What AI should not decide in marketing: brand, offer and approval

In marketing, AI should not decide on its own the brand positioning, the offer and commercial terms, numbers and promises, the use of testimonials and customer case studies, budget allocation across channels and final approval of what goes live. In these areas it can suggest, draft and analyze, but a person decides. The reason is simple: when an ad promises what the company doesn't deliver, the company is accountable, not the tool.

This article details each of these limits and shows how to set up an approval flow that protects the brand without slowing production down. It's part of the guide to AI in marketing for B2B companies.

Why set limits, if AI writes well?

Writing well isn't the same as knowing what can be said. An AI model generates the most likely text for the request it received. It doesn't know which promises the company has already made to customers, which price is valid today, which case study was approved or which topic became a problem last week.

On top of that, AI writes with the same confidence whether it's right or wrong. A made-up number appears in the text as naturally as a real one. Without clear limits and review, the risk isn't a bad text. It's a convincing, wrong text published in the company's name.

Setting limits isn't distrusting the tool. It's dividing responsibility, just as a junior copywriter doesn't approve an ad with pricing on their own.

The six areas that stay with people

1. Positioning and brand

What the company is, who it works for, what sets it apart and how it speaks. AI can help test phrasings, but the choice is strategic and involves information it doesn't have: the company's history, conversations with customers, where the business wants to go.

In practice, this becomes a short document with the company description, the audience, the differentiators, the words the brand uses and the ones it avoids. That document goes into every request made to the AI.

2. Offer, price and terms

What's for sale, for how much, under which terms, timelines and guarantees. These items change, depend on negotiation and have a direct effect on sales. A wrong term in an ad generates leads with expectations the sales rep can't meet.

AI can write the piece from an approved offer. It can't create the offer, give a discount or invent a term to make the copy more appealing.

3. Numbers and promises

Any results figure, comparison, timeline or guarantee needs a source and has to be true. That includes phrases like "the fastest on the market" or "guaranteed results." Besides the reputational risk, promises that can't be backed up often violate platform ad policies and advertising rules.

The practical rule: if there's no source someone at the company can show, the number comes out of the copy.

4. Testimonials, case studies and customers

Testimonials and customer case studies only go in if they're real and authorized. AI can't write a "sample" testimonial that goes live, generate an image that looks like a real customer or mention the name of a client company without authorization.

This is where the mistake is most serious, because it directly misleads the reader. The precautions with generated images in ads are in AI creatives for paid media.

5. Budget and channel priorities

AI can analyze data and suggest where the budget performs best. The final decision involves context that isn't in the data: business seasonality, the sales team's capacity to handle more leads, company goals, long-term bets. It also involves responsibility for the money.

A good practice is to let AI prepare the analysis and the recommendation, and have marketing and sales decide together based on it.

6. Final approval

Every piece that goes live in the company's name has someone responsible for approving it. It can be a different person for each type of piece, but it can't be no one.

Want to implement AI in marketing with review rules that protect your brand? See Nexo's AI for marketing

What about when AI talks directly to the public?

There's a special case: AI agents that answer messages, comments and questions on behalf of the brand, without prior review of each reply. Here you can't approve piece by piece, so the limits need to be built into the agent's instructions and the knowledge base.

Three rules help:

• Closed scope. The agent only answers about what's in the knowledge base: products, timelines, hours, FAQs. Anything else goes to a person.
• No offers. The agent shares published terms, but doesn't negotiate, doesn't give discounts and doesn't promise exceptions.
• Criticism and complaints go to people. Negative public comments, threats of a formal complaint or an angry customer leave the agent and go to someone on the team, with the history.

And someone reads a sample of conversations frequently to correct the knowledge base. The agent answers on its own, but it doesn't go without an owner.

Matrix: what AI does and who decides

How to set up an approval flow that doesn't stall

The common fear is that review cancels out AI's speed gain. It doesn't have to. Three measures solve it:

Review proportional to risk. Classify pieces into two or three levels. A post variation with no offer gets a quick read. An ad with pricing, material with a number or any piece involving a customer goes through full review, with source checking.

Written rules. A short document with what can never be said, the allowed promises, the current offer and the brand's vocabulary. That document goes into the request to the AI and is the reference for review. The clearer it is, the less rework. The AI usage policy for your company helps define what can be sent to the tools.

Named owners. For each type of piece, who reviews and who approves. Without a name, review becomes everyone's responsibility, which in practice means no one's.

Approval checklist

Before publishing, the approver checks:

• Is the piece consistent with the brand's positioning and tone?
• Are the offer, price and terms the current ones?
• Does every number and every promise have a source?
• Is there a testimonial, case study, name or image of a customer? Is it authorized?
• Does the piece follow the policies of the platform where it will run?
• Does the call to action lead to a next step that sales can actually handle?

The last item is often forgotten. A campaign that generates leads for an offer sales can't handle in time hurts the sale as much as wrong copy. The review precautions for the content itself are in AI content without losing quality.

Common mistakes

• Letting AI create an offer or a term to "improve" the ad.
• Publishing an AI-generated testimonial or case study as an example.
• Not having a named owner for approval.
• Reviewing everything with the same rigor and stalling production.
• Reviewing nothing because "AI writes well."
• Forgetting to tell sales about the offer that's live.

Conclusion

AI in marketing delivers more when the limits are clear. It speeds up production, variation and analysis. Brand, offer, promises, testimonials, budget and approval stay with people, because they involve accountability and information the tool doesn't have.

With written rules, review proportional to risk and named owners, the company gains speed without putting the brand at risk. And every approved piece reaches sales with a promise the rep can keep.

Want to know whether your marketing promises what sales delivers? Take Nexo's free assessment.
Take the free assessment