Your Shop Was Never in the Room

Hi, I'm Berkem. I lead the AI ops team at Storyly, which is also where this newsletter lives.

I have a bad habit: when someone tells me AI has changed something, I go and check. Usually it hasn't, or not in the way they said. Occasionally it has, and then it deserves better than a LinkedIn carousel.

I run these checks for work anyway. Writing one up each month costs me an hour and saves you an afternoon, which is the entire reason this exists.

It is called The Human Layer because the interesting part is never what the machine did. It is what still needs a person, and where that has quietly moved to.

This month's test went somewhere I did not expect.

The test

I asked ChatGPT a normal shopping question: I want a portable monitor that works over a single USB-C cable, what should I buy?

It recommended the ASUS ZenScreen MB16ACV at £144 and cited four sources. All four were brand-owned: ASUS, Dell, and Lenovo twice. If you have been told AI answers are stitched out of SEO blog spam, that is not what happened here. The brands won outright.

Then I looked at which page won.

It cited the spec sheet

Not the shop. Not the campaign page. The technical specifications table, the least loved page ASUS owns.

So I fetched that page the way a crawler reads it. Every spec present. No price. No warranty period. No returns window.

Then I went looking for the page where you actually buy it. ASUS keeps its shop on a separate subdomain, uk.store.asus.com, which returns a 403 to AI crawlers. The model never cited it once, because it never could.

So I asked ChatGPT directly for the official store page. It handed back the marketing page on www.asus.com and called it the store.

Then it started filling in the gaps

I asked whether it would work with my laptop on one cable, the question that actually decides the purchase. Zero sources came back, plus a product capability ASUS's own spec page does not state either way.

I asked about returns and warranty. It cited the MB16AC, a different model, and took the returns conditions from a parts shop FAQ.

Then it quoted me £144, down from £189, and credited ASUS's overview page.

Open that page in a browser and the price is sitting in the top right corner next to a Buy button. Fetch the same page as a crawler and there is no price in it anywhere. The number a person sees and the number a machine sees are not the same number.

None of this is the model being stupid. It is a model doing its best with a brand that put the facts where machines cannot reach.

This is not one unlucky brand

Adobe scored US retail pages this year on how much of their content machines can read. Returns pages came in at 82%. FAQ 80%. Homepages 75%. Product pages, last, at 66%.

Every page that does not sell anything is more legible than the page that does. Your returns policy is a static wall of text nobody has redesigned since 2019, and that is precisely why it wins. Your product page is where all the JavaScript lives.

The more you invest in a page's experience, the less of it survives the trip to a machine.

Fix order: facts into plain text, one canonical page per policy, one name per product everywhere it appears.

I am not going to tell you this cost ASUS a sale. I cannot prove that, and neither can anyone selling you a fix for it. What I can tell you is that a £144 buying decision got answered from a spec table, the wrong model's warranty page, and a parts FAQ, and the shop was never in the room.

Getting recommended is downstream of being readable.

Next month: why the click, not the citation, is where AI traffic actually leaks.

Cheers, berkem

Readability scores from Adobe's Q1 2026 AI traffic report. Everything else I ran myself in July 2026, and you can repeat it. The 403 is ordinary bot protection, not an anti-AI policy: it blocks other automated traffic too. That is rather the point.

More from the Storyly AI Hub