We Tracked 66 ChatGPT Sessions Across 120 Shopify Stores. One Landed on a Blog Post.
Merchants are told to write content for AI, then judge it by blog traffic. We tracked 66 AI-referred sessions across 120 Shopify stores — 85% landed on a product page, one on a blog article. Why blog pageviews are the wrong scoreboard.
Every guide to "GEO" or "AI SEO" tells Shopify merchants the same thing: publish helpful content, and AI assistants will send you shoppers.
So merchants publish. Then they open their blog analytics, see a flat line, and conclude it didn't work.
We instrumented it. Between August 8 and August 16 we recorded every AI-referred session across our merchant base — 120 stores — and logged the exact page each one landed on. Here is the distribution.
| Landing page | Sessions | Share |
|---|---|---|
| Product page | 56 | 85% |
| Collection | 5 | 8% |
| Home page | 3 | 5% |
| Other page | 1 | 1.5% |
| Blog article | 1 | 1.5% |
| Total | 66 |
One session in sixty-six landed on a blog article.
Those same stores are not short of blog content. Between them they have 339 published articles carrying 7,137 links to their own products. One store alone has 100 articles and 5,014 product links — about 50 links per article.
The content exists. It is being read. It is simply not where the shopper arrives.
What is actually happening
An assistant answering "what's a good gift for a 30th anniversary" does not hand the shopper a reading list. It reads the article, extracts the specific product the article recommends, and links to that.
Shopper asks ChatGPT
→ ChatGPT reads /blogs/guide/best-anniversary-gifts
→ ChatGPT recommends "Oxidized Silver Chandbali Earrings"
→ Shopper lands on /products/oxidized-silver-chandbali-earrings
The article did every bit of the work. It never appears in the session log.
This is not a quirk of one assistant or one template. It follows from what a shopping answer is for. The article is the evidence; the product is the answer. Assistants link to answers. (It's the same pattern we watched product-by-product in one jewelry store's live session feed — every landing a product page.)
Why this matters more than it sounds
It means the instrument most merchants use to evaluate AI content reports close to zero regardless of how well the content performs.
If you judge AI content by blog pageviews, you are measuring 1.5% of the effect and calling it the whole result. Worse, the failure is silent and one-directional — it never over-reports, so it never prompts anyone to check. Merchants quietly conclude the strategy is dead while it is working.
We had this exact bug in our own product. Our attribution asked "did this session land on the article URL?" and returned zero, forever, for every store — including one with 125 published articles and 17 of them cited by ChatGPT. The feature wasn't broken. The question was.
What to measure instead
Build the reverse map. For every product page an AI session lands on, ask which of your articles link to that product.
- Crawl your own live blog and record every
/products/href in each article. You now have an article → product map. - When an AI-referred session lands on a product, look up which articles link to it.
- Credit one article per session, not all of them.
Step 3 is not optional. In our data the average product is linked by roughly 42 different articles. Credit them all and a store with 7 real sessions renders a table summing to 63. Any merchant who adds up the column will catch it and stop trusting every number you show them. Prefer the article the assistant has actually been seen citing; fall back to the earliest.
This is also why blog traffic is the wrong scoreboard for AI content in the first place — the honest instruments are session arrivals and mentions, not pageviews. (The full measurement guide is here.)
The limits of this data
Worth stating plainly rather than burying:
- 66 sessions is a small sample. Four stores carried measurable traffic.
- Ten of 120 stores had measurement enabled, so this describes the stores that instrumented early, not a random sample.
- An 8-day window. Long enough for a direction, not for a seasonal claim.
- We are not claiming the sessions caused sales. Sessions and orders come from different systems and we do not divide one by the other.
What we are claiming is narrow and, we think, solid: when an AI assistant sends a shopper to a Shopify store, the shopper lands on a product, not on the article that convinced the assistant to recommend it.
If your AI content strategy is being judged on blog traffic, it is being judged by the wrong number.
About the author
Rahul — Founder, FoundGPT
Rahul built FoundGPT and its first-party AI-session pixel. Across 111 Shopify stores he has run 23,000+ real AI-search checks on ChatGPT, Gemini, Google AI Mode and Perplexity, and applied ~99,000 structured-data fixes.