Learn Automation #3: AI Lead Qualifier in Zapier, GitHub Agentic Workflows, and Model Pinning

One workflow. One tool. One tip. Every week.

Want this in your inbox every week? Subscribe to Learn Automation →

⚡ This Week’s Workflow

Build an AI Lead Qualifier with Zapier + ChatGPT

New leads come in. Some are ready to buy. Others are just browsing. The problem: manually reading every submission, Googling their company, and deciding who gets a follow-up eats hours you don’t have.

This workflow uses ChatGPT inside Zapier to automatically score every new lead, extract key details, and route them to the right place — hot leads get an instant Slack alert, cold ones go into a nurture sequence.

Tools you’ll need: Zapier (free plan works for testing, paid for production), a ChatGPT (OpenAI) connection in Zapier, Google Sheets or your CRM, and Slack.

Here’s how to build it:

1. Set up the trigger. Create a new Zap and choose your lead source as the trigger. This could be Zapier Forms (simplest), Typeform, Google Forms, or a webhook from your website’s contact form. The trigger fires every time someone submits their info.

2. Add a ChatGPT “Conversation” step. Select ChatGPT (OpenAI) as your action app and choose the “Conversation” action. In the User Message field, pass in the lead’s form data. Use a system prompt like this:

   “You are a lead qualification assistant. Analyze the following lead submission and return a JSON object with these fields: score (hot, warm, or cold), reasoning (one sentence explaining why), company_size (if mentioned or inferrable), use_case (what they want to automate), and follow_up_priority (1-3, where 1 is highest). Base your scoring on: budget mentioned = hot, specific use case = warm, vague inquiry = cold.”

   Map your form fields into the message — name, email, company, message, etc.

3. Parse the response with Formatter. Add a Formatter step to extract the structured data from ChatGPT’s response. Use “Text → Split” or the “Extract Pattern” action to pull out the score, reasoning, and priority fields. Alternatively, use ChatGPT’s “Extract Structured Data” action instead of “Conversation” — it returns clean JSON fields you can map directly without parsing.

4. Add a Paths step for routing. This is where it gets powerful. Add Zapier Paths to branch your workflow based on the lead score:

   • Path A (Hot leads): Score = “hot” → Send a Slack message to your #sales channel with the lead’s name, company, reasoning, and a direct link to their form submission. Also create a row in your CRM or Google Sheet tagged as “hot.”
   • Path B (Warm leads): Score = “warm” → Add to Google Sheets with the full analysis. Optionally trigger a follow-up email sequence via Mailchimp or your email tool.
   • Path C (Cold leads): Score = “cold” → Log to a separate sheet for monthly review. No immediate action needed.

5. Add a timestamp and log everything. On each path, include a Google Sheets step that logs the original submission plus ChatGPT’s full analysis — score, reasoning, priority. This gives you a searchable history of every lead and how the AI assessed it.

6. Test with real data. Submit a few test leads with varying levels of detail. Check that hot leads actually trigger Slack notifications, warm leads land in the right sheet, and cold leads get logged quietly. Adjust your system prompt based on what you see — the scoring criteria are the most important thing to tune.

Take it further: Replace Google Sheets with HubSpot or Salesforce using Zapier’s native CRM integrations. Or add a second ChatGPT step that drafts a personalized follow-up email for hot leads based on their specific use case — then send it automatically via Gmail.

Inspired by Zapier’s guide to building AI-powered workflows (Jan 2026) and Zapier’s automated system guide

🔧 Tool of the Week

GitHub Agentic Workflows

GitHub just released Agentic Workflows in technical preview — an open-source framework that lets you automate repository tasks using AI agents inside GitHub Actions. The twist: you write workflows in plain Markdown instead of YAML.

Describe what you want in natural language (“When a new issue is labeled as a bug, analyze the error, find related code, and suggest a fix in a comment”), and an AI agent — Copilot CLI, Claude Code, or OpenAI Codex — figures out the steps and executes them inside a sandboxed GitHub Actions environment.

Practical example: Set up a workflow that runs every time a CI pipeline fails. The agent reads the error logs, traces the failure to specific code, checks recent commits for likely causes, and posts a summary comment on the PR with a suggested fix. No YAML gymnastics required.

What makes it interesting: It’s open source (MIT license), security-first (read-only by default, sandboxed execution, network isolation), and works with multiple AI backends. You install it via the gh-aw CLI extension — run gh aw compile on your Markdown file, and it generates a standard GitHub Actions workflow you can commit and run.

Skip if: You need deterministic automation where the same input always produces the same output. AI agents introduce variability — great for tasks requiring judgment (like issue triage), less ideal for strict deployment pipelines. Also still in technical preview, so expect rough edges.

Released February 13, 2026 — GitHub Blog • The New Stack coverage

💡 Quick Tip

Pin Your AI Model Versions — Never Use “Latest”

If any of your Zapier Zaps, n8n workflows, or Make scenarios call an OpenAI model using chatgpt-4o-latest, they will break on February 17, 2026. OpenAI is shutting off that alias, and the recommended replacement is gpt-5.1-chat-latest — which behaves differently.

This isn’t a one-time thing. OpenAI, Anthropic, and Google regularly retire model versions. If you’re using a “latest” alias in your automation, you’re accepting that your workflow’s behavior can change without warning.

The fix takes 30 seconds: Open your AI steps and replace any “latest” alias with a specific, pinned model version (like gpt-4o-2024-08-06 or gpt-5). Then set a calendar reminder to review your model versions quarterly.

In Zapier: edit your ChatGPT action → Model field → select a specific version instead of the default. In n8n: open your OpenAI node → change the model dropdown. In Make: update the model parameter in your HTTP or OpenAI module.

Source: OpenAI Deprecations page (updated Feb 2026)

That’s it for this week.

Got a question about a workflow, or something you’d like me to break down next? Just hit reply — I read everything.

— Sebastian

Subscribe to Learn Automation → — One workflow. One tool. One tip. Every week.