Short answer: in a public scan of 50 commerce sites, the median score for an AI buying agent was 11 out of 100. The Index says a low score means the agent met a barrier, such as a bot challenge or a rejected add-to-cart. The data is one scan, from one location, so read it as a snapshot and not a verdict.

More shoppers now ask an AI assistant to find and compare products. A practical question follows for anyone running a store: if an agent tried to buy here, how far would it get? The Axidort Index tries to answer it by running the journey, so this piece walks through what it measured, what it found, and where its limits are.

What was measured

The Axidort Index scanned 50 well-known commerce sites on October 1, 2026: 25 B2B and 25 B2C. For each, a declared AI browser agent (it identifies itself as AxidortAgentCheck) tried a purchase journey: search for a product, open it, read the price and stock, and head toward checkout. It never places an order, pays, logs in or submits forms. A second, no-browser reader agent checks what can be read from the page text and data.

The score runs from 0 to 100 and is built from seven components. Per the published methodology, agent behaviour carries 90% of the score and standards and files such as structured data carry 10%.

The headline numbers

All figures below are from the Index page and its JSON data.

  • Median score: 11 out of 100.
  • 72% of sites scored below 50.
  • B2B sites averaged 20 out of 100 (25 sites). B2C sites averaged 34 out of 100 (25 sites).
  • The highest score was 97 out of 100, so a full journey is clearly possible for an agent on some sites.

What actually stopped the agent

The honest answer is narrower than the headline. The Index does not publish a per-stage breakdown of where each agent stopped across all 50 sites, so we cannot say how many failed at search versus price versus add-to-cart. What it does state is that a low score means the agent met a barrier, with a bot challenge and a rejected add-to-cart given as examples.

A bot challenge is the example worth understanding, because it can sit in front of the whole journey and stop an agent before it reaches a product page.

Why bot challenges are worth checking

Bot protection exists for good reasons: scraping, credential stuffing, card testing. The side effect is that a declared, well-behaved agent can look like any other automated visitor. Some vendors let site owners allow known bots by verified identity (see Cloudflare's verified bots documentation for one example), but whether a given store has set that up is a configuration question, not something visible from outside.

Two things follow for a store owner:

  1. A rule written years ago for scrapers may now also stop the shopping assistants your customers use. Nobody has to have decided that on purpose.
  2. Passing a standards checklist does not mean an agent can get through. A site can publish every well-known file and still challenge the agent at the door. That is part of why the Index scores behaviour first.

How to read this data fairly

These limits come from the Index itself, and they matter:

  • Each site was scanned from a single location, at one moment. Sites scoring under 70 were scanned twice and the better score is shown.
  • Scores are automated opinions from public pages, not statements of fact about a company's products, quality or security, and not recommendations to buy or avoid.
  • No listed company sponsored, endorsed or reviewed the results, and names are used only to say which site was measured.
  • The scan covers commerce journeys only and does not look behind logins. It says nothing about price competitiveness, range, delivery or service.
  • The scores can shift slightly between runs, and a different agent or location could see something different.

For that reason this article does not name low scorers. A low score tells you an agent met a barrier on that day. It does not tell you the business is doing anything wrong.

What you can check on your own site

You do not need a scanner to start. Four checks cost nothing:

  1. Ask your developer or platform support whether bot protection challenges automated visitors on search, product and cart pages, and whether any rule lets declared agents through.
  2. Look at a product page with scripts off. Is the price in the visible text and in structured data, or only drawn by JavaScript? The guide on structured data for AI agents covers the markup.
  3. Add an item to the cart in a private window, with no cookies. Does it work, or does it demand a session set up elsewhere?
  4. Check that stock status and units (per case, minimum order) are stated in text.

If you would rather see what an agent experiences than guess, the free score at the end runs the same journey on your domain. It is an automated opinion, takes about three minutes and needs a Google sign-in.

FAQ

Does a low score mean my store is losing sales to AI? No. There is no data in the Index about how many purchases agents make on any site, or about revenue. The score shows how far one declared agent got on one day.

Why is the median 11 when the best site scored 97? The scores are uneven: some sites let the agent get a long way, and many scored low. The median describes the middle of the 50, so one strong site does not move it.

Are B2B sites worse than B2C? On average in this sample, yes: 20 versus 34 out of 100. With 25 sites in each group, treat it as a pattern worth watching, not a rule.

Is 50 sites representative of all commerce? No. It is a set of well-known sites, not a random sample.

Check your own site

The free Axidort Score runs a declared agent through search, product, price and stock, and shows where it stopped. A score is a starting point for a conversation with your developers, not a grade on your business. Get my score

Disclosure: ByteSprint is published by the team behind Axidort, which makes the Axidort Index and the Axidort Score discussed above.