AI is moving from answers to actions. Your next growth curve won’t come from adding another channel, it will come from making your brand citable, parsable, and callable by AI agents. That means: machine-legible facts (Schema.org), safe action endpoints (ACP and Agent Payments Protocol [AP2]), and clean, trackable hand-offs to checkout or booking. Do this once and every new assistant whether in ChatGPT Atlas or Google AI Mode can work with you automatically.

Why this matters now 

We’ve entered the agent economy. Rather than traditional search journeys with dozens of page views, assistants will increasingly choose and often transact for customers. You’re already seeing the signals: 

• Google AI Mode for Dining takes prompts like “Italian, 4 people, tomorrow 7pm, €€” and surfaces live tables via major partners linking directly to the booking flow. 

• Google is also trialling agentic shopping: real-time product discovery, price-tracking and “buy-for-me” flows with Google Pay in the U.S. 

• OpenAI’s ChatGPT Atlas introduces an agentic browser where users can seamlessly move from discovery to purchase inside chat. 

• Underpinning this shift are open standards: 

• ACP (Agentic Commerce Protocol) allows agents to complete purchases directly with merchants. (openai.com) 

• AP2 (Agent Payments Protocol) from Google defines how agents, users and payment providers coordinate trust and transactions at scale. (orium.com)

The pragmatic takeaway

Treat AI assistants as a new performance surface. Brands that win will be: 

• Citable — your facts are consistent and verifiable. 

• Parsable — your site is structured for both humans and machines. 

• Callable — agents can trigger bookings, checkouts or reservations. 

• Measurable — you can trace the path from agent referral to transaction. 

What to do in the next 90 days 

1. Facts as data, not prose.

Ship Product + Offer JSON-LD on every relevant page complete with price, VAT, availability, variants, delivery or booking policies. Regularly validate it. 

2. Action endpoints, not just CTAs. 

Expose ACP-compatible product feeds and checkout endpoints. Prepare for AP2-style authorization (mandates, tokens, audit logs). While retail flows are live, travel/booking flows remain emerging. 

3. Surface readiness: Atlas + AI Mode. 

Run real-world tasks in ChatGPT Atlas and Google AI Mode. For each use-case, test: can an agent find the correct item/service, confirm availability, and land on a pre-filled booking or checkout step? 

4. Accessibility and provenance. 

Ensure semantic HTML, clear headings, ARIA roles and transcripts. Mirror key policies (returns, cancellations) in structured data and maintain transparent provenance of assets. 

5. Crawler governance. 

Define which bots you allow (or disallow). Maintain a clear robots.txt policy for AI crawlers and monitor them monthly. 

6. Attribution hygiene. 

Use distinct UTMs/referral tags for assistant-origin traffic (e.g., utm_source=atlas|ai_mode|acp). Ensure you can measure assistant-driven revenue, not just clicks.

What good looks like 

• Complete JSON-LD on PDPs, booking pages and experiences. 

• Agent-friendly deep-links with pre-filled booking or checkout parameters. 

• ACP endpoints live for retail merchants; travel flows staged for readiness. 

• Task tests pass in Atlas + AI Mode for your high-value SKUs or tour offerings. 

• Policies (shipping, returns, cancellation) mirrored in structured data. 

• AI robots policy and assistant UTMs implemented. 

• Quarterly “agent-readiness” audit scheduled. 

The platform context 

OpenAI’s domain still leads GenAI site visits, but Gemini is rising and domain traffic is only part of the story: Google’s embedded AI surfaces (AI Overviews, AI Mode) are still under-counted. The strategic takeaway: optimise for both ecosystems. 

• Domain-level share (Oct 2025): ChatGPT ~ 74%, Gemini ~ 14%. 

• AI referrals (June 2025): ~1.13 billion visits (+357% YoY) from AI platforms to top 1,000 sites. 

“AI isn’t another channel, it’s an agent layer. As Atlas and Google’s AI Mode compress discovery into action, we win by being citable, parsable and callable: machine-legible facts and a checkout agents can trust.” 

Common pitfalls 

• Great copy, missing data. Beautifully written pages but stale or incomplete JSON-LD. 

• Dead-end hand-offs. Agents find the right product/slot but you link to the homepage or open basket with zero pre-fill. 

• Policy ambiguity. Returns, cancellations or delivery windows missing from data, making agents prefer a peer with clearer terms. 

• Attribution blind spots. Assistant-driven purchases showing as “Direct” rather than attributed to the correct surface. 

Practical Examples: Retail & Travel (With Updated Nuance) 

Example 1 – Retail (Ecommerce: Fashion/Sneakers) 

Scenario 

A footwear brand wants agents to find the right size and colour of a flagship trainer, confirm stock and delivery, and complete purchase seamlessly. 

Facts to publish 

• Product JSON-LD: name, brand, sku, gtin, image, size, colour, material. 

• Offer JSON-LD: price, priceCurrency, availability, eligibleRegion, shippingDetails, url. 

• Policies: returns, exchanges, warranty, delivery cut-offs visible and structured. 

Callable actions 

• Deep-link checkout: 

 /checkout?sku=TR-ALPHA-RED-UK9&qty=1&size=UK9&colour=red 

• ACP integration: product feed + POST /checkout_sessions endpoints for in-chat checkout. 

• Payment readiness: delegated token, idempotency, audit log. 

Acceptance test 

 “Buy Alpha Runner, red, UK 9, deliver to Ireland, under €140.” 

 → Agent returns the correct variant, confirms stock, gives delivery ETA, opens pre-filled checkout. 

MVP vs Level-Up 

• MVP: Product + Offer JSON-LD, deep-link checkout, clear returns/shipping data. 

• Level-Up: full ACP checkout, AP2-ready payment mandates, price-drop alerts, size-fit guidance. 

Example 2 – Travel & Attractions (Tours / Experiences) (Emerging Use-Case) 

Scenario 

A coastal kayak tour operator wants agents to find real-time availability, book 2 adults for Saturday 10:00, and upsell a photo package. 

Facts to publish 

• Itinerary (descriptive): TouristTrip with name, description, itinerary, and touristType. 

• Bookable instances: Event with startDate, duration, location as Place; Offer with price, availability, validFrom/validThrough, url. 

• Policies: cancellation window, weather contingencies, cut-off times, what to bring—visible + structured. 

Callable actions 

• Deep-link booking: 

 /booking?productId=KAYAK-COAST&date=2025-07-05&time=10:00&adults=2&addon=PHOTO 

• Future-ready checkout: ACP-compatible product feed & checkout endpoints (once slot/booking logic is supported) 

• Payment readiness: delegate token, idempotency, audit trail. 

Acceptance test 

 “Book 2 adults for Saturday 10:00 on Coastal Kayak Tour, add photo package, show free cancellation ≥48 h.” 

 → Agent picks correct slot, confirms availability, shows total + add-on, lands on pre-filled booking (AI Mode) or in-chat checkout (when supported). 

MVP vs Level-Up 

• MVP: accurate Event + Offer markup for top departures, deep-link booking, clear cancellation policy. 

• Level-Up: full ACP checkout for bookings, AP2-ready mandates, multi-slot bundles (e.g., “sunset + photos”), proactive re-booking when weather affects plan. 

Quick Playbooks 

Retail Playbook 

1. Map your top 50 SKUs → ensure variant-level JSON-LD. 

2. Implement pre-filled checkout links. 

3. Ship ACP Product Feed + checkout endpoints. 

4. Test in ChatGPT Atlas + Google AI Mode. 

5. Add assistant UTMs and revenue dashboards. 

6. Audit quarterly. 

Travel Playbook 

1. Define tour schema: TouristTrip + Event + Offer. 

2. Publish cancellation and cut-off rules in both visible copy and structured data. 

3. Implement deep-link booking. 

4. Prepare for ACP-compatible checkout (when slot logic is fully supported). 

5. Test in AI surfaces and simulate bookings for your highest-value tours. 

6. Track assistant-bookings, add-on rate and re-booking flows. Audit quarterly. 

Final thought 

This isn’t about chasing the latest feature. It’s about building durable foundations clean facts, callable actions and trustworthy hand-offs so whichever assistant your customer uses, you’re the easy choice. 

If you’d like us to run an Agent-Readiness Audit on your site, drop us a message ([email protected]), let’s deliver a punch list and roadmap you can act on this quarter.