The agent arrives on the phone

The machine customer was supposed to arrive as an autonomous shopper with a wallet. For most small businesses it arrived as a phone call. Google's search assistant can call local businesses on a searcher's behalf to ask about prices and availability, and Google gives owners a setting in their business listing to allow or refuse those automated calls and texts.

It is a modest capability and an important precedent. The agent is not buying anything. It is gathering the facts a person will use to choose, and the business is answering software it never invited. The quality of that answer now helps decide whether the business makes the shortlist.

Checkout is standards work, not yet a channel

The purchasing layer is real but early. An Agentic Commerce Protocol stewarded by OpenAI and Stripe is in beta. A Universal Commerce Protocol co-developed by Google, Shopify, and large retailers defines checkout, identity-linking, and order capabilities that a business can publish for agents. Visa has published a protocol that lets merchants verify an agent's identity cryptographically, and Stripe's agent wallet requires the person to approve every purchase.

Volume has not followed the protocols. The most visible in-chat checkout was reportedly scaled back in early 2026 after weak conversion, according to trade press. A small business should treat agent checkout as a pilot channel its commerce platform may switch on, not as a revenue forecast.

What the machine customer needs from you

Every one of these agents needs what a careful human buyer needs, in a form it can read: a current price or price range, hours, service area, availability, policies, and—for a store—product data with identifiers, stock, shipping, and returns. When those facts differ between the website, the business listing, and the person answering the phone, the agent repeats the inconsistency as fact.

The low-regret work is unglamorous. Name one owner-approved source for each fact and keep every other surface in agreement with it. Check that robots rules and bot protection do not block the assistants you want to be described by; assistant makers separate search crawlers from training crawlers, so a business can refuse training without disappearing from answers. For a store, complete the product data the commerce platform already exposes to agents before worrying about protocols.

Answer the machine the way you answer a person

Write a short script for automated callers: what the business will quote, what it will not, and when it asks the customer to call back directly. An AI caller asking for a price is a lead with a strong intent signal. Answer it courteously and consistently, and log it like any other inquiry so the pattern becomes visible.

The same discipline runs in the other direction. If the business uses its own AI voice to call customers back, a 2024 FCC ruling treats AI-generated voices as artificial voices under federal telemarketing law, so those calls need prior consent. The arrival of machine callers is not a license to become one.

The loop

Goal: be described accurately wherever buyers and their agents ask. Actor: the owner or office manager, with a monthly research routine. Tools and context: the website, the business listing, the product catalog, and the script for automated callers. Scoreboard: facts answered correctly across ten buyer questions in three assistants, automated calls logged, and inquiries that name an assistant as the source. Stop condition: a month with no inconsistencies. Human gate: every published price and policy, and the decision to allow automated calls or join an agent checkout channel.

Nothing in that loop requires a new product. It requires one person to own the business's facts.

Agent commerce is not starting with robots spending money. It is starting with software asking questions on a person's behalf.

The businesses that answer those questions accurately will be the ones the person hears about.

sources
  1. Manage automated calls and texts from GoogleGoogle Business Profile Help, accessed September 29, 2026
  2. Deep search and business calling in Google SearchGoogle, accessed September 29, 2026
  3. Agentic Commerce ProtocolOpenAI and Stripe (GitHub), accessed September 29, 2026
  4. Universal Commerce ProtocolUCP contributors (GitHub), accessed September 29, 2026
  5. Trusted Agent ProtocolVisa (GitHub), accessed September 29, 2026
  6. Link CLIStripe (GitHub), accessed September 29, 2026
  7. OpenAI shifts checkout plans in its agentic commerce strategyDigital Commerce 360, accessed September 29, 2026
  8. Does Anthropic crawl data from the web, and how can site owners block the crawler?Anthropic, accessed September 29, 2026
  9. FCC confirms that TCPA applies to AI technologies that generate human voicesFederal Communications Commission, accessed September 29, 2026