AI policy template for a B2B marketing team, with every clause filled in
A two-page AI policy template written for a paid and demand gen team, covering approved tools, data classes, ad account access, claims review and labeling.

An AI policy for a marketing team is a short rule set covering which AI tools people may use, what data goes into them, how AI output is checked before it reaches a customer or an ad account, and who approves exceptions. The filled-in AI policy template below fits all of it on two pages you can adopt this week.
Picture the moment it becomes urgent. A specialist wants a lookalike list, so they paste last quarter's closed-won export, names and emails included, into a personal ChatGPT account. Nobody did anything malicious. Nobody had told them not to, either, and now you're working out with legal whether customer data left the building.
This guide is for the CMO, VP of Marketing or head of demand gen who has to sign that policy. The HR and legal templates that rank for this search are written for a whole company; this one is written for a paid and demand gen team, with the clauses those templates skip: ad account access, AI connectors, generated ad creative and claims. Our article on the generative AI risks a CFO or legal team will ask about covers the risk side; this one gives you the rules.
Your team already uses AI, with or without a policy
Microsoft and LinkedIn's 2024 Work Trend Index, a survey of 31,000 knowledge workers in 31 countries, found that 78% of people who use AI at work bring their own tools (Microsoft). In a marketing team that usually means personal accounts on consumer plans, where the data terms are the weakest.
More recent surveys point the same way. Cybernews research from August 2025, covering more than 1,000 US employees, found that 59% use AI tools their employer hasn't approved and that nearly a quarter of companies have no official policy on AI at work, as Techstrong.ai reported. Senior people were the most likely to do it, which is a useful thing to remember before you write a policy only for juniors.
So the policy decides how your team uses AI: on terms you've read, with data you've classified, under rules they can find. A policy that bans AI outright mostly moves the usage somewhere you can't see. With rules people can follow, a team gets to put AI on real campaign work in the open instead of in personal accounts, and helping teams do exactly that, from choosing the first jobs to building and running them, is what Scalewell is for.
A marketing AI policy answers five questions
The policies that ranking pages offer tend to run to ten pages of principles. A marketing team needs answers to five practical questions, and every clause in the template maps to one of them.
| Question | What the policy decides | Who usually owns it |
|---|---|---|
| Which tools? | The approved AI tools and the plans they must be on | Head of marketing ops, with IT |
| What data? | Which data classes may go into which tools | Marketing ops, with legal or privacy |
| What checks? | How AI output is reviewed before it's published or goes live | The person publishing, plus brand and legal for claims |
| What access? | Which AI connections may read or change ad accounts and the CRM | Head of paid media |
| What disclosure? | When AI involvement must be labeled, and who owns the output | Legal, with the CMO |
If you can answer those five on one page, you have a working policy. The rest is examples and an exceptions process.

The filled-in AI acceptable use policy for a marketing team
An AI acceptable use policy is the part of the policy that tells people what they may and may not do. The version below is written for a hypothetical B2B SaaS company with a ten-person marketing team running Google Ads, LinkedIn Ads and Microsoft Advertising, with HubSpot as the CRM. Replace the bracketed parts with your own names and tools, and have legal read it before you send it.
1. Purpose and scope
This policy covers every member of the marketing team, contractors and agencies working in our accounts. It applies to any AI tool used for marketing work, including chat assistants, AI features inside ad platforms and martech tools, and AI connections to our accounts. Its aim is to let the team use AI for real work while keeping customer data, ad spend and our claims under control.
2. Approved tools
| Tool | Plan required | Approved for |
|---|---|---|
| [Claude or ChatGPT] | [Team or Business plan, company workspace only] | Drafting, analysis, research, summarizing exports |
| AI features inside Google Ads, LinkedIn Campaign Manager and Microsoft Advertising | Our company ad accounts | Asset suggestions, audience expansion, bidding, within the settings the head of paid media approves |
| AI features inside [HubSpot] | Our company portal | Drafting emails, summaries, scoring |
| [Creative tool name] | [Business plan] | Draft creative, size variants, background edits |
Personal accounts on any AI tool may not be used for company work. New tools need approval from [head of marketing ops] before first use, using the request form in section 9.
3. Data rules
| Data class | Examples | Allowed in approved tools? | Allowed in unapproved or personal tools? |
|---|---|---|---|
| Public | Published blog posts, public ads, competitor ads | Yes | Yes |
| Internal | Campaign performance, budgets, plans, ad copy drafts | Yes | No |
| Confidential | Pricing not yet public, board materials, unreleased product details | Yes, with care; never in prompts shared outside the workspace | No |
| Customer personal data | Names, emails, phone numbers, CRM contact records, lead lists | Only through an approved connection or with identifiers removed | Never |
| Regulated or special category data | Health, financial account or government ID data | Never | Never |
When in doubt, remove names, emails and company names before you paste anything. A campaign export with spend and conversions is internal data; the same export with lead emails attached is customer personal data.
4. Review before anything goes live
- Every AI-drafted ad, landing page, email or post is read in full by a person before it's published.
- Any claim with a number, a comparison, a customer name or a guarantee goes through the same claims check as copy written by hand, against [our claims register].
- AI-generated images or video of people, products or customer logos need sign-off from [brand lead] before use.
- Nobody may publish a testimonial, review or quote that a real customer didn't give.
5. Ad account and CRM access
- AI connections to ad accounts and the CRM start read-only.
- Write access for an AI connection needs written approval from [head of paid media], naming the account, the actions allowed and the person who reviews them.
- Budget increases, new campaign launches, placement or exclusion list changes and changes to conversion settings are always approved by a named person before they go live, whatever tool proposes them.
- Platform AI settings are reviewed when a campaign launches and once a quarter, including text customization in Google Ads (formerly automatically created assets, found in each Search campaign's AI Max settings) and Enable Audience Expansion in the Audience section of each LinkedIn ad set, which is on by default where it's available.
6. Disclosure and labeling
- AI-generated people in ads are labeled where the law requires it, and [legal] keeps the list of markets that do.
- We don't present AI-generated content as the personal work or words of a named employee or customer unless that person has approved it.
- Content that is mostly AI-written may be published under the company name or a named author who has reviewed and edited it.
7. Ownership and records
- Work produced with AI tools for the company belongs to the company, subject to each tool's terms.
- Prompts, skills and workflows built for recurring jobs are stored in [shared workspace], not in personal accounts.
- [Head of marketing ops] keeps a list of AI tools in use, their plans and their data terms, reviewed every quarter.
8. Mistakes and incidents
If customer data goes into an unapproved tool, or an AI-made change goes live by mistake, tell [head of marketing ops] the same day. Reporting quickly is expected and won't be held against anyone; hiding it will be.
9. Requesting a new tool or an exception
Send [head of marketing ops] the tool name, the plan, the job it does, the data it would see, and a link to its data terms. Expect an answer within five working days.
10. Review
This policy is reviewed every six months and whenever we add a tool that reads or changes an ad account.
The data rule is the clause to write most carefully
The data clause matters most because it's the one the team will test every day. The useful line to draw is between consumer plans and business plans, because that's where the vendors' own terms change. OpenAI's help center says that by default it doesn't use inputs or outputs from business offerings such as ChatGPT Business, Enterprise and the API to train its models (OpenAI), and Anthropic's Commercial Terms govern its business plans separately from its consumer terms.
That's why the template ties every approved tool to a plan. "You may use Claude" is a weak rule; "you may use Claude in the company Team workspace" is one a person can follow and an auditor can check. Read the current terms yourself before you approve a tool, because they change, and keep a link to the version you approved.
The second half of the rule is about identifiers. Most marketing analysis doesn't need names or emails. A spend and conversion export by campaign answers nearly every question a paid team asks, so make removing identifiers the habit and the CRM connection the exception.
Ad account access needs its own clause
Ad account access is where a marketing AI policy differs most from a company one. An AI connection that can change bids or budgets can spend money, and company-wide HR templates rarely mention it.
The connection routes vary in what they can do, which is why the template starts everything read-only. Google's own Google Ads MCP server can create assets, upload videos and manage labels with propose-and-confirm validation, but its developer page says it "cannot create, modify, pause, or enable campaigns, ad groups, ads, budgets, or bids", while LinkedIn's own MCP server, which reads and writes ad accounts, is in beta and in use but not yet publicly documented as of November 2026. Our guides to the Google Ads MCP server and the LinkedIn Ads MCP routes cover what each route can change.
The four decisions in clause 5, budget increases, launches, placement lists and conversion settings, are the ones we'd always keep with a person. Our article on where humans stay in the loop on ad spend explains why each of those deserves a named signature even when the platform's own AI proposes it.
Generated content needs a claims check, and sometimes a label
AI-written copy carries the same legal risk as copy written by your team, so the review clause applies the claims register to both. The model doesn't know which customer stories you're allowed to tell or which numbers legal has cleared, and it will happily write a confident statistic that nobody can source.
Generated people are where labels now come in. New York's General Business Law section 396-b, in force since June 9, 2026, requires anyone who creates an ad with actual knowledge that it contains a synthetic performer to "conspicuously disclose" it, with civil penalties of $1,000 for a first violation and $5,000 for each one after, per the New York State Senate's text. The FTC's 2024 rule on fake reviews and testimonials separately covers reviews attributed to people who don't exist (FTC).
Keep the market list in the policy short and owned by legal, and keep the platform settings in a separate guardrail document. Our AI brand safety policy maps the settings and defaults in each ad platform that the policy's rules depend on.
Roll the policy out in one week
A policy nobody reads changes nothing, so the rollout matters as much as the wording. This is the order we'd use for a ten-person team.
- Monday: list what's in use. Ask everyone, in a no-blame message, which AI tools and plans they use for work. You'll find more than you expected, and the list becomes section 2.
- Tuesday: settle the plans. Move everyone onto the approved business plan and close the personal accounts used for work. This is the step that changes the data terms.
- Wednesday: fill in the template. Replace the brackets, cut anything that doesn't apply, and send it to legal for a read.
- Thursday: walk the team through it. Spend thirty minutes on the data table and the access clause with real examples from your own accounts. People remember examples, not principles.
- Friday: check the ad accounts. Review which AI connections and platform AI settings are switched on in each account, and make them match clause 5. In Google Ads, people and app access sits under Admin, then Access and security; in LinkedIn Campaign Manager, under Manage access in the account menu. Connectors that sign in with a person's own login may not show there, so ask whoever set each one up.
Training then makes the policy stick. Our guide to AI training for marketing teams covers how to build the policy into role-based training, so the rules show up in the work instead of in a PDF.
AI policies fail when they ban, sprawl or never get updated
The most common failure is a ban. A policy that forbids AI tools without offering an approved one pushes the same work onto personal accounts, which is the outcome the policy was meant to prevent. Approve at least one tool on a business plan on the day the policy goes out.
Writing for the whole company. A company-wide policy written by HR rarely mentions ad accounts, connectors or ad claims. If yours already exists, add a marketing annex with clauses 2 to 6 instead of waiting for a rewrite.
Too long to use. Twelve pages of principles get skimmed once. Keep the working policy to two pages, with the data table and the access rules on the first.
No owner and no review date. Platforms add AI features every quarter, and several switch on by default. A policy without a named owner and a review date describes last year's tools within six months.
How Scalewell turns an approved tool list into work taken off the team
A policy gives the team permission, and the hours come back once the approved tools take recurring work off people's desks. Scalewell's first step is a free 30-minute conversation. Next, the team's week gets audited, listing the recurring jobs and the time each one takes, and it turns into a ranked roadmap of the jobs to hand to AI, in order. The build sprint then works inside the tools your team already uses rather than adding new ones to the approved list, and it typically runs 8 to 12 weeks. It finishes with your own people trained to run every workflow without Scalewell.
Where to start
List the AI tools your team uses today, move them onto business plans, and send the filled-in template above to legal with your names in the brackets. When you're ready to choose which jobs AI should take on under the new rules, Scalewell is glad to talk that through on a free call.
Frequently asked questions
Do agencies and contractors have to follow our AI policy?
Yes, if they work in your ad accounts, CRM or brand assets. Name them in the scope clause, add the policy to their contract or statement of work, and ask them for their own list of AI tools and plans before they start. Their AI connections to your accounts should go through the same write-access approval as your team's.
What is the difference between an AI policy and an AI acceptable use policy?
An AI policy is the full rule set, including ownership, governance, review dates and who decides on new tools. An AI acceptable use policy is the part staff read day to day: what they may and may not do with AI tools and data. Many marketing teams combine the two in one short document, as the template above does.
Is an AI policy the same as an AI governance framework?
Not quite. A governance framework is the wider system: who decides on AI use, how risks are assessed, what inventory of tools and models is kept and how often it's audited. For a marketing team, the policy, its tool list and its review date usually are the practical governance, with the company framework sitting above it.
Should employees sign the marketing AI policy?
A written acknowledgement helps, because it shows each person received the current version. Ask everyone, including contractors, to confirm by email or in your HR system when the policy goes out and again after each six-month review, and keep the confirmations with the policy's owner.
How often should a marketing AI policy be updated?
Review it every six months, and update it whenever the team adds a tool that can read or change an ad account or the CRM. Ad platforms add AI features regularly, and some switch on by default, so the review should include checking each account's AI settings against the policy.