Skip to content
A global perspective on digital financeAbout The Token Press
CRYPTO. FINANCE. WHAT’S NEXT.The Token PressYour daily briefing ↗

Beginner · 6 min read

Why Does My AI Website Form Say Sent but Send No Email?

Trace a contact form from button click to saved submission and email delivery. Diagnose false success messages with a fictional log exercise and a focused AI repair prompt.

The Token Press education desk · Updated

Learning path: Build with AI: ChatGPT, Astra, Gemini and a Practical Learning Path →

A success banner is only the first clue

An AI-built contact form can look complete while its button only changes the screen. The useful question is where the submission stopped: in the browser, at the form handler, in a saved record, at the email provider or inside the recipient's mail system. Check those boundaries in order before asking AI to rebuild the whole page.

This guide uses a fictional workshop enquiry form. Its intended behavior is to save one enquiry and then notify the organizer. A saved enquiry remains available if the notification fails. Other designs may send email without keeping a submission record; document that choice because the evidence available for troubleshooting will differ.

Our website-building guide covers the broader launch process. Here, the task is narrower: prove that one submission followed the intended route and make the success message describe only what the system has established.

Identify the real handler before changing code

Ask the builder to identify the form component, submission handler, destination endpoint and deployment configuration. A handler is the code or service that accepts the request. If the only behavior is a timer, an alert or a local success flag, there is no demonstrated delivery route. Keep the form labeled as a demo until that route exists.

A managed form service and a custom server endpoint need different integration steps. In Netlify Forms, detection must be enabled and the deployed HTML must expose the form. A form rendered only in JavaScript needs the documented HTML detection counterpart. Its AJAX submissions use the supported form encoding; the setup documentation currently says JSON form data is unsupported. These are Netlify requirements, not rules for every host.

For a custom endpoint, verify the actual deployed route, accepted fields and response format with its own implementation or documentation. A development preview may have a local handler that the production host does not run. Do not assume an AI tool's working preview proves that production has the same backend.

Before choosing a service, check its current plan limits, where submissions are retained, who can read them and how they are deleted. A visually free form can still depend on paid processing or email usage. This guide does not assume a particular subscription or promise a zero-cost backend.

Inspect one request and its actual response

On a site you control, open the browser's developer tools and Network panel. Submit one clearly marked test using fictional message content and an inbox you own. Look for the request generated at that moment. Record the endpoint, method, HTTP status, response type and a non-sensitive request reference. Avoid copying cookies, authorization headers or the full form contents into an AI chat.

No request suggests that the front-end handler never reached the network, although you should first rule out browser validation or a disabled button. A request to the wrong route points toward wiring or deployment. A response containing the website's HTML instead of the expected acknowledgement can indicate a fallback page, even if its status is 200.

MDN documents a common JavaScript trap: fetch() does not reject merely because the server returned an HTTP error such as 404. The handler must inspect response.ok or the status. Then it must check the expected response body. Neither a fulfilled promise nor a generic 200 proves that an enquiry was stored or an email was delivered.

Define the acknowledgement contract with the backend. In our fictional design, success means a validated enquiry was durably saved and the response contains its reference. The browser can then say 'Enquiry received' and show that reference. It should not claim the organizer read the message.

Separate the submission record from its notification

Look for the test enquiry in the intended submission store using its reference and time. If it exists, investigate the notification stage without asking the visitor to submit the whole enquiry again. If it does not exist, inspect the handler's error record before treating an inbox search as the main diagnostic.

Netlify configures form notifications separately for verified submissions. Its current documentation restricts email recipients to active team members or eligible outside recipients who have confirmed receipt; the outside-recipient confirmation feature is rolling out gradually. Check that the intended notification was actually saved and enabled for the right form. A captured submission alone does not establish that this configuration is complete.

For a custom email provider, find the accepted send request and delivery event. Resend distinguishes provider processing from acceptance by the recipient's mail server. A delivered event does not prove inbox placement or that a person read the email. A bounce means rejection; a delayed event means delivery has been deferred. Use the actual event and reason before changing the form.

Resend's July 2026 update exposes Message-ID in email events and retrieval responses. That mail identifier can help correlate the message with recipient-server logs. Keep it distinct from your enquiry reference and the provider's API email ID. Each identifies a different part of the journey.

Worked example: three tests, three different faults

These records are invented and contain no real customer data. Test FORM-A at 09:00 shows a green banner, but the Network panel contains no submission request and the store has no enquiry. This evidence does not support a claim that email delivery failed: the demonstrated problem is earlier, in the submission path.

Test FORM-B at 09:05 makes a POST request. The server returns 500 with a failure response, while the page still shows success. The handler needs a visible failure state and must check the response. Investigate the server failure separately; hiding the banner would improve honesty without fixing the underlying fault.

Test FORM-C at 09:10 receives an acknowledgement with enquiry reference Q-103. The store contains Q-103 and the email provider has a send record, but its next event is a bounce identifying an invalid recipient. The enquiry was captured; the notification failed. Correct the configured recipient after verification, then retry that notification according to the system's rules rather than create a second enquiry.

Now count the outcomes: three banners, one stored enquiry and zero demonstrated inbox arrivals. Calling this a 100% successful form would confuse presentation with evidence. The exercise illustrates how to classify a failure; it is not a reliability benchmark for any product.

If a later delivery event appears for Q-103, update the notification record while retaining the original failure history. Do not replace the original submission time with the retry time. That distinction makes it possible to explain both when the visitor contacted you and when the organizer was notified.

Visual explainerThree banners do not prove three deliveries
FORM-ANo requestTrace the browser handler before investigating email.
FORM-BHTTP 500The error response contradicts the success banner.
FORM-CSaved, then bouncedKeep enquiry Q-103; investigate its notification.
Three displayed successes; one stored enquiry; zero demonstrated inbox arrivals.Original fictional diagnostic records. No real customer messages or provider reliability measurement.

Give AI a bounded repair task

Provide the relevant component and handler with secrets removed, the expected acknowledgement contract and the failing test case. Ask for the smallest repair and a way to demonstrate it. Do not paste production credentials or a customer enquiry to make the example feel realistic.

Prompt to adapt: Inspect this form and its deployed handler. Map the path from submit to saved enquiry to notification. Identify what the current success message actually proves. Preserve the page design, URLs and metadata. Show success only after the documented acknowledgement; handle HTTP errors and unexpected responses; preserve entered text when submission fails. Keep provider credentials on the server. Return a minimal patch and a test plan for the supplied fictional cases. Do not send emails or change hosting settings.

Treat duplicate prevention and server-side validation as backend requirements, not just disabled-button styling. If a response is lost after the server saved the enquiry, a blind retry can create another record. Use the backend's documented request-reference or idempotency mechanism where supported. Our idempotency guide explains that failure pattern in more detail.

Keep a separate notification retry path when the design supports it. Replaying an entire submission to retry one email can duplicate stored enquiries or trigger unrelated actions. Review this behavior before enabling automatic retries.

Five-minute practice before calling the form working

Use the three invented records above and write the next check for each: FORM-A needs a real submission request; FORM-B needs truthful response handling and server-error investigation; FORM-C needs recipient correction and notification verification. None justifies rewriting every page or changing DNS without further evidence.

For your own staging form, repeat one valid submission and one rejected submission. Use only your own authorized inbox for a delivery test. Confirm the stored record and notification independently, and check that a rejected submission retains useful input and exposes an accessible error.

Finally, review the form on a narrow screen and with a keyboard. A technically correct handler is still difficult to use if its error is off-screen or focus disappears. Keep the existing website-testing guide as the next step, and record which checks were performed instead of declaring the form fully tested from a single successful click.

  • The deployed form has a known handler and acknowledgement contract.
  • The success message matches evidence from the backend.
  • Stored enquiry, notification request and delivery event are separate checks.
  • Failure and retry paths avoid silently losing input or duplicating records.
  • No production secrets or real customer messages are used in the practice exercise.

Sources & further reading

Educational content. Examples are illustrative. Protocols, products and rules can change; consult the linked documentation for current details.

Explore the newsroom

Connect what you have learned with the latest reporting.