People ask me this a lot, at AI Expo Europe, on LinkedIn or over coffee with startup founders: “Fine, but how do you actually use AI in marketing?” It is a fair question. Talking about AI on stage is easy. Showing what you do with it on a Monday morning is harder.
So I wrote down how I work across my own projects: AI Expo Europe, Joylo, Slide, Desirex and this blog. No numbers, and no recommendations for a particular tool, because tools change from one month to the next. Just approaches and workflows. And one rule that ties them together: AI does the heavy lifting, people make the decisions.
Research and positioning: AI reads, I judge
Every good campaign starts with the customer. We interview users and partners, and AI helps me summarise those conversations, group the recurring themes and surface the exact words people use. Same with competitor analysis: I feed it public websites, reviews and product pages and ask it to show me what each player promises and where the gap is.
But the model does not pick the final angle. AI gives me ten positioning options; I choose the one I can back up with the real product. A summary is only as good as the questions you asked, and the model has no idea what you forgot to ask.
Content: the first draft is the machine’s, the voice is mine
This is where AI helps me most. First drafts, article outlines, alternative headlines, a list of questions a reader might have. Then repurposing: one good idea becomes an article, a LinkedIn post, a newsletter and a short video script, in both Romanian and English. That used to be a full day of work. Now it is an hour plus editing.
What I don’t delegate: opinions and tone. If a piece sounds like everything else on the internet, there is no point publishing it. And I am open about it: AI helps me write, including on this blog, but the ideas, experiences and conclusions are mine. I went into more detail when I wrote about what AI taught me about productivity.
Creative testing: many variants, one identity
In advertising, AI lets me test fast: dozens of ad copy variants, images and short video drafts for a Joylo campaign or for promoting AI Expo Europe. We run them on small budgets, see what works and keep only a few.
Two limits I don’t negotiate. First, brand consistency: every variant goes past a human who knows how the brand looks and sounds. Second, no false promises. If the AI writes “guaranteed delivery in 10 minutes” and we cannot guarantee that, the copy goes in the bin. A model optimises for clicks, not for trust.
SEO and visibility in AI answers
More and more people ask their questions straight to an AI assistant rather than a search engine. I covered this in my piece on SEO in the age of AI-generated answers. In short, here is what I do: clearly structured content with direct questions and answers; verifiable information; pages that say who the author is and why they know what they are talking about. You want to be cited, not just found. And you cannot fake it: real expertise shows.
Email and CRM: personalisation, not spam
I have a head start here. During my Retargeting years we built exactly this: marketing automation and personalisation for online stores. Back then we wrote the rules by hand. I have long argued that email marketing isn’t dead, and now AI makes segmentation sharper and helps with personalised messages and send timing.
The principle has not changed, though: a relevant message is a service, an irrelevant one is noise. More automation does not mean more emails. It means fewer, better ones.
Support, community and analytics
In apps like Slide or Desirex, AI helps triage support messages, answer frequent questions and spot recurring problems quickly. For delicate situations, such as a user’s safety, a human takes over. I described a workflow like this in my AI automation case study.
For analytics, I like to “talk” to the data: why did a conversion drop, which segment responds better. But I always check the numbers at the source. AI can invent a percentage that does not exist, with total confidence.
What I never hand over to AI
- Final decisions: budget, positioning, what we launch and what we stop.
- Relationships with partners, sponsors and investors. There, the person matters, not the perfect text.
- Sensitive conversations, with unhappy customers or people going through a hard time.
- Fake reviews, invented personas, fictional accounts “praising” the product. Never. It is dishonest and, increasingly, illegal.
Ethics and the law
I label AI-generated content where it is required or where readers could be misled. Customer data is handled under GDPR (the EU data protection regulation): I don’t put personal data into tools without knowing where it ends up and what happens to it. And the EU AI Act brings transparency duties, for example when a user is talking to a chatbot or sees realistic synthetic content, phased in gradually, most of them from August 2026. Better to get ready now than in the last week.
A simple weekly routine
- Monday: an hour of research, summarising customer feedback and what competitors did.
- Tuesday and Wednesday: one content idea, turned into several formats and languages, edited by me.
- Thursday: new ad variants, tested on a small budget.
- Friday: questions asked of the data, numbers checked, one decision made by a human.
My advice for small businesses: don’t start with ten tools. Start with one process that eats your time and make it better with AI. Then the next one. As I said in my predictions for 2026, the edge does not go to whoever uses the most AI, but to whoever uses it wisely.
How do you use AI in your marketing? And where have you drawn the line, what would you never hand over to it?