Automate data collection with n8n and webhooks

What you can automate now

Use Tellseed with n8n to create a draft from supplied questions, hand it to the owner for approval, publish the approved version, or create a reviewed dataset export. These workflows use the native MCP Client node and do not need an AI model call. Your own agent can still help write the questions before the workflow runs.

Automatic webhooks notify your HTTPS endpoint when an answer arrives, a questionnaire round finishes, an answer is finalized by review, a campaign is published or an export becomes ready. This works for surveys, customer feedback and research as well as human labels and training data. Tellseed supplies the campaign and event delivery; you configure your downstream tool and audience.

• Restricted keys: choose actions, one campaign or the whole workspace, and an expiry.
• Signed events: answer.received, round.completed, review.finalized, campaign.published and export.ready.
• Delivery controls: test, inspect attempts, replay, disable and revoke keys from your workspace.

• Open workspace automations
• Download the n8n workflows

Choose a workflow for your data collection

Start with the decision your data will support, then choose the event and destination. The same collection workflow supports a customer questionnaire, a research study or a dataset for model training. You bring the participants and configure the destination in n8n.

• Customer feedback: store each answer by its submission ID and use round.completed to trigger a follow-up workflow when a person finishes. Check the actual answered count and stop reason before treating the round as a complete questionnaire.
• Research data: keep original responses separate from review decisions. Use review.finalized as a signal to retrieve the current reviewed target, then create a frozen export for analysis.
• AI training data: collect text examples or classification labels, review corrections and export accepted targets with their provenance. Verify the frozen file hashes before converting the dataset to the format your training pipeline expects.

• Customer feedback survey recipe
• Human data collection for model training
• Reviewed labels and downloadable dataset example

1. Create a key with only the access it needs

Sign in and open Automations. Give the key a recognisable label, choose a campaign scope, select its actions and set an expiry. Copy the key once into an HTTP Bearer Auth credential in n8n, without the Bearer prefix. Select that credential in each Tellseed MCP node.

For reviewed exports, select Read campaigns, Read results and Create export snapshots. For draft creation, use a workspace-wide key with Read campaigns and Create and edit drafts. Publication needs Read campaigns and Publish approved drafts. A campaign-scoped key cannot create a new campaign.

Automation keys cannot approve campaigns, manage billing, mint owner or approval credentials, or administer webhooks. Revocation stops subsequent requests. The workspace shows when a key was created, when it expires and when it was last used. Keep keys in credential storage, not URLs or shared workflow JSON.

• MCP connection reference

2. Create, approve and publish

Import create-draft.json and replace Campaign input with your own questions and collection settings. The example contains ten feedback questions shown in order. Run it to get the campaign ID and an owner approval URL. Creating the same slug again returns the existing campaign; choose a new slug for a separate campaign, or explicitly update the existing draft.

Open the returned URL as the signed-in owner, inspect the contributor experience and approve the exact version. Then import publish-approved.json, enter that campaign ID and run it with the publication key. It checks approval, publishes and returns a participant link that expires after 30 days. Editing the draft invalidates approval and blocks publication until you approve again.

Creating a participant link grants access to answering, so share it only with your intended audience. Re-running the publication workflow can mint another link. It does not approve, invent questions or recruit participants for you.

3. Receive signed events

Import webhook-events.json in n8n. Add a destination in Tellseed Automations using the n8n production webhook URL and select the campaigns and event types. Tellseed reveals a signing secret once. Save it in an n8n Crypto credential, in Hmac Secret, and select it in Calculate HMAC. Publish the n8n workflow, then send a test event from Tellseed.

The receiver reads the original request bytes, verifies the HMAC-SHA256 signature and checks a five-minute timestamp window before accepting the event. It acknowledges only after the last node succeeds. Keep that verification before every downstream action. Do not replace it with an immediate acknowledgement or accept unsigned requests.

Events contain an event ID, schema version, workspace and campaign IDs, occurrence time and a small data object. An answer event identifies the submission and campaign version; an export event identifies the frozen snapshot. Answer text, names, email addresses and access tokens are excluded.

Use round.completed for completed questionnaire rounds. It reports the actual answer count, skipped questions across the visit and the stop reason. A ten-question campaign does not finish after its first answer. All-skip visits do not emit completion. Use review.finalized after enough reviewers accept an answer without dissent; a later review can change that state.

To retrieve an answer, use GET /api/automation/answers/{submissionId} with a results:read Bearer key for that campaign. The response includes the current review status and corrected target beside the original answer. Recheck review.finalized before using a target because events can arrive late. Export manifests and files also accept a results:read key. Results remain subject to the workspace and trial access rules. Only retrieve fields needed by your destination and set its retention policy.

4. Make downstream actions safe to retry

Delivery is at least once. Use event.id as the unique key when writing to your destination, such as an upsert in a database. A timeout may happen after your receiver has already saved an event. Replaying it must not add a second record or send an unintended second notification. The receiver template verifies events; you must configure the idempotent destination.

Tellseed makes up to eight attempts, with delays of 30 seconds, 2 minutes, 10 minutes, 30 minutes, 2 hours, 6 hours and 24 hours. The destination must answer within 10 seconds. Use a durable queue for longer work. Delivery history and replay are available for seven days from the original event. Events are created for active matching destinations, with no automatic historical backfill.

Disabling a destination cancels queued work. A request already in flight may finish. Rotate a signing secret by disabling the destination and adding a replacement. Revoking an automation key stops authenticated reads but does not disable independent webhook subscriptions; disable those separately.

Use reviewed-export.json to freeze a dataset after review. The reviewed JSONL contains accepted, finalized targets. Raw CSV includes unreviewed and excluded answers. Reuse the snapshot ID for downstream retries; running the export creation step again creates a new snapshot.

• Worked training-data collection and review example

A complete live automation example

On 3 October 2026 we ran these workflows against public Tellseed HTTPS in a separate Hosted Free workspace, using real n8n 2.41.6. The operator approved a ten-question synthetic campaign. Ten answers, ten submission retries and ten scripted reviews produced ten saved answers, one completed round and three export files whose hashes matched the manifest.

The production webhook worker delivered 23 business events across all five event types, plus a test event. The n8n Crypto node verified signatures before a SQLite destination saved each unique event ID. We interrupted that destination, observed a production retry and confirmed that replay did not add a second record. Revoking the campaign key blocked the next export download.

This was an internal functional pilot, not a customer testimonial or a human-labelled training dataset. The workspace was provisioned by the operator, and the receiver was temporary. Email signup, a persistent customer destination, n8n Cloud and a real payment were outside this test. The test campaign was closed and its access revoked afterwards.

• Read the pilot checks, counts and limitations (JSON)
• Reproduce the workflow with the public n8n templates

Compatibility and current limits

The templates target n8n 2.41.6 and use its native MCP Client and Crypto nodes. Other versions and n8n Cloud need their own validation. No native Tellseed n8n node, Make listing or Zapier app is required for the current workflow.

Hosted Tellseed and the Node production server include webhook delivery. Destinations must use public HTTPS on port 443 with public IPv4 DNS. Redirects and private networks are refused. Self-hosted Worker deployments need a separately implemented delivery runner and do not expose webhook configuration by default.

Ready-made spreadsheet destinations and platform marketplace listings are still planned. Voice collection, participant recruitment and automatic model training are separate roadmap items.