--- title: "Can an AI agent book a demo or request a quote on your website?" description: "Audit the complete lead form journey, from discoverable fields and validation to consent, CRM delivery, confirmation and duplicate prevention." canonical: "https://getdedica.com/blog/can-ai-agents-book-a-demo-or-request-a-quote" last-updated: "2026-10-30" --- # Can an AI agent book a demo or request a quote on your website? A “Book a demo” button is easy to find in a screenshot. It is harder to know whether an agent can complete the form without inventing a phone number, selecting the wrong company size or submitting the request twice. For a B2B site, this is not a cosmetic test. If a form represents an inbound lead, the quality of the data and the person's consent matter. The objective is not to let an agent create as many leads as possible. It is to help an interested visitor reach the right next step while keeping the request accurate and intentional. ## Trace the whole form journey Begin at the page where a visitor would naturally arrive: a product page, pricing page or campaign landing page. Ask an agent to find the right route and prepare a request for a realistic scenario. Then check each transition. **Can it identify the form?** Give each field a visible label that describes the information you need. “Email” is clearer than “Your details.” If the form asks for work email, state that. Avoid relying on placeholder text alone, since it disappears after entry and can be interpreted inconsistently. **Can it understand conditional questions?** Some forms reveal extra fields based on company size, region, product or project timing. Test at least one path where a follow-up field appears and one where it does not. Confirm the required state changes with the selection and that hidden fields are not accidentally submitted. **Can it recover from validation?** Submit an intentionally incomplete test request. The page should say which field needs attention and preserve the values the visitor already entered. A generic error at the top of a long page can send both people and agents searching for the problem. **Does the visitor understand what will be sent?** A quote request may include company information, work email, budget or a description of a project. Explain which details go to your team and what happens next. Keep a human-visible review step before sending the form. **Does the request arrive once?** Submit a test lead, check the CRM and confirmation email, then retry after a simulated timeout. The user should not create two opportunities because the first confirmation was delayed. Use an idempotency key or equivalent duplicate handling where the underlying API supports it. ## Separate page controls from the lead system A browser agent may be able to fill a form but cannot prove that the lead reached your CRM. The page could show a success state after a failed network request, or a third-party widget could silently drop a submission. Run a test all the way through the server response, lead routing and confirmation. If you are experimenting with WebMCP, its declarative API can expose an HTML form as a browser tool. Chrome's documentation allows a form to remain visible for the person to submit manually, or to use `toolautosubmit` for agent-triggered submission. For a lead form, start with manual submission. That gives the visitor a chance to correct the record and makes the disclosure clear. Review [Chrome's Declarative API guide](https://developer.chrome.com/docs/ai/webmcp/declarative-api) and [tool security guidance](https://developer.chrome.com/docs/ai/webmcp/secure-tools); WebMCP remains an early browser feature, not a capability every agent supports. If you are not implementing WebMCP, still test the ordinary HTML form with a supported browser agent and an assistive technology check. Native labels, sensible input types, visible errors and keyboard support help people too. A tool schema does not compensate for a confusing form. ## Use a small test matrix Run the same journey with a valid request, a missing required field, an invalid email, a conditional question, a slow network and a server error. For each run, record whether the visitor can understand the state, correct the issue and avoid duplicate submission. Verify what the CRM receives and whether the confirmation message matches it. Do not test by sending fabricated leads to a production sales team. Use a sandbox CRM or a clearly labeled internal test record, then verify that test data cannot trigger the normal sales sequence. ## Measure completion without confusing it with agent visibility Track the start of the form, validation errors, successful server acceptance and downstream lead creation separately. A click on the button is not a completed lead. A tool invocation is not necessarily a conversion. Compare the agent journey with a human baseline and look for the exact step where completion diverges. Dedica's [free agent-readiness scan](/) checks public site signals that can prevent a browser from reaching or understanding the lead form. It cannot validate your CRM delivery or consent design. Add a test submission to your own release checklist. If you are experimenting with a browser tool, the [React and Next.js WebMCP example](/blog/build-a-safe-webmcp-tool-in-react-nextjs) shows a draft-first form pattern. ## Sources - [WebMCP Declarative API](https://developer.chrome.com/docs/ai/webmcp/declarative-api), Chrome for Developers. - [WebMCP secure tools](https://developer.chrome.com/docs/ai/webmcp/secure-tools), Chrome for Developers. - [WebMCP](https://developer.chrome.com/docs/ai/webmcp), Chrome for Developers.