> ## Documentation Index > Fetch the complete documentation index at: https://docs.traceten.com/llms.txt > Use this file to discover all available pages before exploring further. # Event reference > Every webhook event type Traceten sends: type strings, full example payloads, and every field documented. ## What this lets you do Build against the exact payload of every event type Traceten delivers. All examples on this page are complete, valid deliveries under `api_version: "2026-07-01"`. Every delivery shares the same [envelope](/webhooks/overview#the-envelope): `id`, `type`, `api_version`, `created`, and a type-specific `data` object. This page documents each `type` and its `data`. All wire fields are `snake_case`. The taxonomy is additive-only. New types may be added; existing types and fields are never removed or renamed. Route on `type` and ignore values you do not recognize. ## `ai_session.classified` Fires when a session on your site is classified as AI-referred. Exactly one delivery per AI session, deduplicated per session, sent the moment classification completes. This is the default subscription for new endpoints. ```json theme={null} { "id": "whd_01j9x7v3k8q4r2m5n6p7s8t9u0", "type": "ai_session.classified", "api_version": "2026-07-01", "created": 1752307200000, "data": { "event_id": "evt_01j9x7v3k1e2f3g4h5i6j7k8l9", "site_id": "0b6f3c2a-9d4e-4f1b-8a7c-2d5e6f7a8b9c", "session_id": "9f2c4b1d-7e3a-42b5-8c1d-2e5f6a7b8c9d", "visitor_id": "123e4567-e89b-42d3-a456-426614174000", "url": "https://acme.com/pricing", "ai_source": "ChatGPT", "ai_confidence": 0.94, "detection_method": "rules", "country": "US", "timestamp": "2026-07-12T08:00:00.000Z" } } ``` ### `data` fields | Field | Type | Description | | ------------------ | -------------- | --------------------------------------------------------------------------------------------------------------------------------- | | `event_id` | string | Unique id of the classified event, assigned at the edge. | | `site_id` | string | Your site key: the `ttid_…` value shown on the install page and used as `data-site`. Not the site UUID from the dashboard URL. | | `session_id` | string | Browser-session identifier the classification belongs to. | | `visitor_id` | string | Persistent pseudonymous visitor identifier. | | `url` | string | Page URL at the time of the event, reduced to **origin + path**. Query strings and fragments are stripped before delivery. | | `ai_source` | string | Canonical AI source name, e.g. `"ChatGPT"`, `"Perplexity"`, `"Claude"`. New sources appear over time; treat as an open set. | | `ai_confidence` | number | Classification confidence, `0.0` to `1.0`. | | `detection_method` | string | Which detection method produced the classification, for example `"rules"` (or its legacy alias `"rule"`) for rule-based matching. | | `country` | string \| null | ISO 3166-1 alpha-2 country code, or `null` when unavailable. | | `timestamp` | string | ISO-8601 UTC time of the event. | ## `conversion.attributed` Fires when a conversion (a purchase or any custom goal) is attributed to an AI source, immediately after Traceten records the attribution. A conversion with no AI-referred session in its journey never fires this event. Exactly **one event per conversion**, not one per attribution model: the top-level fields carry the canonical model (last-touch in v1), and `models[]` carries the full four-model breakdown for the same AI source. Because the breakdown is scoped to the canonical (last-touch) source, a model that credits a different source shows `credit_fraction: 0` for it. A money-carrying conversion (Stripe or Shopify purchase). This journey had two AI touches, ChatGPT first and Perplexity last, so first-touch assigns Perplexity nothing: ```json theme={null} { "id": "whd_01j9x8a4m2n3p4q5r6s7t8u9v0", "type": "conversion.attributed", "api_version": "2026-07-01", "created": 1752307260000, "data": { "conversion_id": "conv_01j9x8a4m9w8x7y6z5a4b3c2d1", "site_id": "0b6f3c2a-9d4e-4f1b-8a7c-2d5e6f7a8b9c", "session_id": "9f2c4b1d-7e3a-42b5-8c1d-2e5f6a7b8c9d", "ai_source": "Perplexity", "event_name": "purchase", "attribution_model": "last_touch", "credit_fraction": 1, "revenue_cents_attributed": 12999, "currency": "USD", "models": [ { "model": "first_touch", "credit_fraction": 0, "revenue_cents_attributed": 0 }, { "model": "last_touch", "credit_fraction": 1, "revenue_cents_attributed": 12999 }, { "model": "linear", "credit_fraction": 0.5, "revenue_cents_attributed": 6500 }, { "model": "time_decay", "credit_fraction": 0.6, "revenue_cents_attributed": 7800 } ], "timestamp": "2026-07-12T08:01:00.000Z" } } ``` For a non-purchase goal, `revenue_cents_attributed` and `currency` are **absent** at the top level and every `models[].revenue_cents_attributed` is `null`. There is no money to split, and Traceten never fabricates a `$0`. A signup attributed to a single AI touch: ```json theme={null} { "id": "whd_01j9x8c5n3p4q5r6s7t8u9v0w1", "type": "conversion.attributed", "api_version": "2026-07-01", "created": 1752307320000, "data": { "conversion_id": "conv_01j9x8c5n1a2b3c4d5e6f7g8h9", "site_id": "0b6f3c2a-9d4e-4f1b-8a7c-2d5e6f7a8b9c", "session_id": "2d8e6a4c-1b3f-42a5-9c7d-1e2f3a4b5c6d", "ai_source": "Claude", "event_name": "signup", "attribution_model": "last_touch", "credit_fraction": 1, "models": [ { "model": "first_touch", "credit_fraction": 1, "revenue_cents_attributed": null }, { "model": "last_touch", "credit_fraction": 1, "revenue_cents_attributed": null }, { "model": "linear", "credit_fraction": 1, "revenue_cents_attributed": null }, { "model": "time_decay", "credit_fraction": 1, "revenue_cents_attributed": null } ], "timestamp": "2026-07-12T08:02:00.000Z" } } ``` Note the distinction: `0` means "this model assigned no credit to a real-money conversion"; `null` means "this conversion carries no money at all". ### `data` fields | Field | Type | Description | | ----------------------------------- | --------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `conversion_id` | string | Unique id of the conversion event. | | `site_id` | string | Your site key: the `ttid_…` value shown on the install page and used as `data-site`. Not the site UUID from the dashboard URL. | | `session_id` | string | Browser-session token credited under the canonical model. | | `ai_source` | string | Canonical AI source name for the credited (last-touch) session. | | `event_name` | string | The goal name, e.g. `"purchase"`, `"signup"`. | | `attribution_model` | string | The canonical model whose credit the top-level fields carry. `"last_touch"` in v1. One of `first_touch`, `last_touch`, `linear`, `time_decay`. | | `credit_fraction` | number | Fraction of credit assigned to the AI source under the canonical model, `0.0` to `1.0`. | | `revenue_cents_attributed` | integer | Revenue attributed under the canonical model, in integer minor units of `currency` (e.g. `12999` for \$129.99). **Present only for money-carrying conversions**, always together with `currency`. | | `currency` | string | Uppercase ISO 4217 code, e.g. `"USD"`. Present only alongside `revenue_cents_attributed`. | | `models` | array | Full four-model breakdown; one entry per model, always all four. Each entry reports the credit that model assigns to the canonical `ai_source`. | | `models[].model` | string | `first_touch`, `last_touch`, `linear`, or `time_decay`. | | `models[].credit_fraction` | number | Credit fraction that model assigns to the canonical `ai_source`, `0.0` to `1.0`. `0` when the model credits a different source. | | `models[].revenue_cents_attributed` | integer \| null | Revenue under that model in minor units, or `null` for non-purchase goals. | | `timestamp` | string | ISO-8601 UTC time of the conversion. | ## `webhook.ping` The test and verification event. Sent in exactly two situations: once automatically when you register an endpoint (to verify it), and whenever you click **Send test event** on the endpoint detail page. It never fires as part of your site's traffic. It travels the exact same signing and delivery path as real events, so a successful ping proves the real path works, and its outcome drives the endpoint's [verification status](/webhooks/setup#verification-status). Unlike real events, a ping gets exactly **one attempt**. It never enters the [retry schedule](/webhooks/delivery#retry-schedule); if it fails, click **Send test event** again. ```json theme={null} { "id": "whd_01j9x6t2j7p3q1l4m5n6o7p8q9", "type": "webhook.ping", "api_version": "2026-07-01", "created": 1752307140000, "data": { "site_id": "0b6f3c2a-9d4e-4f1b-8a7c-2d5e6f7a8b9c", "message": "Traceten webhook test — your endpoint is receiving deliveries." } } ``` ### `data` fields | Field | Type | Description | | --------- | ------ | ------------------------------------------------------------------------------------------------------- | | `site_id` | string | UUID of the site the ping was triggered for: the site's **internal id**, the one in your dashboard URL. | | `message` | string | Human-readable confirmation. Informational only; do not parse it. | **`site_id` means two different things across event types.** On `ai_session.classified` and `conversion.attributed`, `site_id` is your **site key**, the `ttid_…` value used as `data-site` in the snippet. On `webhook.ping` it is the site's **internal UUID** from the dashboard URL. The two never equal each other. If your handler keys anything on `site_id`, branch on `event` type first, or ignore the ping's `site_id` and match on the endpoint instead. ## What webhooks never carry Payloads are stable projections of Traceten's internal records, not the records themselves. The following fields exist internally but are deliberately excluded from every event, permanently: | Excluded | Why | | ---------------------------- | -------------------------------------------------------------------------------------- | | `ip_hash` | Derived from the visitor's IP address. Network-level identifiers never leave Traceten. | | `user_agent` | High-entropy fingerprinting surface. | | `referrer` | Referring URLs can embed personal data (search terms, profile URLs, tokens). | | `asn` | Network (autonomous system) data is an indirect location and identity signal. | | Emails, names, phone numbers | Traceten does not collect PII on the default path, so there is nothing to send. | Additionally, the `url` field in `ai_session.classified` is reduced to origin + path before delivery. Query strings and fragments routinely carry emails, tokens, and session state, so they are stripped and structurally cannot ride along. `conversion.attributed` carries no `visitor_id`. That is not a privacy exclusion (`visitor_id` is pseudonymous and is included in `ai_session.classified`); the credited `session_id` is simply sufficient to attribute a conversion. Join the two event types on `session_id` if you need visitor-level analysis. See [what data we collect](/privacy/data-collected) for the full picture of what Traceten does and does not store. ## Next * [Verify signatures](/webhooks/verify-signatures) * [Delivery semantics](/webhooks/delivery)