AI Search's Next Move: Trust, Then Transactions

Quick read: In July 2026, Search Engine Land published pieces extending AI-citation coverage into two stages beyond "cited" and "recommended": trust and transaction. On the trust side, a Search Engine Land analysis by Journey Further found that 93% of AI search citations across top commercial sectors come from third-party sources rather than a brand's own pages (Search Engine Land). On the transaction side, Google introduced Universal Cart at its I/O 2026 conference in May 2026 — a shared shopping cart that works across Search, Gemini, YouTube, and Gmail, built on Google's Universal Commerce Protocol (UCP) — alongside the Agent Payments Protocol (AP2), which lets a person set spending guardrails an AI agent must follow before it can complete a purchase, rolling out in the coming months starting with Gemini Spark (Google). The same underlying question — can an AI agent complete an action on your business's behalf without a human double-checking every step — applies to service bookings, not just retail carts.

Search Engine Land ran back-to-back pieces this month on what makes AI search engines trust a brand enough to name it, and, separately, what it takes for an AI agent to actually complete a transaction with it. Google already shipped the infrastructure for that second part back in May — a cart and a payment protocol built so an AI agent can finish a purchase, not just describe one. Getting an AI engine to mention your business, it turns out, was never the finish line. It's step two of five.

I've written before about the gap between being cited by an AI answer and actually being recommended by one — that gap is real, and it's measurable. What's new is that the industry stopped stopping at "recommended." The conversation has moved to what happens after: does the AI system trust you enough to pick you over the other names it considered, and can it actually finish the job — book you, buy from you, add you to a cart — without handing the whole thing back to a human. That second part used to be science fiction. It isn't anymore.

The five stages, in order

Here's the ladder as it's being described right now, stitched together from a week of trade coverage on the same theme:

1. Used. An AI system pulls information from your site into an answer without naming you or linking to you. You did the work; you got no credit.
2. Cited. The AI system names your page as a source — a link, an attribution, a "according to."
3. Recommended. The AI system actually tells the person asking "go here." This is the step most local businesses assume citation guarantees and it doesn't — a page can get pulled in as a source and still lose the recommendation to a competitor.
4. Trusted. The AI system picks you consistently, across repeated questions, because independent sources back up what you say about yourself.
5. Transacted. The AI system — or an agent acting on someone's behalf — completes an action with you directly: books the appointment, places the order, adds you to a cart.

Getting cited is the foundation of that ladder — none of the later stages happen for a business that never clears the first one. But foundation isn't finish, and this month's trade coverage is specifically about the two rungs most small business owners haven't heard mentioned yet: trust and transaction.

Why this matters even if you don't sell anything online

If you run a plumbing company or a dog groomer or a bakery, "agentic commerce" sounds like a retail problem — carts, checkout, Shopify integrations, none of it yours. Google's Universal Cart, announced at its I/O 2026 conference in May, is explicitly a shopping tool: a cart that follows a person across Search, Gemini, YouTube, and Gmail, with checkout support from launch partners like Target, Walmart, and Sephora (Google). None of that is aimed at a one-location service business.

But the mechanism underneath it isn't retail-specific. Universal Cart runs on the Universal Commerce Protocol, a common language that lets an AI agent understand what's for sale and complete a transaction on the spot instead of just describing it. Paired with the Agent Payments Protocol — which lets someone tell their agent "these criteria, this budget, go" and the agent only acts once those conditions are met — the stack is solving one problem: can an AI agent finish something on a business's behalf without a person manually taking over halfway through.

A booking form is the same problem in a different outfit. If someone asks an AI assistant to "find a groomer near me who can take my dog Saturday morning and book it," the agent needs what a shopping agent needs: current availability, clear pricing, a request flow it can actually complete without getting stuck on a form that emails an inbox nobody checks until Monday. The protocol names are new. The requirement — be legible and completable by a machine acting on a person's behalf — isn't, and it doesn't go away just because your business doesn't have a "buy now" button.

I ran a dog grooming salon for years before any of this existed, and I know exactly how much bandwidth an owner has left at the end of a Tuesday to think about "agentic commerce protocols": none. Nobody's asking you to build an AP2 integration this week. What's worth understanding is the direction this is moving, so the next protocol name in a headline doesn't feel like it came from nowhere.

The mechanics: what actually builds trust and transaction-readiness

Third-party corroboration does most of the trust work. Search Engine Land's Journey Further analysis found that across top commercial sectors, 93% of AI search citations pull from third-party sources — reviews, write-ups, mentions — rather than a brand's own website (Search Engine Land). That tracks with what's already true for AI search generally: a business's own claims about itself carry less weight than what other people say about it. The fastest lever isn't rewriting your homepage — it's getting more real, specific, independent references to your business out in the world.

Your booking or contact flow needs to be something a machine can complete, not just read. Stating hours, services, and pricing clearly is necessary but not sufficient anymore. The next question is whether the actual request — book, quote, order — can be finished without a human intervening at every step. A contact form that silently fails, a booking widget that requires seventeen clicks, a phone number as the only path to schedule anything — those are the small-business equivalent of a checkout flow an AI agent can't parse.

Consistency across sources matters more than polish on any one source. UCP and AP2 both exist because agents need to trust the data they're acting on before they'll commit to it. The same name, address, and phone number everywhere your business appears is easier for a system to act on confidently than a business whose Facebook page says something different from its website.

What won't help

• Buying visibility in AI chat tools while your booking flow is broken. The transaction layer is where the real shift is happening, not ad placement inside a chatbot. If an agent can't complete an action with you, being mentioned by name doesn't close the gap.
• Treating "trust" as a setting to switch on. It's accumulated third-party corroboration, not a checkbox. Three fresh, specific reviews and a decent local write-up do more for it than a polished "why choose us" page nobody else wrote.
• Waiting for "agentic commerce" to obviously apply to your category before doing anything. The infrastructure is being built for retail first, but the underlying requirement — legible and completable to a machine acting for a person — reaches service businesses on the same timeline, just with less press coverage.
• Chasing the jargon instead of the plumbing. Nobody needs to understand AP2's technical spec to benefit from it. What matters is the boring stuff underneath: working forms, current hours, real reviews, consistent facts.

FAQ

What are the five stages of AI search visibility?

Used, cited, recommended, trusted, and transacted. A business can be used (pulled into an answer with no credit) or cited (named as a source) without ever being recommended (actually suggested to the person asking), and recommended without being trusted enough to be picked consistently or transacted with directly.

What's the difference between being recommended and being trusted by AI search?

Being recommended is a single AI system naming your business as the answer to one question. Being trusted is that happening consistently, across repeated and varied questions, because independent sources back up what you claim about yourself. One is an event; the other is a pattern.

What is Google's Universal Cart, and does it apply to a service business?

Universal Cart is a shopping cart Google introduced at its I/O 2026 conference in May that follows a person across Search, Gemini, YouTube, and Gmail, built on the Universal Commerce Protocol so AI agents can complete purchases directly (Google). It's explicitly retail-focused at launch. The underlying requirement — being legible and actionable to an AI agent — applies just as much to a booking or quote request as it does to a product purchase, even though the specific tooling isn't built for service businesses yet.

What is the Agent Payments Protocol (AP2)?

AP2 is Google's protocol for letting an AI agent make a payment on a person's behalf within guardrails that person sets in advance — specific brands, products, and a spending limit — so the agent only acts once those conditions are met. Google says it will begin rolling AP2 into its own products in the coming months, starting with Gemini Spark (Google).

Does any of this apply if I don't sell products online?

Yes, indirectly. The retail tooling is built for e-commerce first, but the mechanism it runs on — an AI agent completing an action instead of just describing a business — applies to appointment booking and quote requests too. A booking flow an agent can't complete cleanly is the service-business version of a checkout it can't finish.

What actually builds "trust" with AI search systems, in practical terms?

Third-party corroboration, mostly. A Search Engine Land analysis found 93% of AI search citations across top sectors come from third-party sources rather than brand-owned pages — reviews, write-ups, and independent mentions do more of this work than your own website's copy.

What should I actually do about this right now?

Nothing exotic. Keep hours, pricing, and services current and consistent everywhere they appear, make sure a booking or quote request can actually be completed without hitting a dead end, and keep collecting real, specific reviews. Same foundational work AEO has always required — this news just shows where it's headed next.

---

Keeping your business information consistent, your booking flow completable, and your third-party signals current isn't a one-time project — it's the kind of ongoing work a Get Found on Google and AI engagement tracks month to month, so you're not the last business in your category still asking a person to double-check a form nobody reads until Monday.