# ChatGPT Business Agent Templates You Can Adapt (2026)

> Copy-ready ChatGPT Business Agent templates for ecommerce and online-service businesses, with placeholders, add-on blocks and a test routine to run before you publish.

Canonical: https://revnu.com/blog/chatgpt-business-agent-templates-you-can-adapt-2026
Published November 19, 2026 by George Jefferson, Co-founder and CEO, Revnu

# ChatGPT Business Agent Templates You Can Adapt (2026)

Yes, you can adapt a ChatGPT Business Agent template: write the instructions as a short set of blocks (role, sources, tone, boundaries, escalation), fill in your own details, then test the agent against real customer questions before you publish. The templates below are starting points to edit, not a final script. They use [placeholders] for everything specific to your business.

A template can't supply facts. A Business Agent answers from the product and policy information your business provides, so the instructions only shape how it answers. If your return window is wrong in your sources, a polished instruction block will state the wrong window politely.

This guide gives you two full templates (ecommerce and online services), three add-on blocks to paste in, and a test-and-revise routine. If you're still working out whether your business can have an agent at all, start with [ChatGPT Business Agents: Eligibility and Setup](https://revnu.com/blog/chatgpt-business-agents-eligibility-and-setup).

## Are there instruction templates for a ChatGPT Business Agent?

There is no single official fill-in-the-blank template you must use, so the practical answer is to build your own from a proven structure. OpenAI's setup guidance has you review your business information and instructions, preview the agent with common customer questions, and fix anything inaccurate before you publish. The templates in this post follow that flow.

Treat every block as a draft. Replace each [placeholder], delete lines that don't apply, and add the rules only you know, such as a promise you never make or a product you never discuss.

A template can't do four things:

- It can't make your product, shipping or returns information correct. Fix those at the source.
- It can't guarantee how the agent will answer every question. OpenAI's terms note that agents can produce inaccurate or unexpected responses, which is why you test before publishing and keep checking afterward ([OpenAI Businesses in ChatGPT Terms](https://openai.com/policies/businesses-in-chatgpt-terms/)).
- It can't widen what you're eligible for. Business Agents are currently offered to eligible US ecommerce and online-service businesses with a website, and your agent must meet OpenAI's Business Agent policy.
- It can't give the agent powers you haven't enabled. Don't tell it to issue refunds or change accounts unless that capability is actually switched on.

The terms put responsibility for what you publish on you. Write instructions you'd be comfortable seeing quoted back to a customer.

### Can I reuse a custom GPT instructions template for a Business Agent?

Yes, the structure carries over, but the content has to change. A custom GPT template and a Business Agent template both need a role, scope, tone, and rules for what to refuse. The difference is audience: a Business Agent speaks to your customers as your company, so it needs your real shipping, returns, pricing, and handoff rules. Start from the templates below, then replace every generic line with a fact from your own site or policies.

## What a good Business Agent instruction set contains

A good instruction set has five parts, and each one answers a different failure you'll see in testing.

**Role.** One sentence on who the agent is and whom it serves: "You are the customer-support assistant for [business name]." Skip long personas. A clear job beats a clever character.

**Sources.** Tell the agent to answer using only the information you've provided, and name what that includes: catalog, shipping policy, returns policy, service descriptions, pricing page. This line does more to prevent invented answers than any other.

**Tone.** Two or three adjectives and one structural habit, for example "friendly and concise; ask one short follow-up question when the request is unclear." Long tone guides get ignored.

**Boundaries.** What it must not do: guess, promise exceptions, discuss topics outside your business, or claim to take actions it can't take.

**Escalation.** Where unresolved or sensitive issues go: a support email, a form, a phone line. Give a real destination, not "contact us."

Keep the whole set short enough to read aloud in a minute. When an answer goes wrong, a short set makes it obvious which line to change.

### What do OpenAI agent builder instruction examples have in common?

Most good examples share the same parts: a one-line role, the tasks the agent should handle, the tasks it must decline, the tone, and a clear step for passing a person to a human. They also state what the agent should do when it doesn't know something. Keep the same skeleton for your Business Agent, but write the details in your own business's words, because generic examples won't know your products or policies.

### What does a system prompt template for business tasks look like?

A usable template has five short blocks: who the agent is, what it can help with, what it must not do, how it should sound, and what to do when it's unsure. Write each block as plain sentences, not clever prompt tricks. For example, state that it should answer only from your listed policies and offer a human contact when a question falls outside them. The two templates in this post fill in these blocks for a store and a service business.

## Template 1: Ecommerce store (products, shipping, returns, recommendations)

Use this block for a store that sells physical or digital products. It's a starting point: edit every bracket, and cut anything you can't back with accurate source information.

> You are the customer-support assistant for **[business name]**, an online store that sells **[product categories]**. Help shoppers understand our products and policies using only the information provided in **[product catalog, shipping policy, returns policy, sizing or compatibility guides]**.
>
> Tone: **[friendly, concise, plain English]**. Ask one brief follow-up question when a request is unclear, such as size, use case or budget.
>
> Products: Describe items using the details in our catalog. Do not invent specifications, materials, colors or stock levels. If a detail is missing, say you do not have it and point the customer to **[support contact]**.
>
> Shipping: Share shipping options, costs and timeframes exactly as written in **[shipping policy]**. Do not promise delivery dates the policy does not state. Note that **[regions we do not ship to]**.
>
> Returns: Explain eligibility, the **[number]**-day window and the steps as written in **[returns policy]**. If a case falls outside the policy, do not offer an exception. Direct the customer to **[support contact]**.
>
> Recommendations: When a customer describes a need, compare two or three relevant products and explain which stated features fit that need. Do not claim a product is best for everyone, and do not recommend items we do not sell.
>
> Never claim to place orders, issue refunds, cancel orders or change account details unless that ability has been enabled for you.

Fill the shipping and returns lines with your real policy language, not a summary from memory. If your return window differs by product type, say so explicitly: "Items in **[category]** have a **[number]**-day window; all others have **[number]** days." Ambiguity here is the most common source of wrong answers.

## Template 2: Online service business (scope, pricing, booking, qualifying questions)

Use this block for consultancies, agencies, coaching, software services and other businesses that sell work rather than boxed goods. The difference from ecommerce is that most questions are about fit, so the agent needs qualifying questions and a clear route to booking.

> You are the assistant for **[business name]**, which provides **[services in one sentence]** to **[ideal client type]**. Answer questions about our services using only the information provided in **[service pages, pricing page, FAQ, booking policy]**.
>
> Tone: **[professional, warm, direct]**. Keep answers short and ask one question at a time.
>
> Scope: We offer **[list of services]**. We do not offer **[services we decline]**. If someone asks for something outside our scope, say so plainly and, where useful, suggest what we can do instead.
>
> Pricing: Share pricing exactly as published on **[pricing page]**. If a service is quoted individually, say that and explain what the quote depends on: **[factors, e.g. scope, timeline, volume]**. Never invent a price, discount or package.
>
> Qualifying questions: To judge fit, ask about **[the client's goal]**, **[their timeline]** and **[their rough budget range]**, one at a time and only when relevant. Do not interrogate; stop asking once you can give a useful answer.
>
> Booking: When a visitor wants to talk, direct them to **[booking link or contact method]** and summarize in one sentence what they said they need, so the conversation starts with context.
>
> Do not promise outcomes, timelines or availability that our materials do not state. Do not claim to book, schedule or invoice on someone's behalf unless that ability has been enabled for you.

The qualifying questions are the part most worth editing. Write them from the questions your team actually asks on a first call, and drop any your sources can't support.

### What makes a custom GPT instructions example worth copying?

The best examples are specific, short, and testable. They name exact tasks, give real numbers such as prices or turnaround times, and list what to refuse. Avoid examples that only say "be helpful and friendly," since they leave the agent guessing. Judge any example by one test: could you check an answer against it and say right or wrong? If yes, it's worth adapting. The service business template here follows that rule.

## Add-on blocks to paste in: out-of-scope questions, handoff to a human, sensitive and order-specific issues

Add these after the main template. Each one closes a gap that testing usually exposes.

**Out-of-scope questions**

> If a customer asks about something unrelated to **[business name]** or not covered in the information provided, say you can only help with questions about our **[products/services]**. Do not speculate, give general advice outside our area, or comment on other companies.

**Handoff to a human**

> If you cannot answer from the information provided, if the customer asks for a person, or if the customer sounds frustrated twice in a row, stop and give them **[support email, form link or phone number]** along with our hours: **[hours]**. Summarize their question in one sentence so they do not have to repeat it.

**Sensitive and order-specific issues**

> For anything involving a specific order, payment problem, account access, a complaint about a charge, a legal or safety concern, or personal information, do not try to resolve it here. Tell the customer you cannot see or change order and account details in this chat, and direct them to **[secure channel]**. Do not ask them to share payment details or passwords.

The last block matters most. An agent that sounds helpful about a missing parcel but can't actually see the order creates a worse experience than one that hands off cleanly. If you've enabled any real actions, name them in the block so the boundary is precise rather than implied.

### What example prompts help a Business Agent handle tricky questions?

Short, direct rules work best. For out-of-scope topics: "If asked about anything outside our products and policies, say you cannot help with that and offer our contact page." For order issues: "Never guess order status; ask the customer to contact support with their order number." For sensitive cases such as refund disputes or complaints: "Hand off to a human and do not argue." Paste these blocks under your main instructions and adjust the wording to match your brand's tone.

## How to test and revise a template before and after publishing

Build a question set of 25 to 40 real questions before you publish, and run it in the preview. Pull them from your support inbox, live chat logs, sales calls and the reviews your products have received. Cover five groups:

- **Easy facts:** return window, shipping time, what a service includes.
- **Edge cases:** an item past the return window, an international order, a service you decline.
- **Comparisons and recommendations:** "which of these fits my situation?"
- **Traps:** a question that tempts the agent to promise a refund, invent a price or discuss a competitor.
- **Handoffs:** a missing order, a billing dispute, a request for a human.

For each answer, check four things: is it accurate against your source information, is it inside your boundaries, is the tone right, and does it hand off when it should.

Then decide where the fix lives. Most teams skip this step.

| What went wrong | Where to fix it |
| --- | --- |
| The answer states a wrong fact (window, price, spec) | The source information, not the instructions |
| The answer invents something your sources do not contain | Instructions: tighten the "only the information provided" line |
| The agent promises an exception or an action | Instructions: strengthen the boundary |
| Tone is off or answers run long | Instructions: adjust the tone block |
| It does not escalate when it should | Instructions: sharpen the handoff block |

Change one thing at a time and rerun the full set, so you know which edit helped. After you publish, keep a small recurring routine: re-run your question set whenever a policy, price or product line changes, and review real conversations for questions you didn't anticipate. Add each new failure to the question set so it stays tested.

### Does OpenAI's prompt engineering guidance apply to Business Agent instructions?

Yes, the core advice applies: be clear and specific, give the agent a role, separate rules from reference facts, and test with real questions before relying on it. Treat OpenAI's own prompt guidance as the general reference and your Business Agent instructions as the business-specific layer on top. After publishing, read real conversations, find answers that missed, and edit one instruction at a time so you can see which change fixed the problem.

## Where Revnu fits

Revnu is an agent that finds customers and grows revenue for a business across search, AI answers, outbound and ads, and reports to you in Slack.

It helps around the agent. Revnu tracks how your business appears in ChatGPT, Claude, Perplexity and Gemini answers and in Google, publishes the pages that give those answers something accurate to cite, and runs outreach and ads within your approvals. Pro is from $99 a month, self-serve. If you want to see how Hub reporting turns into weekly decisions, read [Turn ChatGPT Business Hub Data Into Revenue Decisions](https://revnu.com/blog/turn-chatgpt-business-hub-data-into-revenue-decisions).

## FAQ details: what the three common questions mean for your agent

The short versions are in the FAQ below. Businesses in ChatGPT is in beta, so recheck OpenAI's current help pages and terms before you rely on any detail here, especially eligibility and which actions an agent can take.

## Start with one template, then let the test set decide

Pick the template that matches your business, fill every placeholder from your real policies, and run it against 25 to 40 genuine customer questions before you publish. Fix wrong facts at the source and wrong behavior in the instructions. Once your agent is answering accurately, point your effort at being found: start Revnu Pro at revnu.com to track how your business appears in AI answers and publish the pages that earn those mentions.

## Related Business Hub guides

- [How to Write Instructions for a ChatGPT Business Agent](https://revnu.com/blog/how-to-write-instructions-for-a-chatgpt-business-agent)
- [What Customers Can Ask a ChatGPT Business Agent](https://revnu.com/blog/what-customers-can-ask-a-chatgpt-business-agent)
- [ChatGPT Business Hub for Professional Services Firms](https://revnu.com/blog/chatgpt-business-hub-for-professional-services-firms)

## Frequently asked questions

### Is ChatGPT an AI agent?

ChatGPT is a conversational assistant, and OpenAI also offers agent-style features that can take multi-step actions. A Business Agent is narrower: it answers customer questions for an eligible business using the information and instructions that business provides. Check OpenAI's current documentation for exactly which actions an agent can perform, since the beta is changing.

### Can I use ChatGPT for customer service?

Eligible US ecommerce and online-service businesses with a website can set up a Business Agent to answer customer questions about products and policies. Write instructions, preview with common questions, and publish only once answers are accurate. Keep a human handoff for order-specific and sensitive issues, since agents can still give inaccurate or unexpected responses.

### Can I sell products through ChatGPT?

Businesses in ChatGPT lets eligible businesses share product information, such as through a product feed or a connected Shopify catalog, so products can appear in ChatGPT. Whether a purchase can complete inside ChatGPT depends on features OpenAI has enabled for your account, so confirm current capabilities in OpenAI's help pages before promising customers anything.

### What is an example of an AI agent?

A Business Agent is one example: it answers shopper questions about a store's products, shipping and returns using the business's own information. Others include AI outbound agents that research prospects and send approved messages, and Revnu, which runs search, AI-answer, outbound and ad work for a business and reports results in Slack.
