# AI Policy Template for Small Teams: Editable DOCX

> Free editable AI policy template (DOCX) for small teams: approved tools, data rules, who reviews what, outside actions and incident steps.

- Canonical URL: https://techvendorly.com/ai-policy-template/
- Author: Anil Rawat
- Published: 19 November 2026
- Updated: 21 November 2026
- Publisher: TechVendorly (https://techvendorly.com/)

Your team already uses ChatGPT, Gemini or Copilot, with or without a policy. A short written policy keeps client data safe and makes it clear who checks the output. This page gives you an editable Word template and shows how to fill it in for a small team.

THE SHORT ANSWER

## What should a small business AI policy include?

Keep it to one page. List the **approved tools**, say what **data** staff may share, name **who checks** each kind of output and which actions need approval, and write down **what to do** when something goes wrong. Download the editable template below, replace the example and review it whenever your tools change.

Our verdict

Do this weekDownload the template, list your approved AI tools and name one owner for each.

Must includeWhat data staff may share, who checks the output and what to do when something goes wrong.

Not a substituteFor legal advice. Ask a professional before using AI on regulated, financial or HR decisions.

Template and guidance reviewed 18–20 Nov 2026 · [How we researched](#how-we-researched)

The TechVendorly decision mapDecision route
- 01**Define permission**Tools, tasks and allowed data
- 02**Name the reviewer**Claims and actions need owners
- 03**Keep a record**Approval, incidents and changes

*Rule of thumb*Text version appears in the article.

**TechVendorly**Original editorial framework. Sources and examples in the guide below.

### How we built this template

The template is TechVendorly's own. We updated this guide on 20 November 2026 using the NIST AI Risk Management Framework, the NIST generative AI profile and the OWASP list of risks for LLM applications. We checked the data-control pages for ChatGPT and Microsoft 365 Copilot, and India's DPDP Rules notification, between 18 and 20 November 2026. It has not been reviewed by a lawyer. [Read our evidence standards](https://techvendorly.com/how-we-test/).

New to AI at work? Start with [the main guide to AI in a small business](https://techvendorly.com/ai-for-small-business/).

## Download the template

The Word file includes an example you can overwrite, approval fields and a review record. Fill it in, then share it with the team.

[tv_download_card file="ai-policy-template.docx"]

What goes in a one-page AI policy

The six things a one-page AI policy should cover. TechVendorly guidance.

### Show the data as a table

| Step | What happens |
| --- | --- |
| 1. Approved tools | Service, plan and account owner |
| 2. Allowed data | Four groups, one rule each |
| 3. Who checks | Reviewer by type of output |
| 4. Actions that need approval | Sending, paying, publishing |
| 5. When things go wrong | Stop, record, tell the owner |
| 6. Review date | And what triggers an early review |

## A filled-in example you can adapt

Here is how the first part of the template might read for a five-person marketing agency. It is an example, so replace every detail with your own.

| Section | Example wording |
| --- | --- |
| Approved tools | ChatGPT Business (owner: Priya, operations lead) and Gemini in our Google Workspace. No other AI tools for client work. |
| Allowed tasks | First drafts of blog outlines, social posts and email replies from approved briefs. Summaries of our own meeting notes. |
| Not allowed | Final prices, contract terms, legal or tax answers, and anything sent to a client without review. |
| Data rules | Public and internal material is fine in approved tools. Client personal data only with the client's permission. Never passwords or payment details. |
| Who checks | The account manager approves client-facing text. The owner approves anything with a price or promise. |
| If something goes wrong | Stop, tell Priya the same day, and note what was shared. Priya decides who else must be told. |
| Review | Every six months, or when we add a tool, plan or integration. |

Fictional example. Names and tools are placeholders.

Notice how each line answers a real question a colleague might ask. That is the test for every rule you add.

## Write rules people can use mid-task

A policy works only if someone can use it while writing an email or a blog draft. Test each rule with one question: would it help a colleague decide what to do right now? If not, rewrite it.

Keep it short. One clear page that people read beats ten pages nobody opens.

## List the approved tools and tasks

Name the exact service and plan, who owns the account and what it may be used for. Approval for a chat account does not cover a connector that reads your shared drive. Each new connection needs its own check.

| Field | What to record | Example |
| --- | --- | --- |
| Approved use | What staff may ask the tool to do | Draft a reply from the service factsheet |
| Approved account | Service, plan and account owner | The company's business workspace |
| Not allowed | What the tool must not decide or do | No final quote, refund or legal answer |
| Reviewer | Who checks and approves the output | The person who owns the customer promise |
| Review trigger | What needs a fresh check | A new integration, data type or plan |

## Sort your data before anyone pastes it

Use four groups your team already understands, and give each one a rule.

What staff may paste into an AI tool

Example classification from the template. Adjust the groups to your own data, and check each tool's data settings before approving it.

### Show the data as a table

| Function | Use | Risk level |
| --- | --- | --- |
| Public | Your website text | Low risk |
| Public | Published prices | Low risk |
| Internal | Service factsheets | Needs review |
| Internal | Meeting notes | Needs review |
| Private | Customer names and emails | Needs review |
| Private | Staff records | High risk |
| Restricted | Passwords and API keys | High risk |
| Restricted | Card or bank details | High risk |

- Never paste passwords, API keys or payment details.
- Use made-up or anonymised examples when you try a new tool.
- Check permission before using customer, staff or partner details.
- Share only what the task needs, not a whole folder.

Read the data settings for the exact plan you use. Model training and data retention are separate settings. For example, see [ChatGPT's data controls](https://help.openai.com/en/articles/7730893-data-controls-faq) and [Microsoft 365 Copilot's privacy page](https://learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-privacy).

## Match the reviewer to the risk

A social post and a tax answer carry different risks. Say which drafts need an editor, which need an expert and which tasks stay with a person. Whoever approves a public claim must be able to check its source.

Use [the output review checklist](https://techvendorly.com/ai-output-review-checklist/) for claims, gaps, sums and promises. Asking a second AI tool to check the first is not an independent review.

## Keep a person before anything leaves the business

Drafting and doing are different permissions. A tool may write an email but not send it. It may suggest a database change but not make it. It may read an invoice but not approve the payment. Put a human check before each of those actions.

For automated workflows, record the source, the approval, where the result went and proof it arrived. The [workflow guide](https://techvendorly.com/ai-workflow-guide/) shows how.

## Plan what to do when it goes wrong

- **Stop** the task or workflow.
- **Record** what happened, without copying the sensitive data again.
- **Tell** the policy owner the same day.
- **Check** what was shared and who could see it.
- **Ask** an expert whether a client, staff member or regulator must be told.
- **Fix** the rule so it does not happen again.

Do not promise a client that data is deleted until you have checked with the provider.

## Legal rules: get advice for your use

NIST and OWASP help you think about risk. They do not certify this template. Privacy, employment, copyright and sector rules depend on your business and how you use AI.

In India, the [DPDP Rules notification](https://www.meity.gov.in/static/uploads/2025/11/53450e6e5dc0bfa85ebd78686cadad39.pdf) brings different rules into force on different dates. As of 18 November 2026, not all main rules were in force. Do not describe this template as "DPDP compliant". Check which rules apply to your use, with a professional if needed.

## Roll it out in one meeting

Name the owner, the start date and the next review date. Then walk the team through three real cases: drafting from public material, using an internal document and replying to a customer. Afterwards, each person should be able to say what is allowed, what must stop and whom to ask.

Keep the policy, approved tool list and incident steps where people work, such as a pinned note in your team chat. Review it when you change tools, plans or data, not just once a year.

## Related guides

- [AI output review checklist for a small team](https://techvendorly.com/ai-output-review-checklist/)
- [ChatGPT for Small Business: Uses, Prompts and Limits](https://techvendorly.com/chatgpt-for-small-business/)
- [AI for Small Business: 25 Uses, Tools, Costs & Safeguards](https://techvendorly.com/ai-for-small-business/)

## Questions readers ask

### Is this template legal advice?

No. It is an operating template. Have a professional review it before you use AI for legal, financial, HR or regulated work.

### Is it suitable for a team of two or three?

Yes. Keep only the rules your tasks need. A short rule people follow beats a long document nobody reads.

### Does it make us DPDP compliant?

No. Compliance depends on how you process personal data and which rules are in force. The template is a starting point, not a certificate.

### Do free AI tools need approval too?

Yes. What matters is the data shared and the action taken, not whether the plan is free.

### How often should we update it?

Whenever your tools, plans, permissions, data or tasks change, and at least at a set review date. Record who approved each change.

### Who should own the policy?

One named person, usually the owner or operations lead. They approve tools, handle incidents and run the reviews.

## Sources

Read 18–20 November 2026.

- [NIST: AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework)
- [NIST: generative AI profile](https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence)
- [OWASP: risks for LLM applications](https://owasp.org/www-project-top-10-for-large-language-model-applications/)
- [OpenAI: ChatGPT data controls](https://help.openai.com/en/articles/7730893-data-controls-faq)
- [Microsoft: 365 Copilot privacy](https://learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-privacy)
- [MeitY: DPDP Rules notification](https://www.meity.gov.in/static/uploads/2025/11/53450e6e5dc0bfa85ebd78686cadad39.pdf)

Last editorial update: 20 November 2026.
