InMobi’s “two-thirds of stores aren’t agent-ready” stat counts 37% it hasn’t scanned

On 5 October InMobi announced that nearly two-thirds of US online stores can’t yet support an AI agent buying from them. Trade sites are now repeating that number as fact. InMobi’s own press release doesn’t support it: 37 of those 64 points are stores that, in InMobi’s words, “sit on platforms InMobi has not yet scanned.” If you count only the stores it actually scored, 57% reached its top tier. I also ran InMobi’s free scanner on the five Shopify brands I tested last week. On four of them, InMobi’s own check found a checkout profile that supports cart, checkout and orders. None of the five scored above 50 out of 100, because that checkout check doesn’t count toward the score. If a readiness tool has just told you your store isn’t ready for AI agents, here’s how to read the result.

What InMobi’s release actually says

The release opens with the claim: “Fewer than four in 10 U.S. public e-commerce storefronts can complete a purchase made by an AI agent from start to finish, according to new InMobi research covering 3 million of the nation’s digital merchants.” It then gives four tiers. 36% are “Superagent”, 15% “Special Agent” and 12% “Field Agent”. About the last group it says: “The remaining 37 percent sit on platforms InMobi has not yet scanned.”

The tier list says the same thing a second time: “Sleeper Agent (37 percent; a score below 40). These storefronts have not yet been assigned a readiness score.” The CEO’s quote then adds them back in: “Nearly two-thirds of U.S. storefronts still cannot support the full journey today.” Two-thirds is 15 plus 12 plus 37. So more than half of the “not ready” group are stores nobody has measured.

A store that hasn’t been scanned is an unknown. It isn’t a failure. If you leave the unscanned stores out, 36 of the 63 points InMobi did measure are top-tier, which is 57%. That’s my arithmetic on InMobi’s own numbers, so you can check it yourself. A fair headline would have been “of the stores we scored, most were ready”. MarTech Cube, for one, ran it as “Nearly Two-Thirds of U.S. Stores Aren’t AI-Ready”.

What happened when I ran the scanner

InMobi’s tool is free at inmobi.com/agent-ready. The release describes it as “80 live probes across 11 categories, with every check tied to a step an agent must complete to make a purchase.” On 7 October I ran it on the five brands from my WebMCP test. I chose them because I already knew what an agent can do on each one. I recorded the score each store got on the page, and I also saved the raw probe results the page loads, which list every check and whether it counted toward the score.

Four out of five came out as “Sleeper Agent”. Death Wish Coffee scored highest at 50, which is “Field Agent”. Here’s what drove those scores.

The checkout check runs but doesn’t count. The scanner fetches each store’s UCP profile. UCP is the open protocol Google and Shopify use so an agent can find a store’s cart and checkout. On Tentree, Allbirds, Death Wish Coffee and Gymshark, InMobi’s own probe listed cart, checkout, order, discount and payment handlers as present. The raw results label that probe “INFO” and mark it as not scored. Only 38 of the 78 results InMobi returned count toward the score at all. All four of those stores got 20 out of 100 in the section titled “If AI does recommend your products, can a shopper actually buy from your store through AI?” Ruggable has no UCP profile and got 0.

The browser-agent tools aren’t checked at all. For WebMCP, the scanner reports “not checked” because the test “needs a full browser scan, which this service doesn’t run”. WebMCP is how Shopify lets an agent in the shopper’s own browser search and fill a cart. On 6 October I used those tools to add items to a cart on Tentree and on Death Wish Coffee. Both stores are rated as unready for agent purchases.

Product data is checked on the homepage. Every store failed “Price schema markup”, “Price as structured data” and “Stock availability markup”. In the raw results the probe says “no Product JSON-LD on homepage (check product pages)”. Price data normally lives on product pages, not the homepage. The Allbirds Wool Runners page serves structured product data with a price of 110 USD and “InStock” right in the HTML. Even so, the scanner’s top five fixes for Tentree tell it to add Product JSON-LD.

The same tier name means two different things. In the press release, “Sleeper Agent” means a store nobody scored. In the tool, it means a score below 40. The Tentree result shows “33/100” next to the line “Your storefront has not yet been assigned a readiness score.”

Many of the scored checks that failed were for things like an A2A agent card, an MCP server card, an auth.md file and DNS agent records. I haven’t found a shopping agent that requires any of those today. That’s my reading of the current landscape. InMobi doesn’t claim that agents require them.

What InMobi gets right

The tool shows every probe and why it passed or failed, which is better than a single black-box grade. The checks for robots.txt, sitemaps and bot blocking are useful. The release also makes a point I agree with: “A lot of retailers are further along than they realize because the commerce infrastructure beneath their storefront has already evolved.” That’s true. On Shopify, the platform provides the checkout rails for you. The scanner then fails to count them.

A few limits on my side. Five large Shopify brands are a spot check, not a sample. The 3-million-store study may have used a different method from the public tool, and I couldn’t find a methodology page to compare. Disclosure: I build Pollen, which also runs a free store scan. The scanner page closes with “Want a deeper audit? Our team can run a full agent-readiness review and fix issues for you.”

How to read any agent-readiness score

1. Find out which checks were actually scored. A tool can look at your checkout and leave it out of the grade. Open the per-check results and look for checks labelled info, not scored or not checked.
2. Find out which page it looked at. If it only read your homepage, any “missing product data” result tells you nothing. To check a product page yourself, run curl -A GPTBot on a product URL and look for "@type":"Product" (I explain this check in this post on what crawlers see).
3. Look at the rails yourself. Open yourstore.com/.well-known/ucp. If it returns a profile listing checkout, the agent checkout path exists, whatever score you got. On Shopify, whether ChatGPT, Gemini, Copilot or Meta can actually use it is a setting in your admin. I covered that in the AI channels map.

A readiness score can still be useful. Just treat it as one vendor’s list of checks rather than a measure of whether agents can buy from you. And when a headline says two-thirds of stores aren’t ready, look for the line in the release that says how many of them were actually measured.