AI in marketing: what to delegate and what needs a specialist
What Mtools already delegates to AI, where it gets things wrong, how the human double check works, the risks with client data and where your team can start.
AI in marketing pays off most on high-volume tasks and first drafts: meeting summaries, task organization, data compilation and draft copy. Strategy, published numbers and brand tone stay with specialists, and nothing the AI produces goes out without a human check. At Mtools, that check has a name: the double check.
On Brazil's Innovation Day, October 19 (created by Law No. 12,193 of 2010), we're opening up our process: what we already delegate, where AI gets it wrong, how we review and where your team can start.
What we already delegate to AI
AI comes in where the work is repetitive and easy to check. In our day-to-day, four tasks already go through it:
| Task | Tool | What we check |
|---|---|---|
| Client meeting summaries | Gemini, in Google Meet | The decisions and what only the client said |
| Loose documents turned into ClickUp tasks | Claude | Owners, deadlines and priorities |
| Data compiled for reports | Claude | Every number against the original source |
| First drafts and grammar review | Claude | Brand tone and the final copy, always |
The meeting summary is a good example of the gain. With it and the notes from the people in the room, a few minutes of conversation turn into ideas for next month's content plan. The AI organizes the material, and the plan is still written by people who know the client.
Where artificial intelligence in marketing gets it wrong
AI makes mistakes with confidence, and the mistake looks right at first read. Three failures show up often.
Made-up data
When information is missing, the model fills the gap with something plausible: a percentage, a date, the name of a study. This is what's called a hallucination. In a report or a post, a number with no source becomes a public problem for the client.
Generic copy
Without context, AI writes like the internet average: correct sentences, with nothing only that brand would say. The text passes the grammar check and fails on tone of voice.
Generic landing pages
Any AI can build a landing page in a few minutes. What it can't do on its own is understand who will visit the page, what questions they have and how to organize the information toward conversion. The result is usually a good-looking, interchangeable structure, with a landing page form asking for things sales never uses.
How the double check works
Nothing AI produces goes to the client without passing through a person. The check covers four points:
- The reasoning. Whoever asked checks that the conclusion makes sense, because AI hallucinates now and then.
- Numbers and sources, one by one, against the original document.
- What only exists in the conversation with the client. Context, restrictions and decisions that aren't in any file come in through the people who were there.
- The final copy, adjusted to sound like the brand.
The reason is simple: the name on the publication is the client's. One sentence misread, in the wrong place, becomes an image crisis for them. AI speeds up the operation, and the responsibility stays with us. It's what we sum up on our who we are page: in the age of AI, we are the specialists.
Talk to usWant help planning this at your company?Tell us what you need. We'll reply by email or WhatsApp and, if it makes sense, put together a quote.Contact us
The risks with client data
Before pasting any data into an AI tool, ask whether it can leave your company. Lead lists, customer bases and conversations with names and phone numbers are personal data, and Brazil's data protection law, the LGPD, requires processing to have a legitimate, specific purpose that the person has been informed of. Some practical precautions:
- use company accounts, with the tool's data rules known to the team;
- anonymize spreadsheets before asking for analysis: replace names and emails with codes;
- treat the client's strategy, prices and numbers as confidential, as you would with any vendor;
- record which tools the team may use and for what.
How to start using AI in marketing on your team
Start with a task that is repetitive, easy to check and free of sensitive data. The steps:
- List the tasks that repeat every week. Meeting summaries, task organization and copy review usually come up first.
- Pick one and write the instruction with context: who the client is, the goal, the expected format and what to avoid.
- Define the double check before scaling. Who reviews, what they check and where errors get logged.
- Document what worked. A good instruction becomes a template for the whole team.
It's worth looking at the other side too: AI already sends visits to your website, and you can measure ChatGPT traffic in GA4.
AI and agencies: what to ask before hiring
When hiring an agency, ask how it uses AI in marketing for its clients. Every agency uses some tool today, and the difference is in the process around it. Three questions reveal a lot:
- which deliverables go through AI, and at what stage;
- who reviews, and what exactly is checked before sending;
- which tools receive your company's data, and under what rules.
An agency that answers clearly has a process. These questions fit into the checklist on how to choose a marketing agency.
Frequently asked questions
Will AI replace the marketing team?
It takes over tasks. Strategy, the client relationship, deciding what gets published and responsibility for results stay with people.
Which AI tools for marketing should I use?
It depends on the task. We use Gemini to summarize meetings and Claude to organize tasks, compile data and write first drafts. The instruction and the review matter more than the tool.
How do I stop AI from making up data?
Always ask for the source and check each number in the original document before publishing. If the source doesn't show up, the number doesn't go in.
