Step 2: Connect GotoHuman Data Sources

These nodes pull prior review examples for classification and drafting context. They are required to build prompt context used by the AI agents.

1. Open Retrieve Classifier Examples and keep URL set to https://api.gotohuman.com/fetchResponses.
2. Ensure Send Query is enabled and the query parameters include fieldIds set to meta.emailText,classification and approvedValuesOnly set to true.
3. Credential Required: Connect your gotoHumanApi credentials in Retrieve Classifier Examples.
4. Open Retrieve Reply Examples and verify filterResponseValues contains the expression {{ [ { "field": "classification.value", "operation": "==", "value": $json.output.classifiedAs } ].toJsonString() }}.
5. Credential Required: Connect your gotoHumanApi credentials in Retrieve Reply Examples.

⚠️ Common Pitfall: Leaving formId empty in the GotoHuman queries will return no examples. Populate it with your review form ID before production.

Step 3: Set Up AI Classification and Drafting

Configure the AI models and agents that classify emails, generate draft replies, and prepare ticket details for bugs and feature requests.

1. In Primary Chat Model, select the model gpt-5-nano as configured.
2. Credential Required: Connect your OpenAI credentials in Primary Chat Model (credentials are required on the model node, not on Email Category Analyzer).
3. In Secondary Chat Model, select the model gpt-5-nano and connect OpenAI credentials (used by Draft Reply Generator).
4. In Ticket LLM Model, select gpt-5-nano and connect OpenAI credentials (used by Ticket Drafting Agent).
5. Open Email Category Analyzer and keep the text field expression set to =From: {{ $('Batch Email Queue').item.json.text }} To: {{ $('Batch Email Queue').item.json.to.text }} Subject: {{ $('Batch Email Queue').item.json.subject }} Body: {{ $('Batch Email Queue').item.json.text }}.
6. Confirm Classification Parser uses the manual schema with classifiedAs enum and important boolean. This parser is attached to Email Category Analyzer; credentials are managed by the model node, not the parser.
7. Open Draft Reply Generator and keep the text field expression set to =From: {{ $('Batch Email Queue').item.json.from.text }} To: {{ $('Batch Email Queue').item.json.to.text }} Subject: {{ $('Batch Email Queue').item.json.subject }} Body: {{ $('Batch Email Queue').item.json.text }}.
8. Confirm Draft Output Parser uses {"textEmailDraft": "Hello Jack..."} as the structured output example. Credentials are managed by Secondary Chat Model, not the parser.
9. Open Ticket Drafting Agent and keep the prompt text expression intact to include classification, subject, and text from Batch Email Queue.
10. Confirm Ticket Output Parser uses the manual schema with required fields title, description, and priority. Credentials are managed by Ticket LLM Model, not the parser.

Step 4: Configure Human Review and Routing

Human review ensures the AI draft and classification are approved, rejected, or sent back for revision, and directs the workflow to reply or create a ticket.

1. Open Human Review Task and ensure fields.value.email uses ={{ $('Batch Email Queue').item.json.textAsHtml }} and fields.value.emailReply uses ={{ $json.output?.textEmailDraft ? {ai: {prompt: $('Compose Prompt Text').item.json.prompt}, content: $json.output?.textEmailDraft} : "No reply needed - FYI only" }}.
2. Credential Required: Connect your gotoHumanApi credentials in Human Review Task.
3. Verify Reviewer Decision routes approved to Test Run Switch, rejected to Skip Action, and chat to Update Draft Prompt.
4. Keep Update Draft Prompt assignments so prompt uses ={{ $json.messages[0].content || $('Compose Prompt Text').item.json.prompt }} and reviewToUpdate uses ={{ $json.reviewId }}.
5. Confirm the routing order: Draft Reply Generator → Human Review Task → Reviewer Decision → Test Run Switch.

⚠️ Common Pitfall: If reviewTemplateID is not set in Human Review Task, GotoHuman cannot render the review form and no approvals will occur.

Step 5: Configure Output Actions (Replies and Tickets)

Approved replies are sent to Gmail, and bugs/feature requests are turned into Linear tickets.

1. Open Send Thread Reply and confirm Operation is reply, Message is ={{ $('Human Review Task').item.json.responseValues.emailReply.value }}, and Message ID is ={{ $('Incoming Email Trigger').item.json.id }}.
2. Credential Required: Connect your Google Gmail credentials in Send Thread Reply.
3. Verify Filter Bug or Feature uses the condition expression ={{ ["bug_report", "feature_request"].includes($('Human Review Task').item.json.responseValues.classification.value) }}.
4. Open Create Linear Issue and set Team ID to your Linear team identifier (replace [YOUR_ID]).
5. Confirm Create Linear Issue fields are mapped to ={{ $json.output.title }}, ={{ $json.output.priority }}, and ={{ $json.output.description }}.
6. Credential Required: Connect your Linear credentials in Create Linear Issue.

Step 6: Configure Manual Testing Inputs

Use the seeded email data to test end-to-end processing without real Gmail messages.

1. Use Manual Run Starter to initiate test runs without Gmail.
2. Confirm Manual Run Starter connects to Seed Emails Batch Two and then to Batch Email Queue.
3. Review Utility: Seed Test Emails and Seed Emails Batch Two data to ensure [YOUR_EMAIL] placeholders are replaced with your test address if needed.

⚠️ Common Pitfall: If you test only with Manual Run Starter, Test Run Switch will follow the Test Run branch (no real Gmail thread reply), which is expected behavior.

Step 7: Test and Activate Your Workflow

Run a full test to validate classification, human review, reply sending, and ticket creation, then activate the workflow for production.

1. Click Execute Workflow on Manual Run Starter to process seeded emails through Batch Email Queue.
2. Verify that Email Category Analyzer classifies emails and Draft Reply Generator produces a draft that appears in Human Review Task.
3. Approve a review in Human Review Task and confirm it routes through Reviewer Decision to Test Run Switch.
4. For real emails, ensure Incoming Email Trigger is executed so Test Run Switch sends replies via Send Thread Reply.
5. Confirm bug/feature classifications pass Filter Bug or Feature and create tickets in Create Linear Issue.
6. Once successful, toggle the workflow to Active to enable production processing from Incoming Email Trigger.