One in seven retailers turned away an AI shopping agent. What CPG brands should check before Black Friday

A Shopify agency called Blend Commerce sent a shopping agent to 200 retailers in the U.S. and U.K. this fall and tried to get one item to the checkout page at each. At 29 of them, roughly one in seven, the agent never got through the door. The list includes Home Depot, Best Buy and Macy's, and on the CPG side H-E-B and Chewy.

The detail that should worry brand teams: these stores mostly say AI is welcome. Blend could read the robots.txt file for 24 of the 29, and every one of those 24 allows the search crawlers used by ChatGPT, Perplexity and Claude. The refusals came from bot protection software, not from policy.

Blend published the audit on October 6 and put out a statement on November 4, which PPC Land picked apart in detail the next day. With Black Friday on November 27, it is worth an hour of your week.

What did the Blend Commerce audit test?#

Blend picked 10 retailers in each of 20 sectors, five per country, and ran 17 checks between September 8 and 15, 2026. The checks sit in four layers: can an agent reach the site, find the products, read the product data, and get a basket to checkout. Nothing was bought. The average score was 65 out of 100.

Read it with one caveat in mind. The agent was Blend's own software, running an unmodified Chromium browser. It did not pose as ChatGPT, Gemini or any named agent. Blend's own chart note says it plainly: a retailer may allow named AI agents while refusing automation it does not recognize. So this is a test of how stores treat an honest but unfamiliar machine, which is how most agents look today. It is not proof that ChatGPT is locked out of Chewy.

Blend also sells conversion audits and says so in the report. Ten stores per sector is a small sample. The numbers are still the most concrete look we have at what an agent hits when it shows up at a store.

How did grocery, beauty and pet retailers score?#

The categories where CPG brands live spread across the whole table. Baby and nursery led every sector. Food and drink sat seven points under the overall average.

The size split runs against intuition. Large retailers refused the agent 20% of the time (16 of 80), mid-sized ones 15% (12 of 80) and emerging ones 2.5% (1 of 40). Big stores have bigger security teams, and those teams are doing what they were asked to do. Nobody asked them about agents with a customer's money.

Here are the U.S. retailers in the audit that sell the categories CPG brands care about most. Filter by sector or show only the stores that refused the agent.

Target scored 48, Safeway 72, ShopRite 40 and Giant Food 34. Ulta scored 72 and Petco 69. If your brand does most of its online volume at a few of these names, their front door is part of your AI visibility whether you like it or not.

Which product data fields are missing?#

Blend could read product pages at 171 of the 200 stores. Most had the basics. The fields that let an agent compare and commit were rare.

GTIN is the one CPG brands own. Every UPC on a shelf is a GTIN, and you assigned it. Yet none of the beauty, pet or department store retailers Blend could read published one in their product data. Without it, Blend notes, an agent comparing prices has to guess whether two listings are the same product. For a 12-count versus a 24-count of the same SKU, that guess is how the wrong pack ends up in the cart. We covered why that mismatch turns into returns in yesterday's post on AI-recommended returns.

Two more numbers: 107 of the 200 publish a product sitemap, and 38 have a live agent checkout endpoint, 37 of them on Shopify. Shopify stores averaged 91 against 59 for everyone else, and still 89 against 66 with the endpoint checks removed. That matches what Shopify told investors about AI orders, which we wrote up on November 4.

Why does this matter before the holidays?#

Because the traffic that agents and AI answers send is now the best traffic a retailer gets. Adobe reported on October 29 that in September, AI traffic to U.S. retail sites converted 90% better than non-AI traffic and generated 83% more revenue per visit. Adobe expects AI traffic to rise 159% year over year on Thanksgiving and 95% on Black Friday.

Adobe also scores how much of a page AI systems can read and cite. The average U.S. retail homepage scored 63%. Grocery scored 55%, behind electronics at 79%.

The fight over access is getting louder too. Amazon blocked Meta's Muse agent in September, while Walmart opened Agent Profiles and an Agent Gateway on October 21, Reuters reported. On November 5, Cloudflare published a paper arguing that in nearly every agent problem it lists, the site bears the primary cost. Its preferred fix is Web Bot Auth, which lets an agent sign its requests so a site can tell who it is. Until that is common, an unfamiliar agent will keep getting the same treatment as a scraper.

What should CPG brands check this week?#

You can't change H-E-B's firewall. You can fix your own site, the data you syndicate, and the questions you ask your account teams. Adam Pearce, Blend's CEO, put it this way in the November 4 statement: "I'd bet most marketing teams don't even know it's happening."

Work through this with your ecommerce and web leads. Tick what you have already confirmed.

Then check the other side of the problem: whether the assistants name you once they can read you. Ask ChatGPT, Gemini and Perplexity the category questions your shoppers ask, every week through December. Our free SKU audit does that per product, and the Searchpanel homepage explains the score. For definitions, see AI shopping agent and structured data in our glossary, or book a demo to walk through your catalog.

Our view: being recommended and being reachable are two different jobs, and most brands only fund the first. Spend the next two weeks on the second.