# Decide between native AI Steps and a Custom GPT in 10 minutes Source: https://getstartedwithai.net/tutorials/zapier-ai-features-vs-custom-gpt Site: Get Started With AI Tagline: Build AI automations that actually run your business. Audience: solo professionals, freelancers, consultants and AI implementers in training Contact: hello@getstartedwithai.co.uk This Markdown companion is generated from the prerendered page for AI crawlers. # Decide between native AI Steps and a Custom GPT in 10 minutes Pick the wrong primitive and you spend a half-day refactoring later. Pick the right one in 10 minutes and the workflow ships clean. Zapier rolled out three AI features through 2025 and 2026 (Copilot for build help, AI Steps for in-Zap LLM calls, AI Agents for multi-turn flows). Custom GPTs do something different. The decision tree below is four branches deep and lands on the right answer for inbound triage, discovery summaries and inbound sales chat. The lever: match the shape of the work to the shape of the tool, not the other way round. ## What each feature actually does - Zapier Copilot: helps you build a Zap. You describe what you want; Copilot drafts the trigger, actions and field mapping. A build-time tool, not a runtime tool. - Zapier AI Steps: an action inside a Zap that calls an LLM (ChatGPT, Claude or Gemini) on the data flowing through. Useful for summarisation, classification, drafting. - Zapier Agents: a separate product. A conversational agent with a goal and tools. Multi-turn, multi-step. Charged separately from regular Zap tasks. - Custom GPT: a packaged ChatGPT, accessible via the GPT Store or a private link. No automation; just a configured chat tool. ## Decision tree Pick the option that matches the question, in this order. - Reacting to an event (form submission, email arrival, deal status change)? Build a Zap, optionally with AI Steps inside it. - Need a chat tool a person opens, types into, and gets a packaged answer from? Build a Custom GPT. - Need a back-and-forth conversation that books, refunds or escalates over multiple turns? Reach for Zapier Agents or a dedicated agent platform (Vapi for voice, Claude Code for ops). - Only building a Zap and want help writing it? Switch on Zapier Copilot at build time, then turn it off. ## Workflow 1: Inbound email triage with AI Steps Trigger: New email in Gmail with label "client". The first AI Step (Claude) classifies the email as question, request, complaint or FYI. A second AI Step drafts a reply in your voice. The final action saves to Notion in the right database row. AI Steps fit because the trigger is an event and the AI runs once per event. ## Workflow 2: Discovery call summariser as Custom GPT You finish a discovery call. You paste the raw notes into a Custom GPT named Discovery Summariser. The GPT returns the structured intake brief. There is no event, no action, no chain. A Custom GPT is exactly the right tool. Building this as a Zap would over-engineer it. ## Workflow 3: Inbound enquiry chat as AI Agent A prospect lands on your site, asks "What does a typical engagement cost?" The agent should ask a clarifying question, share the rate range, ask for the email, and book a call. Multi-turn, multi-step, with tools (calendar lookup, CRM write). Use the AI Agent shape here. Try Zapier Agents first if you already pay for Zapier; consider a dedicated tool if the volume is high. ## How to add an AI Step to an existing Zap (10 minutes) 1. Open the Zap. Click the plus between two steps to insert a new action. 2. Search for AI by Zapier or for the specific provider (OpenAI, Anthropic, Google). 3. Choose Send Prompt. In the Prompt field, write a structured instruction: "Read the email below. Return one of: question, request, complaint, FYI. Reply with only the label." 4. Map the email body into the prompt with the field picker. 5. Test. The output appears in the next step as a usable variable: {{AI by Zapier output}}. 6. Use that output to branch the rest of the Zap with Paths or Filter. ## What you ship A clear primitive choice for any new automation, a four-branch decision tree to keep on your desk, and one real AI Step inserted into an existing Zap. The next time a client asks for a chat tool, you start with a Custom GPT in 30 minutes rather than over-building a Zap. ## How to use this on a discovery call Most small business clients ask for a Custom GPT first because it is the fastest demo. Build the GPT in 30 minutes, ship it, then upsell the Zap with AI Steps when the team realises they need automation around the GPT, not just the GPT itself. Charge £600 to £1,200 for the GPT, £1,500 to £4,000 for the wrapped automation. ## Common mistakes - Using AI Agents for jobs that are really Zaps with AI Steps. Agents are the most expensive shape; only use them when you need multi-turn dialogue. - Building everything as Custom GPTs. A GPT cannot react to a calendar event or write to your CRM. Some work needs a Zap. - Using Copilot at runtime. Copilot is a build-time helper, not a worker. Turn it off once the Zap is shipped. - Skipping the test step on AI Steps. LLM outputs vary. Run three real test cases before you trust the Zap to send anything to a client. ## What to build next Pair this with the Custom GPT tutorial and the first Zapier automation tutorial so you have one of each pattern running in your own business. ## Related builds More tutorials in Automation. Each one runs in 60 to 90 minutes and ships a finished asset. ### Capture every lead automatically: a 30-minute Zapier build 30 to 60 minutes. Solo Pro. [Open tutorial](/tutorials/first-zapier-automation-30-min) ### Run automations Zapier cannot: the first Make scenario 60 to 90 minutes. Solo Pro. [Open tutorial](/tutorials/first-make-scenario-visual) ### Self-host the automation stack: when n8n is worth the effort 60 to 90 minutes. Solo Pro / Implementer. [Open tutorial](/tutorials/n8n-self-hosted-when-worth-it) ### Pick the right automation tool in 5 minutes: Zapier vs Make vs n8n 30 to 60 minutes. Solo Pro / Implementer. [Open tutorial](/tutorials/zapier-vs-make-vs-n8n) ## Questions people ask before they start ### Do I need to pay extra for AI Steps? AI Steps run on the same Zapier task budget plus the cost of the underlying model (you provide your OpenAI or Anthropic key). One Zap run with one AI Step costs one Zapier task and one model API call. Budget about £0.01 per AI Step on a typical workload. ### Are AI Agents worth the separate fee? Only if the work is genuinely multi-turn (booking, qualifying and escalating). Most flows that look agentic are just chained Zaps with AI Steps and cost less. ### Can a Custom GPT trigger a Zap? Yes. A Custom GPT can call a webhook, which lands as a Zap trigger. Useful when you want a chat front-end on a Zap workflow. Setup takes about 20 minutes. ### Will Copilot replace consultants? Copilot writes the Zap, but it does not pick the work to automate. The WIRE Method work (audit, identify, run, embed) is what the client pays you for. The build is the easy part once the work is named.