What AI agents mean for ecommerce and retail

What agents do in ecommerce today

Written for the head of ecommerce, whoever owns the store platform and the payments relationship, and the marketing lead who pays for traffic that agents now handle.

• 20 Sep 2026

  Amazon blocked Muse from its store around 20 September 2026; Shopify keeps direct checkout through Meta and Muse on by default for eligible stores, with an off switch for merchants.

  Trade press report. Retail Gazette: Amazon blocks Meta's Muse (22 Sep 2026), read 4 Nov 2026
• 16 Sep 2026

  Google said on 16 September 2026 that its Universal Commerce Protocol already enables direct checkout for hundreds of thousands of brands and retailers across Google, and that merchants can now transfer a shopper's cart to their own site.

  Google: holiday shopping updates (16 Sep 2026), read 4 Nov 2026
• 8 Sep 2026

  Meta launched Muse, a personal agent that runs its own browser and can make purchases with the user's approval, in the US on 8 September 2026.

  Meta: Introducing Muse (8 Sep 2026), read 4 Nov 2026
• 13 May 2026

  Amazon's Alexa for Shopping, which replaced Rufus in the US in May 2026, includes Buy for Me, which completes purchases of eligible products on other retailers' sites using the customer's primary address and card.

  Amazon: Meet Alexa for Shopping, read 4 Nov 2026
• 12 Mar 2026

  Modern Retail reported on 12 March 2026 that OpenAI was moving Instant Checkout to apps and that Shopify products in ChatGPT would typically be bought on the merchant's own storefront.

  Trade press report. Modern Retail: OpenAI retreats on Instant Checkout (12 Mar 2026), read 4 Nov 2026
• 8 Jan 2026

  Microsoft launched Copilot Checkout on 8 January 2026: shoppers buy inside Copilot, payments run through PayPal, Shopify or Stripe, and the merchant stays the merchant of record.

  Microsoft Advertising: Copilot Checkout and Brand Agents (Jan 2026), read 4 Nov 2026

Decisions for leadership

These three belong to the head of ecommerce, not to the developers. Each can be settled in a meeting, and each changes what the web team is asked to build.

1. Which AI shopping surfaces may sell our products, and which are switched on today by platform default.
2. Whether an agent may complete a purchase without the customer confirming the final step.
3. Whether personal agents that do not identify themselves are allowed in, and what our terms say.

• Whether an agent may complete a purchaseWhether a customer's agent may finish the purchase or must hand back the last step.
• Whether prices and policies are readable as textWhich prices, fees and policies are plain text on the page an agent reads.
• Whether to accept agent paymentsWhich agent payment routes you accept, and through which provider.

The numbers that frame the risk

Figures labelled as vendor claims come from Ora or another vendor's own publication and are given with the date they were read. The crawl figures are this site's own.

• 5 Nov 2026

  On Ora's public leaderboard read 5 Nov 2026, the 153 entries in the E-commerce category averaged 52.7 out of 100; 18 scored B (70) or above and 3 scored A (86) or above.

  Vendor claim. Ora public leaderboard, category averages, read 5 Nov 2026
• 4 Nov 2026

  In the network's own crawl, run of 4 Nov 2026 (13,071 of 13,071 domains from Tranco list 746WX plus Ora leaderboard domains), 6.9 percent of 8,624 sites with a readable robots.txt disallowed ChatGPT-User at the root and 5.9 percent disallowed Perplexity-User, the fetchers that act on a person's question.

  Scorecard methodology: the crawl of 4 Nov 2026, read 4 Nov 2026
• 4 Nov 2026

  In the network's own crawl, run of 4 Nov 2026 (13,071 of 13,071 domains from Tranco list 746WX plus Ora leaderboard domains), 2.1 percent of 10,385 sites published an MCP manifest, OpenAPI file, agent card or agents.json at the standard paths, and 0.59 percent published a UCP profile at /.well-known/ucp.json.

  Scorecard methodology: the crawl of 4 Nov 2026, read 4 Nov 2026
• 8 Oct 2026

  Ora's lab found (8 Oct 2026) that agents clearly recommended agent-ready businesses in 20 percent of answers against 11 percent for businesses that were not agent ready, across 37,927 runs on 1,056 business sites.

  Vendor claim. Ora research: AX is the new AEO (8 Oct 2026), read 4 Nov 2026

Where ecommerce stands on Ora's leaderboard

Ora ranks sites on a public leaderboard, grouped by category. On 5 November 2026 it listed 184 sites in the Commerce Tools and E-commerce categories, the ones that stand in for ecommerce in this site's crawl. The median score was 55 out of 100 and the top score was 98. 29 of the 184 scored B (70) or above. These are Ora's figures, vendor-published, and they describe the sites Ora lists, not every business in the sector.

The middle half of the listed sites scored between about 42 and 65. A score of 41 or lower sits in the bottom quarter of the ecommerce sites Ora lists, and 66 or higher in the top quarter. To place your own site, scan it with Ora (free, public pages only) and read the result against these numbers, then see how competitors in your sector score.

Where ecommerce stands in the crawl

• 95%allow the four main AI agents
  190 of 201 sites with a readable robots.txt let GPTBot, ClaudeBot, PerplexityBot and Google-Extended in at the root.
• 83%show readable text on the home page
  176 of 211 home pages gave a plain fetch 150 words or more of text.

The sample is 256 ecommerce domains from the crawl of 4 Nov 2026. The sector figures use Ora leaderboard categories as stand-ins, not a definition of the sector: E-commerce (153 rows), Commerce Tools (31 rows), plus 72 domains placed by a keyword on the domain name. The full ecommerce scorecard has all four measures and the method.

Check your own business

• Self-check for ecommerceThe 18 core questions with examples from this sector, plus two extra questions. Scored in your browser.
• Boardroom brief for ecommerceOne page you can paste into a deck, with the decisions and the numbers above.
• Scorecard for ecommerceThe four crawl measures for this sector and how they were counted.