Does AI Understand Your Store? — Built for Humans, Read by AI

Ask ChatGPT or Gemini "what does this store sell?" — can the AI actually read your site correctly? Much of today's web is built for human eyes, and its meaning doesn't come through to AI agents. This article works through why that happens, and lays out the two fronts of preparation — a presentation that gets you found by humans, and data that lets AI read you correctly — using data from a sample EC store.

TL;DR#

1. Today's web is optimized for humans and hard for AI agents to read

   Text baked into images and decoration-first layouts are friendly to the human eye, but unfriendly to an AI that has to read meaning and act on it.

2. The entry point for finding information is shifting from search clicks to AI answers

   AI increasingly picks candidates and presents them in one summary, and being read and chosen correctly by AI is starting to matter.

3. An EC store's preparation runs on two fronts: "found by humans × read correctly by AI"

   A human-facing presentation alone isn't enough, and AI-facing data alone isn't either. Only both together make you ready.

4. Two-front preparation starts today by measuring AI-referred traffic

   AI-referred traffic is already happening, yet it hides inside Direct and unknown. Making it visible is the first step.

1. Why Today's Web Is Hard for AI Agents to Use#

Bottom line: today's web pages assume human eyes and human operation. So on the very same site, what's friendly to a person can be unfriendly to an AI agent that reads meaning and acts on it.

The EC sites we look at every day sell the appeal of products through photos, catch the eye with decoration, and prompt action with buttons. Humans draw meaning from that look. But what an AI agent wants to read isn't the look — it's data that carries meaning: whether price, stock, shipping terms, and product specs are laid out as clear text or structure. This is where human optimization and AI needs diverge. Text baked into images, content painted onto the screen after the fact, information buried inside decoration — all of it becomes hard for AI to read.

AI researcher Shota Imai said on the PIVOT program that "an AI agent is useless on a human-only web," and pointed out that the web will start being reorganized for agents [1][2]. In other words, a way of building that assumed only humans as readers is being questioned now that AI has arrived as a new reader. If your site is easy for humans to see but AI can't read it, you risk being left behind in AI-driven acquisition.

2. A New Entry Point Is Emerging: the AI Browser#

Bottom line: the entry point for finding information is shifting from clicking search-result links yourself to AI presenting a summarized answer. AI-embedded browsers are the symbol of this.

In search as we knew it, users compared the listed links and chose which site to open. Ask an AI, though, and it picks candidates from multiple sources and hands you a summary. AI-native browsers like OpenAI's "ChatGPT Atlas" have appeared. Users receive the answer AI chose instead of opening links one by one. For EC, this change means that "whether you can become a source AI reads and chooses" drives your exposure.

One thing to watch here is not to get ahead of expectations. You can't pin down on the spot "whether AI is citing you." What you can know is the traffic side — how many people reached your site by way of an AI answer. That's exactly why you need both: preparing a design AI can read correctly, and measuring how much traffic arrives via AI. How AI chooses its sources, and why a bare AI can only return generalities, is also covered in AI gets smarter by connecting, not training.

3. Prepare on Two Fronts#

Bottom line: prepare on two fronts — "a presentation that gets you found by humans (GEO)" and "data that lets AI read you correctly." Either one alone falls short.

GEO is the craft of presentation that gets you found within human search and AI answers. The other front, AI-facing data, is preparing your product information in a clear, structured form so AI can read it directly. For example, an MCP (a mechanism that connects AI to your own data) that lets AI read your numbers embodies this idea of handing over data in a form AI can interpret. If humans find you but the content can't be read, you won't be chosen; if it's readable but no one finds you, it won't reach anyone. So you need both fronts in place.

Which front you're weak on can be roughly gauged with a few questions. Is product info readable as text? Can you measure AI-referred traffic? Do you have data in a form AI can read? Starting from this diagnosis, the practical move is to begin with the weaker front. The full picture of EC preparation for the agent era is in preparing EC for the agent era, and the concrete way to let AI read your own data is in connect and try before you build.

4. You Can Start Measuring It Today#

Bottom line: two-front preparation starts not from a big rebuild, but today, by measuring how much AI-referred traffic reaches your site. AI-referred traffic is already happening, yet in most cases it hides inside Direct and unknown.

The volume of AI-referred traffic itself often still sits at a tiny fraction of the whole. But now that the entry point for finding information is shifting toward AI answers, this traffic is trending upward. The problem is that it's hard to see. Someone who followed a link from an AI answer tends to land without a clue as to where they came from, and GA4 lumps them into Direct or unknown.

By hand, you can open GA4's referrer report, pick out AI-service domains, and confirm part of your AI-referred traffic by eye. But treat this as no more than a rough estimate. AI assistants don't always pass a referrer when they follow a link. When it isn't passed, GA4 drops it into Direct or unknown, and doing this by hand can't avoid misses and undercounting. On top of that, new AI services keep appearing, and having a person keep judging which referrers count as AI traffic every single time is too heavy to repeat monthly. The idea is simple, but running it by hand is heavy — that's the line where you hand it to a machine.

Note that AI-referred traffic first disappears inside Direct. How to tell this hard-to-see traffic apart is covered in detail in why AI-referred traffic hides in Direct.

RevenueScope helps

Bottom line: RevenueScope makes AI-answer-referred traffic visible as an independent channel, and finds pages that are cite-worthy yet missing traffic. Its job is to carve out the AI-referred traffic that GA4 buries in Direct, without the manual repetition.

RevenueScope's get_ai_traffic carves out, as a single channel, the traffic that arrives via answers from tools like ChatGPT, Claude, Gemini, Perplexity, and Copilot. Alongside that, it lists pages that are cite-worthy yet aren't receiving traffic as "missed." GA4 is an indispensable tool for a whole-site health check, and RevenueScope is not a replacement but a complement. It's just that in GA4, AI-referred traffic gets buried in Direct or unknown and is hard to carve out as an independent channel. That's the gap it fills. Because the MCP is read-only, letting AI connect and read your data carries no risk of rewriting it.

Here's the actual view, with sample-store data.

Asking the sample store about AI-referred traffic (sample data)

Figures from a fictional EC store with sample data (RevenueScope demo). AI-referred traffic is a classification based on referrer and browser information; whatever isn't passed goes missed.

Let me draw the line honestly. What RevenueScope produces is the channel-level record of how much traffic arrived via AI answers, plus a lead on pages that are cite-worthy yet aren't getting traffic — and no further. Since AI assistants don't always pass a referrer, it can't be fully comprehensive and may come out low. It doesn't pin down which AI answer generated each visit one by one, nor does it judge whether AI is citing you. Showing the honest range of what it can do, its job is to lighten, as much as possible, the monthly chore of picking AI-referred traffic out of Direct by eye.

FAQ#

Q. If I work on GEO, will I know whether AI is citing me?

A. GEO (the craft of presentation that gets you found by AI and search) is preparation to "make yourself easier to find" — separate from judging "whether you're being cited." RevenueScope's get_ai_traffic, too, makes traffic arriving via AI answers visible as a channel, but it doesn't pin down which AI answer cited your page one by one. The practical use is to first measure how much AI-referred traffic arrives, then find pages that look like they're being missed.

Q. I hear AI-referred traffic is small in volume. Is there any point in measuring it?

A. Right now, AI-referred traffic often stays a tiny fraction of the whole. But as the entry point for finding information shifts from search to AI answers, this traffic is trending upward. Setting up how you measure it while the volume is small keeps you from later discovering "it had quietly grown to a size we couldn't ignore." The value of making it visible now only grows from here.

Q. Where should I start?

A. You don't need a big rebuild. Start by measuring how much AI-referred traffic reaches your site. Your GA4 can give you a rough estimate, but AI doesn't always pass a referrer, so doing it by hand misses some. Once you've made the measurement something you can repeat automatically, move on to the two-front design: a human-facing presentation, and data that lets AI read you correctly.

Summary#

Today's web is built for humans and hard for AI agents to read. Now that the entry point for finding information is shifting from search to AI answers, an EC store's preparation runs on two fronts: "a presentation that gets you found by humans" and "data that lets AI read you correctly." And that preparation starts not from a big rebuild, but today, by measuring AI-referred traffic. Your GA4 can give you a rough estimate, but since AI doesn't always pass a referrer and doing it by hand misses some, the shortcut is to first make the measurement something you can repeat automatically. Let's move the two-front preparation forward one step at a time, starting from measuring.

References#

• [1] PIVOT official channel "Shota Imai: 'An AI agent is useless on a human-only web' (1on1 Tech)" 2025
• [2] TBS NEWS DIG "The web's ruler shifts from humans to AI — Shota Imai: 'An AI agent is useless on a human-only web' (1on1 Tech)" 2025