Is your website ready for AI agents that shop, book and call for your customers? 10 checks

An agent ready website is one a customer’s AI agent can use: it can find the company, read the right price and reach checkout. Ten checks show what a customer’s agent gets on the website, the phone line and the support chat. Each check has a pass mark and an owner, so a company can decide which agents to block, allow or check first.

An agent ready website is one a customer’s AI agent can use: it can find the company, read the right price and reach checkout. An AI agent is software that searches, fills a basket, books or phones a company for a person. Ten checks show what a customer’s agent gets on a company’s website, phone line and chat. They are written for heads of ecommerce, digital and contact centres, who decide which agents to block, allow or check first.

Meta launched its Muse agent on 8 September 2026, and Google began testing Gemini’s Call for Me, which phones businesses, on 24 September. Black Friday, on Friday 27 November, is the next big sale. From its own analytics data, Adobe forecasts that clicks from AI services to US retail sites will rise 130% year on year over the 2026 holiday season, and 95% on Black Friday. That counts people following an AI’s link, not agents shopping for them.

Meta’s Muse AI agent and the others arriving now

Each of these agents works for one person, the customer, and reaches a company through its website, its phone line or a platform the company has joined. They belong to a wider group called machine customers: nonhuman economic actors that obtain goods or services in exchange for payment, in a definition CMSWire attributes to Gartner.

• Meta’s Muse. Meta launched Muse in the US on 8 September 2026 as a personal AI agent that sends emails, books travel, opens a browser, fills out forms and negotiates for its user. Meta says it keeps working after the app is closed and checks with the person before a purchase. On 6 October Meta told businesses that Muse already makes purchases, books appointments and resolves support questions. It does this on a business’s own website or through a connector, a direct link into the business’s systems.
• Amazon’s block. Implicator reported that Amazon blocked Muse from its store on 20 September 2026. It added that the next day Shopify’s chief executive announced Muse checkout with Shop Pay, Shopify’s payment wallet, on all Shopify stores.
• Shopify’s Shop app. Shop’s help page says shoppers can connect a personal AI agent, such as OpenClaw or Hermes, or an agentic platform such as Muse, to their Shop account. The agent can search, prepare checkouts, place orders the shopper approves, track orders and read a store’s published return policy. The store stays responsible for fulfilment, returns and refunds.
• Google’s calls. On 24 September 2026 Google began testing Gemini’s Call for Me with US Pixel 11 owners on a paid Gemini plan, TechCrunch reported. Gemini calls a business from the user’s own number to check stock, book a table, move an appointment or put an item on hold. It works through phone menus and waits on hold.
• Walmart’s Agent Gateway. TechCrunch reported on 21 October 2026 that Walmart had announced a gateway through which approved agents can search and buy. The same announcement added Agent Profiles in a customer’s account, with approval for each purchase or a spending limit. Walmart said agents that don’t comply will be blocked.

Forter’s network data, published on 23 October, tracks agents that load a store’s pages and click through them as a shopper would, Muse among them. Their share of agentic orders, the orders an AI agent places, rose from 1% in late July 2026 to 70% in the week of 12 October. Muse bought at 74% of the stores that received any agentic order. Forter has not published how many orders sit behind these shares.

In a Checkout.com survey of UK and US merchants, published on 9 June 2026, 3% of transactions involved AI agents. That was measured before Muse launched. The same survey found 89% of merchants preparing for agentic commerce, shopping in which an AI agent searches and builds a basket for a person.

AI agent traffic: what a company can and can’t see

Inside a company, a customer’s agent shows up in 3 places, and each one shows part of the picture.

• Analytics. Google Analytics added an AI Assistant channel on 13 May 2026. It labels a visit only when the visit comes from a recognised assistant’s link, such as ChatGPT, Gemini or Claude. Its release note says nothing about AI agents that browse a site themselves.
• Bot settings. The security team’s bot tools decide which automated visits get through. Akamai says 47.9% of the AI bot traffic on its network from July to December 2025 was in commerce. Training crawlers, which collect pages to train AI models, made up more than 70% of the AI bot activity it flagged in commerce. Over 90% of that commerce AI bot activity was set to monitor, which records it without blocking it.
• Phone lines. Invoca’s call data from July to November 2025 found Google’s AI pricing calls to businesses rose 162% month on month in October and 324% in November. Of those calls, 26% went unanswered, and 49% of the answered calls gave Google’s AI no price.

What a company can’t see from inside is what the agent got: the page, the price, the basket and the answer. Outside tests show the gap. In its 12 October 2026 snapshot of 77 stores, ARC, the Agentic Report Card, found all 5 of its test shoppers added a product to the basket at 32 stores. None did at 18.

Should you block AI agents?

A company has 3 choices for each agent: block it, allow it on terms, or check what it gets first. The largest platforms have already chosen differently.

• Block. Implicator reported that Amazon blocked Muse. eBay’s User Agreement, in force from 28 June 2026, prohibits buy for me agents and any flow that places orders without human review. Blocking suits a company whose terms, stock or fraud risk rule out orders no person has reviewed.
• Allow on terms. Walmart says approved agents can search and buy through its gateway and that agents that don’t comply will be blocked. Shopify lets shoppers connect their own agent. A Shopify store’s agents.md, a file that sets rules for AI agents, such as the one served on colourpop.com, tells agents to identify their requests as coming from an agent. For requests sent from a server or a hosted browser, that means signing them with Web Bot Auth, a signature that lets a website check which company’s agent is visiting. The file also tells agents to get the buyer’s approval before ordering and to shop through the store’s agent tools.
• Check first. Run the 10 checks, then decide agent by agent, with the results in front of the security, ecommerce and contact centre leads.

Common rules for allowing agents are being written. On 6 October 2026 Sierra and Meta announced the Personal Agent Protocol with Genesys, Instinct, Rocket, Shopify, Stripe and Walmart. They describe it as an open standard for how personal agents sign in to a business, start as guests and get read only or write access. Its first specification, version 0.1, is still to come.

A default setting can make the choice before anyone does. From 15 September 2026, Cloudflare blocks its Agent bot category by default on pages that show ads, for every new domain it takes on. Search engine crawlers, the bots that index pages for search, stay allowed. When Muse failed at Walmart, the retailer told TechCrunch the failures were unintentional and happened when a ‘verify you’re human’ step was interrupted. Zillow said its anti scraping defences can block some consumer agents.

Whichever choice a company makes, write it down for each agent and check that the settings carry it out.

Agent ready website checks and pass marks

Run each check from outside the company, logged out, on a phone and a laptop. Log every result in 1 sheet with the date, the agent used and what it got, and stop every test before payment.

1. Found: an assistant names the company

Ask the assistants and agents customers use for the company’s category in a customer’s words, for example “a waterproof jacket under £150 delivered by Friday”. Use ChatGPT, Gemini, Perplexity and Copilot, and Muse where it is available. Ask 3 times, in a new chat each time, because answers change between runs. A pass: the company is named in at least 2 of 3 runs, linked to its own site. The guide to what ChatGPT says about your store sets out a full scoring method.

2. Reached: the site loads for an identified agent

Ask an agent to open a product page and report its price and stock. Then ask a developer to request the same page with a user agent string, the name a visitor sends with each request, that declares an AI agent. In a test the proxy vendor Decodo published on 26 August 2026, Nike blocked requests declaring an AI agent in 5 of 5 trials. It served the same page to a browser. Blend Commerce’s audit of 200 UK and US retailers, run from 8 to 15 September 2026, found 29 refused its automated browser on 2 attempts. A pass: the page loads with its price and stock, with no block page and no ‘verify you’re human’ loop.

3. Answered: a question gets a usable answer

Ask an agent 3 questions a customer would ask about one top seller: its price, whether a size is in stock, and the returns policy for an opened item. Check each answer against the product and policy pages. Bill Hunt audited 122 leading Shopify stores and asked about returning used or open box items. Ecommerce Paradise reported the audit on 14 October 2026. 88 stores gave an answer, and 5 of the 88 answers were usable. A pass: every answer matches the page and gives the customer enough to decide.

4. Price: the agent sees the price on the page

Compare the price, currency, discount and delivery cost an agent reports with the product page at the same minute. Then compare both with the product feed the company sends to Google and other platforms. Google Merchant Center makes the same comparison: it checks the price in a merchant’s product data against the landing page and may disapprove a product when the two differ. A pass: the same price, currency and delivery cost everywhere, including during a sale.

5. Basket: the agent adds the right item

Give an agent a precise task: a named product, size and colour, quantity 2, added to the basket. A pass: the right variant, quantity and price are in the basket, and any welcome or sale code applies. In ARC’s snapshot, all 5 test shoppers added a product to the basket at 32 of 77 stores.

6. Checkout handoff: the agent reaches checkout

Ask the agent to go to checkout with the basket from check 5 and stop before payment. Some agents hand the customer a checkout link instead of opening the page. At 18 stores no ARC test shopper could add to the basket. At 13 of those 18, an agent that asked for a checkout link through a commerce protocol, a standard way for agents to request one, still got a link. ARC’s wider checks returned that link at 81.1% of 1,561 Shopify stores and 2.6% of 1,757 other stores, and ARC stresses that a link does not mean a completed order. A pass: the first checkout page, or a checkout link, opens with the basket intact, the right total and the delivery options a person sees.

7. Phone: a call from an AI is answered

Have an AI agent call the main number in opening hours and say at the start that it is an AI calling for a customer. It then asks for a price and a booking. Gemini’s Call for Me, where it is available, or a voice agent the company sets up for testing will do. Retail Insider reported on 17 August 2026 that Google’s AI says it is automated when it calls local businesses about prices and availability. Retail Insider adds that owners can opt out in their Business Profile settings. Pindrop detects AI callers for banks and other contact centres. It says weekly Muse call volume at its customers reached 8 times the first week’s by week 4, and 64% of Muse calls reached a human agent. A pass: the call is answered and the AI caller gets a price, a booking or a time when a person will call back. Nobody hangs up because the caller is an AI.

8. Chat: the support chat answers an AI

Open the support chat, say in the first message that it is an AI agent asking for a customer, then ask about an order’s status and the returns window. Meta says Muse resolves support questions on business websites, so a support chat may meet it. In the UK, the Competition and Markets Authority’s guidance of 9 March 2026 holds a business responsible for what its own AI agent does, as it is for an employee. A pass: the chat gives the AI the answer it gives a person, matches the policy page and offers a person when asked. The guide on how to test a chatbot has 8 questions and a scoring sheet.

9. Rules: the settings say what the company means

List every setting that decides what agents get. On the website, that is the bot categories in the security tool, robots.txt (the file that tells automated visitors which pages they may fetch) and agents.md. On the platform and the phone line, it is the store’s AI shopping settings, such as Shopify’s agentic storefronts, and the phone system’s rules for automated callers. Robots.txt alone may not stop an agent. Google’s documentation says tools that fetch a page because a user asked, such as Google-Agent, generally ignore robots.txt, and that a user agent string can be faked. A signature is a firmer check than a user agent string. Cloudflare says OpenAI, Google and AWS now sign their agents’ requests with Web Bot Auth, while Google’s own documentation says it is still experimenting with Web Bot Auth for Google-Agent. It also says daily requests from AI agents on its network grew more than 1,700% in a year. A pass: each setting matches a decision someone made and wrote down, and the signed agents the company chose to allow get through.

10. Owner: someone reads the results

Give each check an owner. The website, price and basket checks go to the ecommerce lead, the bot settings to security, the phone check to the contact centre lead and the chat to the support lead. A pass: every fail has a named owner and a fix date. Run the checks before your code freeze, the date after which your team stops changing the site for the sale, and again on the morning of Black Friday, Friday 27 November 2026.

Bot dashboards, AI visibility tools and audits compared

Obsession makes AI customers, so every other option is described from its provider’s own pages as of 27 October 2026.

Bot and fraud dashboards. Cloudflare sorts automated visitors into categories such as Agent, sets a rule for each, and verifies agents that sign their requests. Forter’s Agentic Commerce Engine, launched on 13 October 2026, includes an AI Activity Dashboard that shows merchants which AI agents transact on their site. HUMAN measures agent and bot traffic across its customers and publishes it in reports such as its 2026 benchmark.

AI visibility tools. Profound’s pricing page offers a free trial of 50 prompts over 7 days and an agency plan billed per client workspace. Its custom Enterprise plans cover up to 9 AI search tools, ChatGPT Shopping and agent analytics. OtterlyAI launched Agent Analytics on 13 August 2026. Its launch release says the tool reads server logs, the records a web server keeps of every request. It shows which AI agents and crawlers visit a site and which pages they request.

Agency audits. Mgroup sells a Shopify agentic commerce readiness audit for $3,000 over 2 to 3 weeks. Crasman sells an Agentic Readiness Audit from €1,250 plus VAT, and WebToffee gives its readiness audit free with its other services.

Doing nothing is reasonable for a company whose customers rarely send agents yet. In that case, run the 10 checks by hand once a quarter and watch the share of orders placed by agents.

Limits and next steps

• A test by hand covers 1 agent, 1 country and 1 day. Agents change month to month, so repeat the checks after each platform announcement and each change to bot settings.
• Every test stops before payment. To see a full order, place 1 real order through each agent the company allows, then refund it.
• Tests say what they are. Tell the security and contact centre teams before a test runs, and have every AI caller and chat test say it is an AI, so staff can tell tests from customers.

Obsession’s Voice agent check calls your AI receptionist every morning as new customers, books a slot, asks a price and a person, and checks every answer. Its Support bot check asks your support bot what your customers ask, on every channel, every day, and checks each answer against your policy.

Questions about agent ready websites