Zapier's ChatGPT and OpenAI actions are stateless. Every Zap runs in isolation — the AI has no memory of the same contact's previous interactions, no knowledge of what was discussed last week, no context beyond what you manually include in the prompt for this exact run.
For simple Zaps that trigger once and finish, this is fine. For AI-powered workflows that interact with the same customers repeatedly — email assistants, support bots, lead follow-up sequences — statelessness is a fundamental problem. This guide shows the complete fix.
Why Zapier ChatGPT Zaps Have No Memory
Zapier executes each Zap run independently. There is no shared state between runs, no built-in session storage, and no way for the ChatGPT action to access anything from a previous execution.
The ChatGPT action accepts a prompt and returns a response. That is all. If you want the AI to know that this customer submitted a support ticket three weeks ago and was unhappy with the response, you have to include that context in the prompt yourself — which means retrieving it from somewhere first.
That somewhere is retainr.
The Pattern: Search Before, Store After
Adding persistent memory to any Zapier ChatGPT Zap follows a single pattern:
- Webhooks GET — before the ChatGPT action, retrieve relevant memory for this contact
- ChatGPT action — inject the retrieved memory into the system prompt
- Webhooks POST — after the ChatGPT response, store what happened for next time
This pattern works with any Zapier AI action: ChatGPT, OpenAI, Claude, Gemini. The memory layer is model-agnostic.
Step-by-Step Setup
Step 1: Get your retainr API key
Create a free account at retainr.dev/dashboard. Copy your API key. You get 1,000 memory operations per month on the free plan — enough to test and run a small workflow.
Step 2: Add Webhooks GET before the ChatGPT action
In your Zap, add a Webhooks by Zapier action before the ChatGPT step:
- Method: GET
- URL:
https://api.retainr.dev/v1/memories/search - Query string params:
namespace:customer:{{email}}(map the email field from your trigger)q:{{subject_or_topic}}(the current query or subject)limit:5
- Headers:
Authorization: Bearer YOUR_API_KEY
The response contains a results array with up to 5 relevant past memories. If no memories exist yet (new customer), the array is empty — handle this gracefully.
Step 3: Inject memory into the ChatGPT system prompt
In the ChatGPT action, add the memory results to the system prompt:
You are a helpful assistant.
Prior context for this customer:
{{webhook_results_formatted}}
Use this context to inform your response. If no prior context exists, respond normally.
If the Webhooks GET returns an array and ChatGPT needs a string, add a Formatter by Zapier step between the Webhooks GET and ChatGPT to join the array values into a single text block.
Step 4: Add Webhooks POST after the ChatGPT action
After the ChatGPT step, add another Webhooks by Zapier action:
- Method: POST
- URL:
https://api.retainr.dev/v1/memories - Headers:
Authorization: Bearer YOUR_API_KEY,Content-Type: application/json - Body (JSON):
{
"namespace": "customer:{{email}}",
"content": "{{topic_or_subject}}: {{chatgpt_response_summary}}"
}Store a concise summary — 1-3 sentences — not the full ChatGPT response. The summary is what gets retrieved as context next time.
Namespace Design: How to Scope Memory
The namespace field determines whose memory you're reading and writing. Common patterns:
| Use case | Namespace |
|---|---|
| Per customer email | customer:{email} |
| Per HubSpot contact | hubspot:{contact_id} |
| Per Zendesk ticket requester | zendesk:{requester_email} |
| Per Airtable record | airtable:{record_id} |
| Per Gmail sender | email:{from_email} |
Use the same namespace in both the GET and POST steps. If you change it, the search won't find the stored memories.
Full Zap Structure
A complete Zapier ChatGPT memory Zap for a customer support use case:
- Zendesk — New Ticket (trigger)
- Webhooks GET — search
zendesk:{requester_email}for past ticket history - Formatter — join results array into text (if needed)
- ChatGPT — draft response with customer history in system prompt
- Webhooks POST — store ticket topic + response summary in
zendesk:{requester_email} - Zendesk — Create Reply — post the draft response
Each new ticket from the same customer starts with full context of every previous interaction.
What to Store in Memory
Store summaries, not raw data. The goal is to give the AI enough context to be relevant — not to reconstruct the full history.
Good examples:
"Customer asked about refund policy for order #1234. Resolved: issued full refund.""Lead interested in enterprise plan. Objection: pricing too high vs. competitor X. Follow-up scheduled.""Onboarding call completed. Customer struggling with API auth setup. Sent documentation link."
Bad examples (too verbose):
- Full email bodies
- Complete chat transcripts
- Raw JSON payloads
2-3 sentences per interaction is the right level of detail. retainr's semantic search finds the relevant ones even at scale.
Common Issues
ChatGPT ignores the memory context: Put the memory block in the system prompt, not the user message. System prompt instructions have higher weight in ChatGPT's attention mechanism.
Formatter step needed: Zapier's ChatGPT action expects text fields. If the Webhooks GET returns an array, Zapier may render it as [object Object]. Add a Formatter → Utilities → Line Formatter step to join the array into readable text.
Empty results on first run: A new customer has no memories yet — the GET returns an empty array. Your system prompt should handle this: "If no prior context exists, respond normally."
Multi-step Zap plan required: Webhooks by Zapier requires Zapier Starter or above. The minimum useful memory Zap needs 4 steps: trigger, Webhooks GET, ChatGPT, Webhooks POST.
Give your AI agents a real memory
Free plan includes 1,000 memory operations/month. No credit card required.
Add memory to your Zapier ChatGPT Zap — free plan →Frequently Asked Questions
Does this work with Zapier's OpenAI action as well as ChatGPT? Yes. The Webhooks GET and POST steps work identically regardless of which AI action you use. Swap ChatGPT for OpenAI, Claude, or Gemini — the retainr memory steps are the same.
Can I use this with Zapier AI by Zapier? Yes. Zapier's own AI actions work the same way — add memory context to the prompt input field and store the output via Webhooks POST.
Does retainr store memories across different Zaps? Yes. retainr is keyed by namespace, not by Zap. If you use customer:{email} in a support Zap and a sales Zap, both Zaps share the same customer memory pool. Use different namespace prefixes to keep them separate if needed.
What is the memory search latency? Typically under 50ms. Semantic search over 100k memories returns in under 50ms; 1M memories under 120ms. It never adds meaningful latency to your Zap.
How do I delete a customer's memory? Add a Webhooks DELETE action to https://api.retainr.dev/v1/memories?namespace=customer:{email}. This removes all memories for that namespace. Use it for GDPR deletion requests.
Add memory to any Zapier Zap
Building with a specific app? These guides cover the exact setup:
- Zapier Gmail AI Memory — Per-sender memory for Gmail AI Zaps
- Zapier HubSpot AI Memory — CRM-aware AI memory for HubSpot Zaps
- Zapier Zendesk AI Memory — AI memory for Zendesk support Zaps
- Zapier Airtable AI Memory — Persistent memory for Airtable Zaps
- Zapier WhatsApp AI Memory — Persistent memory for WhatsApp Zaps
- Zapier Slack AI Memory — Persistent memory for Slack Zaps