How to Use ChatGPT & AI in Zendesk (2026)
"Can I just use ChatGPT inside Zendesk?" is one of the most common questions support leaders ask us — and the honest answer is that there isn't one button for it. There are four genuinely different routes, and the right one depends on whether you want Zendesk to run the AI for you, whether you want to wire up OpenAI yourself, and how much control you need over the model and the cost.
This guide walks through all four: Zendesk's own generative AI, the ChatGPT/OpenAI apps in the Zendesk Marketplace, third-party AI agents and copilots that layer GPT on top of your helpdesk, and the build-it-yourself API route. We sell an AI agent ourselves (Macha — more on that, honestly, in route 3), so treat this as a practical map from people who watch this market closely, not a neutral encyclopedia. Each route gets real pros, cons, and the data-privacy notes that actually matter.
First, the one thing to know about Zendesk and OpenAI
Zendesk already uses OpenAI under the hood. Zendesk runs a multi-LLM architecture — pulling from OpenAI, Microsoft Azure OpenAI, Amazon Bedrock, and Google Cloud depending on the feature — and it has a public partnership with OpenAI to power features like generative replies and AI agents (OpenAI). So when people say "I want ChatGPT in Zendesk," they're often already half-there: the native features are GPT-powered, just wrapped in Zendesk's own product and billing.
That matters for two reasons. First, you may not need to bolt anything on. Second, the privacy posture is better than people assume — Zendesk states it has a zero-data-retention agreement with OpenAI, and that third-party LLM providers do not use your Zendesk data to train their models; inputs are processed ephemerally to generate a response (Zendesk help — AI Data Use). If you wire up your own ChatGPT integration instead (routes 2 and 4), you own that privacy posture — so read the privacy note in each section.
Route 1: Zendesk's own generative AI (the no-code default)
This is the path of least resistance. Zendesk ships GPT-powered features inside Agent Workspace and its messaging/ticketing, split into two buckets.
Agent Copilot helps your human agents. It drafts and expands replies (turn bullet points into a polished response), rewrites tone with one click, summarizes long ticket threads so an agent can catch up fast, and suggests next steps. It also powers intelligent triage — auto-detecting intent, sentiment, and language so you can route tickets.
AI agents (Zendesk's rebrand of what used to be "Answer Bot") handle conversations end to end. They read a customer's question, search your connected knowledge — Help Center articles plus external sources via the web crawler — and reply to deflect or resolve the ticket without a human. The newer AI agent builder lets you define procedures in natural language; the agent then plans a course of action with adaptive reasoning and shows a preview of its proposed steps before you publish.
How to turn it on (high level):
- Confirm your plan. Core Suite plans start around $19/agent/mo (Support Team) up to $55 (Suite Team), billed annually (Zendesk pricing).
- Add the AI capability. The Copilot add-on is $50/agent/mo (annual) — and note it's not available on Support Team or Suite Team, so you may need to be on a higher Suite tier first (Zendesk help — Buying Copilot).
- Build your knowledge base. AI agents are only as good as the Help Center and sources you connect — thin docs produce a thin agent.
- Configure agents, triage rules, and escalation, then test before going live.
The cost catch — read this carefully. Beyond seats and add-ons, Zendesk charges per automated resolution: roughly $1.50 per resolution on committed volume and $2.00 pay-as-you-go, with rates dropping at higher committed tiers (eesel AI). Since January 2026, overages above your committed volume are auto-billed with no cap and no grace period, so usage spikes hit the invoice directly. In May 2026 Zendesk softened what counts: under a three-tier model, Assisted Escalation and Contained Resolution are free, and only Verified Resolutions draw down your allowance. Also in flux: Zendesk is absorbing the old $50 Advanced AI add-on into the Suite/Support plans (rollout May 11–June 12, 2026) — the separate SKU is retiring, but the per-resolution overage that drives most of the spend stays (eesel AI). These plan details are mid-rollout as of June 2026; verify against your own Zendesk billing page before you budget.
Pros: zero integration work; native to Agent Workspace; strong privacy posture; one vendor, one invoice. Cons: per-resolution pricing gets unpredictable at scale; answers are largely confined to what's in Zendesk; you don't choose the model. Privacy: handled by Zendesk's zero-retention agreements — the safest of the four routes.
Route 2: ChatGPT apps from the Zendesk Marketplace
If you want raw ChatGPT behavior — your own prompts, your own OpenAI key — without engineering, the Marketplace has dozens of apps that drop a GPT panel into the agent sidebar. Most are bring-your-own-API-key: you paste an OpenAI key, and the app calls the ChatGPT API on your behalf.
Common examples:
- Triggers+ChatGPT — run ChatGPT workflows that categorize, translate, or parse tickets (Marketplace).
- ChatGPT Assistant by Canadesk — summarizes messages, analyzes sentiment, and recommends replies/macros from your own prompt (Marketplace).
- AI Agent Assist — uses the ChatGPT API with custom and predefined prompts for consistent, on-brand replies (Marketplace).
How to set up (typical flow):
- Get an OpenAI API key from your OpenAI account (this is a separate, usage-based bill).
- Install the app from the Zendesk Marketplace into your account.
- Paste your API key and configure prompts/macros.
- Use the panel from the ticket sidebar to summarize, draft, or translate.
Pros: cheap to start (you pay OpenAI usage, often a few dollars to dozens per month); full control over prompts and which GPT model you use; fast to install. Cons: quality and support vary widely between third-party developers; most are copilots for agents, not autonomous resolvers; you're stitching together two vendors. Privacy — important: when you bring your own OpenAI key, Zendesk's zero-retention agreement does not automatically cover that traffic — your data goes to OpenAI under your account terms. Confirm the app's data handling and your own OpenAI data-retention settings before sending real customer data (eesel AI — Zendesk ChatGPT).
Route 3: Third-party AI agents that layer GPT on top of Zendesk
This is the middle ground between "use whatever Zendesk gives you" and "build it yourself." A class of tools connects to your Zendesk, ingests your knowledge from more places than the native agent reaches, lets you pick the underlying model, and either drafts replies or resolves tickets autonomously — while leaving Zendesk as your system of record.
Macha is one honest option here, and it's worth being clear about what that means. **Macha is an AI agent layer on top of Zendesk — not a Zendesk replacement.** Tickets, customers, and reporting stay in Zendesk; Macha adds the AI resolution layer over it. Teams reach for this route when they want answers grounded in sources beyond the Help Center (docs, Notion, Slack, past tickets), more control over model choice and behavior, and pricing structured around AI actions rather than per-resolution overages. Other tools play in this space too (eesel, myaskai, and others), so evaluate a couple against your own ticket mix — the right fit depends on your channels, your knowledge sources, and how much autonomy you want to grant. You can see how the layer works on the Macha for Zendesk page, and our broader rundown of AI agents for Zendesk compares the category.
How to set up (typical flow):
- Connect the tool to Zendesk via OAuth.
- Point it at your knowledge sources (Help Center, websites, docs, internal wikis).
- Configure agents/behaviors and choose a model.
- Run in copilot/draft mode first, then enable autonomous resolution once you trust it.
Pros: broader knowledge grounding than native; model flexibility; keeps Zendesk as the source of truth; often more transparent pricing. Cons: another vendor in the stack; quality still depends on your documentation; you should pilot before trusting full autonomy. Privacy: depends on the vendor — check for SOC 2, a DPA, data residency, and whether they (or their model providers) train on your data. Pricing note: Macha is credit-based, where credits are spent per AI action (the model you pick changes the cost), not per resolution — so you're paying for automation work done, not a flat outcome fee. You can start on the 7-day free trial, no credit card required to see real behavior on your own tickets before committing.
Route 4: The OpenAI API + automation platforms (build your own)
If you have engineering resources — or just want a no/low-code automation — you can wire Zendesk's API to the OpenAI API directly. The two common variants:
- Automation platforms (no/low-code): Zapier, Make, n8n, or Pipedream all have Zendesk and OpenAI connectors. A typical recipe: new Zendesk ticket → send body to OpenAI for summary/classification → post the result back as an internal note or set a tag (Make, Pipedream).
- Fully custom: your developers call the Zendesk REST API (triggers/webhooks fire on ticket events) and the OpenAI API, with your own prompt and business logic in between.
Pros: total control over prompts, models, and routing; can encode bespoke business rules; no per-seat AI markup. Cons: you own the build and the maintenance — prompt drift, model upgrades, error handling, and rate limits all become your problem; no one to call when it breaks. Privacy: entirely on you. You're sending customer data to OpenAI under your own account, so set zero-data-retention where eligible, redact PII where you can, and document the data flow for GDPR (eesel AI — GDPR).
Which route should you pick?
| Route | Best for | Effort | Who controls the model | Watch-out |
|---|---|---|---|---|
| 1. Native Zendesk AI | Teams wanting fastest, lowest-risk start | Low | Zendesk | Per-resolution overages at scale |
| 2. Marketplace ChatGPT apps | Agent-side copilots on a budget | Low | You (your key) | Variable quality; your privacy posture |
| 3. Third-party AI agent layer (e.g. Macha) | Broader knowledge + model choice, Zendesk stays the system of record | Medium | You (pick model) | Pilot before full autonomy |
| 4. OpenAI API / automation | Custom logic, full control | High | You | You own the build and upkeep |
A reasonable default: start with Route 1 to see how far native AI gets you, add a Route 2 app if you just need agent-assist on the cheap, and graduate to Route 3 when you want resolutions grounded in knowledge beyond the Help Center or pricing that scales with actions rather than outcomes. Reserve Route 4 for genuinely custom workflows where you have engineering to maintain it.
Frequently asked questions
Does Zendesk use ChatGPT/OpenAI? Yes. Zendesk's native generative AI runs on a multi-LLM architecture that includes OpenAI, alongside Azure, Amazon Bedrock, and Google Cloud, under an official OpenAI partnership.
Is my customer data used to train OpenAI's models? Not for Zendesk's native AI — Zendesk states it has a zero-data-retention agreement with OpenAI and that third-party LLMs do not train on your data. If you bring your own OpenAI key (Marketplace apps or the API), that traffic runs under your OpenAI terms, so set retention and PII handling yourself.
Can ChatGPT actually reply to and resolve Zendesk tickets automatically? Yes — via Zendesk's own AI agents (Route 1) or a third-party AI agent layer (Route 3). Most Marketplace apps (Route 2) assist human agents rather than resolve autonomously.
How much does AI in Zendesk cost? Native AI adds the Copilot add-on (~$50/agent/mo, annual, not on the lowest tiers) plus per-resolution charges (~$1.50 committed to $2.00 pay-as-you-go). Marketplace/API routes mostly cost your OpenAI usage. Verify current numbers on Zendesk's pricing page — the AI plan structure is mid-change in 2026.
Do I need to be a developer? No for Routes 1–3. Route 4's automation-platform variant is low-code; the fully custom variant needs engineering.
The bottom line
There is no single "ChatGPT button" in Zendesk, but there are four solid ways to get LLM help into your helpdesk. The native features are the fastest and most privacy-clean start; Marketplace apps are the cheap agent-assist option; a third-party AI agent layer (Macha among them) gives you broader knowledge grounding and model choice while keeping Zendesk as your system of record; and the API route trades control for maintenance burden. Match the route to how much control, autonomy, and predictability you actually need — and pilot before you let any of them answer customers unsupervised.
If the AI-agent-layer route fits, you can start a 7-day free trial, no credit card required and watch it work on your own tickets before committing.
Sources: Zendesk pricing, Zendesk AI Data Use, Buying the Copilot add-on, Moving to automated resolutions, OpenAI × Zendesk, and Zendesk Marketplace listings linked above. AI features and pricing change quickly — figures verified June 2026; re-check before budgeting.
