AI assistants are booking for customers: is your website ready to let them in?

AI assistants such as ChatGPT and Claude are starting to make enquiries and bookings on people's behalf, and large companies are split on whether to let them in. For a smaller business, that decision is often being made by a bot-protection or firewall setting nobody has looked at. The practical response is to test your own booking and enquiry pages, decide on purpose, and mark AI-sourced leads in your CRM before any automated follow-up goes out.

What has actually happened in the last fortnight?

Two strands of news have pulled in opposite directions.

On the blocking side, a Hotel-Online report summarises coverage by The Verge on 17 October: agents from OpenAI, Meta, xAI and Instinct were being blocked on major sites. Its examples were Delta stopping agents from booking tickets, and Yelp only letting in agents that pay for its data.

On the welcoming side, Wyndham used its third-quarter earnings call on 22 October to describe AI booking agents as a new type of guest. According to Skift, its CEO said the group feeds room rates and availability straight to AI assistants, and that its booking engine is live in Google's AI Mode, ChatGPT and Claude.

So customers are beginning to hand over the job of "find me a slot and book it". The open question is whether your website lets that happen.

Why do some websites end up blocking AI assistants by accident?

On 18 October, independent live tests pointed Claude and ChatGPT at the websites of ten large hotel groups. The agents reached the checkout page at five or six of the ten, and Marriott, IHG, Accor and Choice let both agents through. Part of that research was produced with ChatGPT, so treat it as one indicative snapshot rather than a settled finding.

The more useful detail is in the second part of the tests. Bot-protection services, not booking policy, often decided who got in. At Wyndham, an agent's request was dropped one day and worked the next.

That is the bit a small business should take seriously. If a company actively courting AI agents can still have them turned away by its own security layer, a salon, clinic, trades firm or events venue can easily be in the same position. Hosting platforms, firewalls and anti-spam tools on enquiry forms are designed to stop automated traffic. An AI assistant acting for a genuine customer is automated traffic. Whether you want it is a business decision, and right now it may be a side effect of a default setting.

What is Wyndham doing that a smaller business could copy?

The important part of Wyndham's approach isn't the scale. It's the shape. AI assistants can see what's on offer and when it's available, but its ChatGPT app sends guests to wyndhamhotels.com to finish the booking. The final confirmation and payment stay on its own website. Skift also reports that Morgan Stanley expects AI agents could cost less per booking than online travel agents do.

A small business can follow the same principle in a simpler way:

• Make your offer readable. Services, prices, durations, locations and opening hours set out clearly on the page, not buried in images or PDFs.
• Keep booking flows straightforward. Fewer steps, clear labels and no unnecessary pop-ups help human customers and software alike.
• Keep the commitment on your site. Let an assistant gather details or check a slot, but confirm the booking and take any payment through your own system, with a confirmation sent to the real customer.

If you run your own booking system, there's a further step. Conduit's explainer notes that ChatGPT apps are built on the Model Context Protocol (MCP). The same technology can give AI assistants a controlled, read-only view of your availability, without handing over the keys to your booking database. That's the kind of connection we build as part of our MCP servers and AI integrations work.

How are the big CRM vendors changing lead follow-up at the same time?

While assistants start arriving at the front door, CRM vendors are automating what happens after the enquiry lands.

A Microsoft roadmap entry (ID 573700), mirrored here, announces a Sales Development Agent in Microsoft 365 Copilot. It's described as an autonomous agent that researches, qualifies and engages with leads, with a preview due in October 2026 and full release in January 2027. Because so many UK firms already pay for Microsoft 365, automated lead follow-up is about to appear inside tools they already have, as this admin write-up covers.

Salesforce's Dreamforce 2026 announcements include AIforce, which brings Salesforce data and workflows into Claude and Slack using the permissions you've already set, plus reusable AI Skills and A/B testing to compare versions of an agent. A partner write-up adds that agents can now work towards a goal over days or weeks, not just one session.

Put those together and a new situation appears: an AI assistant makes an enquiry for a customer, and an AI agent in your CRM starts nurturing that "lead" automatically. Nobody has checked that a real person is on the other end, or that the details were filled in correctly.

How should AI-sourced enquiries be handled in your CRM?

Treat them as a distinct source, not a mystery.

• Tag the source. Add an option such as "AI assistant" to your lead-source field, and to any "How did you hear about us?" question on your forms.
• Confirm with the person. Send an immediate confirmation email to the customer so they can correct anything the assistant got wrong.
• Hold back the sequence. Don't let automated nurture emails start until the enquiry has been confirmed or reviewed.
• Measure it. If your CRM supports testing different follow-ups, use it. Salesforce's move towards A/B testing agents is a reminder that automated follow-up should be checked against results, not trusted by default.

Sorting and checking incoming enquiries is a good fit for AI workflow automation, as long as a human stays in the loop for anything unclear.

What should you do next?

Start small and practical:

1. Test it yourself. Ask ChatGPT or Claude, in their agent or browsing mode, to make an enquiry or booking on your own website. Note exactly where it gets stuck.
2. Check your bot protection and firewall rules. Find out what your hosting, security tools and form spam filters do with automated visitors, then decide on purpose which pages AI agents may reach.
3. Make services, prices and availability easy to read. Keep the final confirmation and any payment on your own site.
4. Tag AI-sourced leads in your CRM and review them before automated follow-ups go out.
5. Consider a direct connection if you run a bespoke booking system, so assistants can check availability through a channel you control.

If your booking or enquiry flow needs reworking to cope, that's core web app development territory, and it's best done in short steps you can test as you go.