Our guides explain how shopping with AI agents works using what vendors, protocol maintainers, retailers and regulators publish. We do not place test orders or publish scores from use. Instead, we read the primary sources, cross-check them, and date every claim.

What sources do we use?

We rely on primary sources first, in roughly this order:

  1. Vendor documentation and help centres, such as Anthropic's Claude Duet safety article, Amazon's shopping assistant how-to page and Google's agentic checkout announcement.
  2. Protocol specifications and developer docs, such as OpenAI's product feed spec, the ACP repository, the AP2 site, ucp.dev and Stripe's agentic commerce docs.
  3. Retailer and platform statements, including published agent terms, robots.txt files (checked on a stated date) and announcements such as Shopify's.
  4. Regulator and consumer-protection pages, such as the CFPB and FTC on card disputes.
  5. Reputable press, mainly for incidents, disputes and lawsuits, attributed by name.

Social media and forums are used only to find leads that we then try to confirm elsewhere.

  1. CollectGather vendor docs, specs, retailer policies and regulator pages.
  2. Cross-checkCompare against independent press coverage.
  3. Flag conflictsWhere sources disagree, give both versions and say which looks more reliable.
  4. DateMark prices, partner lists and availability "as of" the month checked.
  5. ReviseUpdate pages when a vendor, retailer or protocol changes.

What do we not claim?

We have not tested purchases with these agents. We do not write "we tested", give star ratings or describe personal experiences. When we explain how a checkout works, we are describing what the vendor documents, and we say so. If a vendor has not published something, such as whether a price-triggered auto-buy asks for confirmation, we say it is not documented rather than guessing.

How do we handle conflicts and gaps?

Some facts in this field are genuinely disputed. Reports give three different dates in March 2026 for when OpenAI scaled back Instant Checkout, and parts of the Amazon v. Perplexity appeal history are reported but not confirmed in court records we could access. In those cases we give the range, attribute each claim and avoid stating the uncertain part as fact.

Shopping versus safety

We cover the practical side of buying: which agent, which store, how checkout and returns work, and how merchants integrate. Risks such as scams, unauthorised purchases and liability are covered in depth by our sister site Safe AI Agents, and we link there rather than repeating it.

How often do we update?

Partner lists, protocol versions and store policies change often. We review pages when a vendor or retailer announces a change, and each article shows its last updated date. Read more about this site or our privacy policy.