The short answer (TL;DR)
A generative AI policy for a marketing team should answer a handful of practical questions: what people may use AI for, what they must never put into it, who checks the output and how the brand’s voice is protected. It can fit on two pages. Google itself asks publishers to fact-check and review AI-generated content before it goes live.
In this article
- What is a generative AI policy for marketing?
- Why teams need a policy now
- No policy vs a two-page policy
- What the policy should cover
- Protecting the brand voice
- How to write your first policy, step by step
- Common mistakes to avoid
- What agencies should add
- Where I can help
- Final thought
- Frequently asked questions
What to remember
- People are already using AI. Only 25% of frontline employees say they get enough guidance (BCG).
- Cover approved uses, off-limits uses, data rules, voice, fact-checking, review, disclosure, images and ownership.
- Every number, name, date and quote gets checked against its source before publishing.
- A person owns the final voice of every piece.
- Keep it short, build it into the review step and revisit it every few months.
For marketing leaders who want their team to move faster with AI without putting the brand at risk.
My training includes crisis and issues management and reputation management, and at BBDO Levant I managed mandates in legally sensitive environments. One thing that work teaches you is that reputational problems usually start small: a line nobody checked, a claim nobody could support. AI makes it faster to produce good work, and also faster to produce those small mistakes.
What is a generative AI policy for marketing?
A generative AI policy is a short set of shared rules for how a team uses AI tools: what for, with which data, under whose review and how readers are told. For a marketing team, its main job is to protect three things at once: client and company data, the accuracy of what gets published and the brand’s voice. In my experience, the longer a policy gets, the less of it people remember.
Why teams need a policy now
People are already using AI, often without clear guidance. In BCG’s 2025 survey, only 25% of frontline employees said their leaders gave them enough AI guidance (BCG, June 2025).
Google’s own guidance is a useful reminder of the risk. It explains that generative models don’t retrieve facts, so their output can contain errors, and it asks publishers to manually fact-check and review AI-generated content before publishing, including titles, meta descriptions and image alt text (Google Search Central, updated October 2026).
No policy vs a two-page policy
| No shared policy | A short, shared policy | |
|---|---|---|
| What’s allowed | Each person decides alone | Approved and off-limits uses are listed |
| Client data | Depends on individual judgment | Clear rules on what never goes into AI tools |
| Accuracy | Errors caught late, if at all | Facts checked against sources before publishing |
| Brand voice | Drifts toward generic phrasing | A person owns the final voice |
| Confidence | People hide their use or avoid AI | People use AI where it helps, openly |
What the policy should cover
| Area | What to decide | Example rule |
|---|---|---|
| Approved uses | Tasks AI can help with | First drafts, variations, summaries, resizing assets |
| Off-limits uses | Tasks AI must not do | Final legal or medical claims, crisis statements, quotes attributed to real people |
| Data | What may never be entered | Confidential client information, unreleased results, personal data |
| Brand voice | How output stays on brand | Voice guide and examples supplied; a person edits every piece |
| Fact-checking | What gets verified | Every number, name, date and quote, against the original source |
| Review and sign-off | Who approves what | Scaled to risk: a caption differs from a product claim |
| Disclosure | When and how AI use is shared | Agreed with clients and in line with platform rules |
| Images and rights | Sources and labels | Licenses checked; AI-generated images labeled |
| Ownership | Who maintains the policy | One named owner, reviewed every few months |
Protecting the brand voice
Voice is where AI output most often drifts. A few habits help:
- Give AI the brand’s voice guide, plus examples of writing the brand would publish and writing it wouldn’t.
- Keep a short list of words and phrases the brand never uses.
- Make a person responsible for the final voice of every piece. AI can draft, but it doesn’t own the sound of the brand.
- Read important pieces aloud before they go out. Generic phrasing is easier to hear than to see.
How to write your first policy, step by step
- Ask the team how they use AI today. Be clear that the point is to learn, not to catch anyone out.
- List the approved and off-limits uses based on what you hear.
- Agree on the data rules with your legal or privacy contact.
- Decide who reviews what, scaled to risk.
- Fit the key rules on one page.
- Test the draft with the people who’ll follow it, and adjust.
- Name an owner and set a review date a few months out.
A policy works when it’s part of the workflow. Cover it in training and build the checks into the review step. I wrote about setting up that step in how to build AI into a marketing team’s workflow, and about deciding where AI belongs in the first place in mapping how your team works.
Common mistakes to avoid
- Writing the policy without the team. Rules written far from the work tend to miss how AI is actually used.
- Banning everything. Strict bans push AI use out of sight, which is where the risk grows.
- Making it too long. People remember a page, not a manual.
- Forgetting images and video. Rights and labeling questions come up just as often as text.
- Never updating it. A policy written for last year’s tools will have gaps today.
What agencies should add
Agencies work with many clients, each with its own rules. An agency policy needs a clear default, plus a simple way to record each client’s preferences: whether AI may be used on their work, what disclosure they expect and which data must stay out of AI tools. The account lead is usually the right owner for that record.
Where I can help
I help marketing teams and agencies put AI governance and brand safety in place: a short policy people actually follow, voice safeguards and review steps that match the level of risk.
Final thought
Ask your team how they’re using AI today, then write down the five rules you’d want everyone to follow tomorrow. That’s the core of your first policy.
Frequently asked questions
Why is it important to establish generative AI usage policies?
Because people are already using AI, and without shared rules each person decides alone what is safe. A clear policy protects client data and the brand’s reputation, and it gives people the confidence to use AI where it helps.
How do you create a generative AI policy?
Start with how your team actually uses AI today. Agree on approved and off-limits uses, data rules, review steps and disclosure, then test the draft with the people who will follow it. Keep it short and review it every few months.
How can marketing teams use generative AI without compromising brand voice?
Give AI the voice guide and good examples, keep a list of words the brand avoids and make a person responsible for the final voice of every piece. A read-aloud check before publishing catches most drift.
How often should a generative AI policy be reviewed?
Every few months is a reasonable rhythm, and sooner when the team adopts a new tool, takes on a client with different rules or when the laws or platform rules that apply to you change. Name one owner so the review actually happens.
Should marketing teams tell clients when they use AI?
Agree on it with each client up front, and record what you agreed. Some clients welcome AI-assisted drafts, others have their own rules or restrictions. Check the platform rules and any laws that apply in your markets too, because disclosure requirements vary and keep changing.
What should never go into an AI tool?
As a starting point: confidential client information, unreleased results, personal data and anything covered by a non-disclosure agreement. Your legal or privacy team should confirm the full list for your company and the markets you work in, and the tools you’re allowed to use.
About the author
Perla Richa is a marketing and communications leader helping teams build AI into how they work. Based in Montreal.
Sources
- Boston Consulting Group, “AI at Work 2025: Momentum Builds, But Gaps Remain,” survey findings press release, June 26, 2025. Read on bcg.com
- Google Search Central, guidance on using generative AI content, updated October 1, 2026. Read on developers.google.com
Not sure yet where AI fits in your team?
That’s a good place to start a conversation.
