How long does it take to see results?
Based on what we are observing across accounts, expect a lag of several weeks before changes to existing pages show up in LLM citation behavior. New content structured for GEO from the start tends to get picked up faster.
Should I stop investing in SEO?
Absolutely not. Organic search is still the dominant channel. The content structures that perform well in GEO (specific, direct, data-forward) also tend to perform well in SEO. Build for both.
What does EcomIQ actually do in this area?
SEO and GEO strategy is a core part of what we coach through. That includes GA4 setup, content brief development, site audits, and ongoing citation monitoring. See how our coaching works or apply to join the founding cohort. We also have a deeper post on GEO for Shopify specifically in the resources section.
The bottom line
The study's closing point is worth restating: GEO is not replacing SEO. It is a second evaluation system running alongside the first, one with different criteria and different winners.
The brands that do well going forward will be the ones who understand both systems. They will know which pages perform on Google, which pages get cited by AI, and where those two groups overlap or diverge. Most brands at your stage have not run that audit yet. That is actually your opportunity.
Here is how EcomIQ works if you want to talk through what this looks like for your brand specifically.
