How Agents Shop - Sep 14

Share
How Agents Shop - Sep 14

How Agents Shop | kya labs

How Agents Shop

The week's news and original research on AI shopping.

Issue 01 | 14 September 2026 | kya labs

Five of 23 sites let agents in through a browser but blocked direct page requests.

We tested two ways an agent can reach a website: opening a browser, much as a person would, and requesting the page directly.

Five of the 23 sites responded differently. If you only test one approach, you could miss a problem with the other.

The checks covered real estate, home services, autos and resale.

Worth your attention

Shopify adds Meta as an AI channel

Shopify Changelog | 8 September

Shopify says Meta is now available in Agentic Storefronts, with products shared through Catalog by default. Check which AI channels your catalog reaches and who controls direct checkout.

Instacart launches Clementine for grocery shopping

Instacart | 9 September

Instacart says Clementine turns conversations, recipes and lists into shopping carts. It's available to most customers in the U.S. and Canada. The next test is whether the cart matches what the shopper asked for.

Mastercard announces Agent Connect

Mastercard | 9 September

Mastercard announced Agent Connect and an expanded Agent Suite to help merchants connect with shopping agents and support purchases customers authorize. Before choosing a platform, decide which actions an agent should be able to take for your customers.

kya Observatory | Original Research | 25-31 August 2026

Agent navigation: real estate sites were easier to enter.

Four of the six real estate sites let in all the types of agent we tested. That compared with three of five home services sites, and one of six in both autos and resale.

How the 23 sites responded to shopping agents

Agents used a browser or requested pages directly. Numbers inside bars count sites.

Real estate
4 2
6 sites
Home services
3 2
5 sites
Autos
1 1 2 2
6 sites
Resale
1 2 2 1
6 sites
Both approaches allowed Browser visits only Both approaches blocked Unclear result

Home services shows why the distinction matters. At two of the five sites, agents could get in through a browser but were turned away when they requested pages directly. A single test could give the site owner a misleading picture.

One check for your team

Ask someone to try the same shopping task using each approach. Can the agent reach the right product or listing? Where does it stop? Compare the results with the experience you want customers to have.

See how agents navigate your site

Register interest in a weekly custom tracker from kya Observatory. $79/month.

Tell us your domain Explore kya Observatory

About the research

We observed 23 selected sites from 25-31 August 2026 and compiled these findings on 4 September. Twelve results were first readings and remain provisional. The chart shows whether agents were admitted, not whether they finished shopping. This small sample does not represent the whole industry. Site names are withheld; unclear results are shown separately.