
Create the current web-form agent
Zendesk's creation guide starts in the AI agents workspace. Select Dashboard, Create AI agent, then for Email: Zendesk includes API and web form channels in that path. Select a brand and knowledge base, optionally add a crawler, and ensure at least one knowledge source is connected. Set the factual business profile, tone, default language, translations, and the escalation acknowledgement on Set up your AI agent on email and web form. Choose Save and close, or choose Go to activation, select the web form, and click Activate on channels. Only one agent can be active for that form at a time.
Use this current flow for a new agent; existing Essential, legacy, or pre-March-2026 Advanced accounts may still expose different screens. The new agent is channel-specific, so do not use a Messaging result as evidence that a web-form journey works.
Design the form and the next step together
A form can collect the facts that make the reply or human handoff useful. Ask only for information the next step needs: an order reference, product, account identifier, or issue category. Every extra question adds friction and makes an urgent customer repeat themselves.
For each request type, write down the expected outcome. A straightforward policy question may receive a sourced answer. A refund request may need an identity check and a human route. A failed action needs a clear customer message and a verified operational next step. Do not claim a ticket is owned merely because the agent showed a fallback reply.
Test the full channel, not a sample sentence
Submit representative web forms in a safe test environment. Include a clear question, incomplete information, a policy edge case, a request for a person, and an action that is designed to fail. Inspect what was sent to the customer, whether a ticket was created, which fields and tags are present, the destination queue, the action result, and the failed-escalation behavior. Re-run these cases whenever the form, source content, permissions, or routing changes.
Turn a form's requirements into an eesel review
A support owner can use Claude Code, Codex, or Cursor with eesel CLI to inspect the same eesel teammate and workspace used in the dashboard. The useful job here is checking whether the teammate's instructions match what the form actually collects.
Give the coding agent an approved, non-customer field list—for example, issue category, order reference, and an optional description. Ask it to run instructions --agent zendesk-web-form and inspect the connected sources, then propose how to handle a damaged-order submission with no order reference. Replace the example agent name with the existing teammate's name or ID.
The proposal should answer three questions: which missing detail should the reply request, what replacement promise must wait for a person, and which team should receive the exception? This can expose a mismatch before rollout: the form may allow an empty order reference while the teammate's instructions assume one is always available.
CLI commands use Node.js 18.17+ and return JSON. A person can read the output directly; a script or coding agent can compare it with the field list and produce a proposed instruction diff. Keep personal records and attachments out of this initial review. The workspace owner reviews permissions and approves any instruction write, using command help and --dry-run where supported before it is sent. Inspect the applied instructions in the dashboard or CLI.
Then run approved synthetic cases in fresh teammate conversations: missing order reference, supplied reference, and an explicit human request. Check responses and activity, remembering that chat can perform real, billed work and only some actions are held for approval. Finally submit the actual Zendesk form and inspect the resulting fields, acknowledgement, ticket, and receiving queue. The CLI helps review the eesel teammate; it does not configure the Zendesk form or prove its routing from a text response.
Try an AI teammate for Zendesk forms
eesel can work as an AI helpdesk teammate in your Zendesk form workflow. The CLI and dashboard operate the same workspace teammate, so the team can inspect evidence before it widens the job. Try eesel

Frequently Asked Questions
What are Zendesk AI agents in web form channels?
They are AI agents that can handle web-form requests using connected knowledge and a configured channel workflow.
How do I create a web form AI agent?
In the AI agents workspace, choose Dashboard, Create AI agent, then for Email. Complete the email and web-form setup, then select the web form during activation.
What should a web form AI agent collect?
Collect only details the support team needs to act, such as an order number or issue category. Avoid a long form that delays a human request.
How should web form AI agents be tested?
Submit realistic form cases and inspect the reply, ticket creation, captured fields, action result, escalation path, and failed-case behavior.







