If you are judging your AI content by blog pageviews, you will see a flat line even when it is working

SEO
, ,
1

A lot of us are being told to write content so that AI assistants recommend our products. Far fewer of us have a way to check whether it is happening. Here is how I would approach it without buying anything.

Start with the referrer, and know what you are actually looking at.

Shopify does not have an AI traffic segment. What it has is Analytics, Reports, Sessions by referrer. Look for chatgpt.com there. One thing worth understanding before you read that number: a lot of what shows up is there because ChatGPT appends utm_source=chatgpt.com to the links it puts in an answer, not because a clean browser referrer was passed. That is still a real signal, someone clicked a link inside a ChatGPT answer, but it is the reason the number is a floor rather than a total.

Expect to undercount, and know why.

The ChatGPT mobile apps often do not pass a normal web referrer at all. Those sessions land as direct and blend into the rest of your unattributed traffic. So the honest reading of whatever you find is “at least this much”, not “this much”.

Now the part that trips people up, and the reason I am posting.

Do not judge your AI content by blog pageviews. Assistants read your article, then link the shopper to the product the article recommends. The article does the work and never shows up as a pageview. So if you are checking whether your AI content is landing by watching blog traffic, you will see a flat line even when it is working. I have watched people conclude their content failed on exactly this evidence.

What to do instead.

Pick a product that has been getting direct traffic you cannot explain. Then check which of your own blog posts link to that product. If a post links it, and the product started picking up unexplained direct visits after that post went live, that is your signal. It is imperfect and it is not proof of causation. It is a lot closer to the truth than blog pageviews.

To be clear about what this is not. It is not an attribution method. It is a way to stop drawing the wrong conclusion from the one number that is easiest to look at. Nothing here separates a ChatGPT driven direct visit from someone who typed your URL, and I cannot close that gap with the data I have.

The thing I cannot answer on my own: has anyone here actually found chatgpt.com sitting in their Sessions by referrer report, and roughly what share of sessions was it? I would like to know whether it shows up cleanly for other people or whether everyone is looking at the same fog I am.

2

I’d track this as a before/after pattern, not try to prove every visit came from AI.

  • Keep a simple sheet with the post publish date, products linked, and each product URL.
  • In Shopify, open Sessions by landing page, filter to that product, then compare 28 days before and after publishing.
  • Break those sessions down by referrer. Count chatgpt.com separately, but watch direct traffic for a sustained lift rather than a one-day spike.
  • Also compare add-to-cart and conversion activity for that product. Ten qualified product visits matter more than 100 article views.

I would not add UTMs to internal blog-to-product links. They can overwrite the shopper’s original attribution and make the report harder to trust.

3

This is better than what I posted; especially the UTM point. Completely agreed:
tag your internal blog→product links and you overwrite the original source, then
spend a week wondering why ChatGPT traffic is attributed to your own campaign.

One thing I’d add to the before/after method, because it caught me out: the
article→product mapping is usually many-to-many and much denser than people
expect. When I actually crawled the blogs I was working with, the average product
was linked from around 40 different articles. So “28 days before this post went
live” often isn’t a clean baseline — other posts linking that same product
published inside the window too.

Two things that helped:

  • For your first tests, pick products linked by only one or two articles. Fewer
    products, much cleaner signal.
  • Or stagger publishing so each product gets one new linking article at a time.

On watching direct traffic for sustained lift rather than a spike — that’s the
right instinct, and there’s a concrete reason for it. The ChatGPT mobile app
frequently sends no web referrer at all, or an android-app:// one. So the
chatgpt.com line in your referrer report is a floor, not a total, and the
remainder lands in direct.

And strongly agree on add-to-cart over article views. Ten people arriving on a
product page with intent is a different event from a hundred people reading a
guide.