Zapier Text Classification: AI by Zapier Classify vs Vertex Classify Text vs HTTP → Jev

In Zapier, classification is usually a routing step: normalize an inbound field (ticket body, form answer, chat transcript), ask a model to pick from a short label list, then branch Paths or Filters before you write to Zendesk, Slack, or a CRM. Two first-party-ish options dominate docs today: (1) AI by Zapier with a Classify prompt template and structured output fields; (2) the Google Vertex AI Classify Text action when you already have a GCP project. A third pattern is Webhooks by Zapier or the HTTP/Custom Request style action that POSTs to an external decision API such as JevModel. None of these replaces a labeled holdout from your own traffic.

According to Zapier Help, AI by Zapier lets you write a prompt (or pick a template under categories such as Classify), select a model tier—Standard (1× tasks), Advanced (3×), Premium (5×, often the default for new steps), or Bring Your Own Key (1× with your provider)—and define Output Fields so later steps can map structured results. Supported providers include OpenAI, Anthropic, Google Gemini, Azure OpenAI, and Amazon Bedrock. Content moderation is on by default for Zapier-provided models. Strengths: fast authoring inside the Zap editor, preview before publish, and flexible models without leaving Zapier. Limits: each run resets knowledge (no learning across runs); confidence you can calibrate across providers is not a first-class product contract; the decision schema lives in this Zap’s fields unless you copy it elsewhere; and task multipliers make Advanced/Premium expensive at high volume. Treat template output as prompt-shaped JSON until you measure precision on your tickets.

Zapier’s Vertex guide describes a Classify Text action: map Text, list Categories, select a Google Cloud Project ID, and keep Model Region at the default us-central1 unless your org requires another region. Optional Advanced Fields expose LLM parameters. Typical use cases in the doc include spam, sentiment, topic detection, and content moderation. This path ties classification to your GCP billing and IAM—not to Zapier’s AI task multipliers. It still returns a category-style answer that you branch in Paths; it does not give you a portable Choice/Score/Noul contract for non-Zapier apps. Prefer it when Vertex is already approved by security and you want classification to stay inside Google’s stack.

The table contrasts workflow shape and decision contract—not marketing accuracy scores. Calibrate any confidence on your own labeled sample.

Stay on AI by Zapier Classify or Vertex Classify Text when labels are few and stable, false routes are cheap to undo, the whole automation already lives in Zapier, and you mainly need editor-native field mapping. Write precise category names (and descriptions in the prompt for AI by Zapier), keep an Other/ambiguous Path, Preview with real samples, and log the model tier or Vertex region with every route so provider upgrades are auditable. Prefer Vertex when GCP is already the system of record for AI spend.

A common pattern is: Trigger → Formatter/Code to normalize text into state → Webhooks by Zapier POST (or another HTTP-capable action) to JevModel’s /api/v1/systemone with Choice questions (2–12 labels) and optional Score/Noul → Paths/Filters on the chosen label and score-like fields → side effects only on high-confidence paths → human-review queue otherwise. Keep the API key in Zapier Credentials / connected accounts—never hard-code keys in field mappings or commit them to git. Respect product budgets: validated state is capped (about 8000 characters after validation), 1–8 questions per request, and free verified accounts get 5 successful runs per UTC day. JevModel hosts typesafe/jev-1.13 via OpenRouter; it is independent of TypeSafe AI (which develops Jev), is not multimodal on this site, and does not ship downloadable weights.

// Illustrative only — map fields in the Zap editor; keep secrets in Credentials.
POST /api/v1/systemone
{
  "state": "Customer asks for a refund on order #1842 after late delivery.",
  "questions": [
    {
      "id": "route",
      "type": "choice",
      "prompt": "Support route",
      "labels": ["billing", "shipping", "product", "other"]
    }
  ]
}

Whether you parse a free-form confidence field from AI by Zapier, trust a Vertex score, or use Jev Score/Noul, calibrate on a labeled sample before auto-applying tickets. A practical policy: auto-apply safe labels above a high band, send mid-band items to a human queue with the model’s top labels as suggestions, and force review below a floor. Wire mid and low Paths before any write action. Vendor numbers are weak signals until you measure them—see confidence-and-abstention and support-triage.

1) Freeze a label dictionary with edge-case examples. 2) Decide Other vs discard for unmatched items. 3) Pick AI by Zapier, Vertex, or shared Jev API based on reuse, GCP posture, and abstention needs. 4) Store keys only in Credentials. 5) Measure precision/recall on a holdout after every model, tier, or prompt change. 6) Gate destructive actions behind thresholds and a human Path. 7) Do not claim ranking or accuracy gains you have not measured.