# I found two prices for the same product on the same website.
> One price was for people. The other was sitting in the version of the page machines could read. What I found testing real product pages, and what my own AI traffic looks like.

AUDIENCE TWO · ISSUE 001 · SEPT 2026
Researching the machines between brands and customers.

Canonical URL: https://audiencetwo.com/newsletter/001
Markdown twin: https://audiencetwo.com/newsletter/001.md
Author: Cam Burley
Published: 2026-09-17
All issues: https://audiencetwo.com/newsletter

---

That's not supposed to happen.

One was for people. The other was sitting in the version of the page machines could read.

And that sent me down a rabbit hole.

## THE SECOND AUDIENCE
We're used to thinking about websites as something people visit. That's not really true anymore.

ChatGPT, Claude, Gemini, Perplexity and a growing list of AI systems are showing up too. They read pages, gather information and use what they find to answer questions for people.

I've started calling them the second audience. And here's the weird part:

**They don't necessarily see the same website we do.**

### FIG. 1 — BOTH PATHS END AT A PERSON
```
                       YOUR WEBSITE
                            │
                   ┌────────┴────────┐
                   │                 │
                HUMANS            MACHINES
                   │                 │
                browse          read / retrieve
                compare              │
                purchase             │
                   │                 ▼
                   │          ChatGPT / Claude /
                   │          Gemini / Perplexity
                   │                 │
                   └────────┬────────┘
                            ▼
                         CUSTOMER
```
- FIRST AUDIENCE: People
- SECOND AUDIENCE: Machines acting for people

The machines aren't an abstract bot audience. They're increasingly intermediating information for the human one.

## THE FINDING: real product pages
I tested real product pages to see what information a machine could pull from them.

Ride1Up was particularly weird.

### RIDE1UP · eBIKE · SAME PRODUCT, SAME URL
- HUMAN: $1,795
- MACHINE-READABLE PAGE: $700

It also surfaced conflicting information about whether the product was in stock and whether it could be returned.

Then I kept looking.

### FISHWIFE
- Price seen by humans: $40; price in the machine-readable page: $44
- Availability seen by humans: Add to Cart; availability for machines: Sold Out

### ALLBIRDS
- Found the $140 price, shipping details and fit information.
- Could also walk away thinking the product was out of stock.

### TRACKSMITH
- Amounts on the page: $80 product and $200 free-shipping threshold
- Flatten the page into text and the machine has to work out what those two dollar amounts actually mean.

This kept happening.

None of this means an AI assistant necessarily gives the wrong answer. It means the information available to machines can be ambiguous or contradictory.

**The information is there. The question is what the machine thinks it means.**

## WHY I'M OBSESSED WITH THIS
Think about the last time you researched something with ChatGPT. You probably never visited half the websites it used.

ChatGPT did. That's the shift.

For 25 years brands have gotten very good at understanding the people who visit their websites. Where they came from. What they clicked. What they bought. Where they dropped.

Now there's another audience showing up between the brand and the customer. And we know remarkably little about it.

## THE FINDING: I CHECKED MY OWN TRAFFIC
This made the whole thing more real for me. On one small site I instrumented, I recorded 397 AI requests. 40 were live-reader fetches.

On another, 555 AI requests. Those aren't all the same thing.

### 555 requests. Very different intentions.
WHO SHOWED UP — ONE SITE, ONE MONTH

| Requester | Role | Requests |
| --- | --- | --- |
| Amazonbot | crawl / index / train | 181 |
| ChatGPT live reader | live reader — a person asked, right now | 137 |
| PerplexityBot | indexing | 51 |
| GPTBot | training | 50 |
| Other | crawl / index / train | 136 |

Some machines are crawling. Some are indexing. Some are training. And some appear because a person asked an AI system something right now.

Calling all of that "bot traffic" feels increasingly useless.

## TRY THIS ON YOUR BRAND (5 MIN)
Pick one product. Ask ChatGPT, Claude and Gemini:
- How much does it cost?
- Is it in stock?
- What does shipping cost?
- Can I return it?

Compare their answers with the actual product page.
FIND SOMETHING WEIRD? REPLY TO CAM: cam@audiencetwo.com
Screenshot is fine. I'm collecting these.

## WHAT I'M TRYING TO FIGURE OUT
I made a course about the Second Audience a little while ago. And almost immediately realized this thing is changing too quickly for me to treat it like a finished subject.

So I'm changing the experiment. I'm getting a small group of people from brands together and comparing notes as this develops.

I'll keep doing the research and sharing the interesting stuff I find. You send me the weird stuff you're seeing. We'll figure out what's actually happening together.

Because right now I have a lot more questions than answers.

- How quickly does ChatGPT notice a product-page change?
- Do different AI systems understand the same page differently?
- Does more crawling mean more visibility?
- What happens when you block one bot and allow another?

And the big one: **What, if anything, should brands actually be doing differently?**
That's what I want to find out.

## Sign-off
Have a great day.

If you work on e-commerce, digital product or growth at a brand, hit reply. Tell me what you're seeing. If you've found something weird, I really want to see it.

And if you know somebody who's already thinking about this stuff, forward this to them.

— Cam

## COMING IN 002 — FROM THE GROUP
Someone runs the test above on their own product. ChatGPT nails it. Claude doesn't. Three answers, side by side. Your result could be that issue.