AI in marketing: what it can really do and where it falls shortAI in marketing

The most suitable tasks have clear inputs, verifiable results and limited potential for harm if a draft is wrong.

First, research support. AI can organise subjects, collect search terms or summarise long documents. Every factual claim and source must then be checked against the original.

Second, first drafts. A rough draft for a newsletter, landing page or blog article can make it easier to get started. Expertise, brand voice and approval still come from people.

Third, variants. AI can suggest several wording options for Google Ads or social adverts. Claims, policies, the landing page and labelling are checked before testing; only a properly evaluated test will show the effect.

Fourth, analysis support. AI can help structure approved, adequately protected data. Results remain hypotheses until a person has checked the method, data quality and context.

Strategy, subject-matter claims and legally sensitive communications require particular oversight. A model does not know your full business context or automatically have the current legal position and source material.

Outputs can sound plausible while still being wrong, outdated or invented. Check figures, quotations, links, product features and performance claims against the original sources. Medical, legal, financial and other high-risk subjects also require appropriate specialist expertise in the approval process.

Do not enter confidential information or personal data into an AI tool without review. Use such data only once the purpose, legal basis, data minimisation, access rights, contracts and approved tool have been clarified. If in doubt, anonymise the data or do not enter it at all.

Human review involves more than spelling: a named person checks the content, sources, rights, data protection, brand voice, potential discrimination and the meaning conveyed by the final presentation. Only then is it published.

GEO, short for Generative Engine Optimization, describes presenting content clearly for conventional and AI-assisted search. Systems such as ChatGPT, Perplexity and Google's AI features decide for themselves whether and how to use or display sources.

A clear structure, first-hand expertise, specific answers, reliable sources and sound technical foundations are useful here. Much of this is already part of good SEO and the fundamentals in our SEO for beginners guide.

FAQ sections, structured data and consistent company information can provide context. They guarantee no mention, citation, recommendation, ranking or particular number of visits. Such outcomes must be monitored and must not be sold as assured.

Since 2 August 2026, the transparency obligations under Article 50 of the EU AI Act have applied. Depending on their role, providers must make certain AI outputs technically detectable, while deployers must disclose deepfakes and certain AI-generated text concerning matters of public interest, among other things; exceptions and the specific form of disclosure depend on the use. Before publishing realistic synthetic or manipulated image, audio or video content, you must therefore determine which labelling obligation applies.

In practical marketing terms, document the tool used, sources, rights, editing and human approval. Ensure the perceptible disclosure required in the specific case for realistic synthetic or manipulated media and deepfakes. Do not present an AI-generated character as a supposedly real customer, employee or case study.

The EU Data Act is not a general seal of approval for a marketing website. It primarily concerns access to and use of data from connected products and related services, as well as certain data and cloud scenarios. Its relevance to a particular offer must be assessed separately.

This overview is not a substitute for legal advice. Keep responsibilities, approvals and AI literacy within the team up to date, and have high-risk uses reviewed by the appropriate subject-matter and legal specialists.

You need neither an AI team nor a four-figure tool subscription. You need a practical starting point. Here is how to proceed:

Step one: choose a single use case, such as first drafts for your newsletter or variants for your adverts. Do not tackle five fronts at once.

Step two: define your audience, offer, tone of voice and non-negotiables. Use only information permitted for the approved tool and purpose; exclude confidential information and personal data until the requirements have been clarified.

Step three: establish source verification, human review, rights clearance and a responsible approver. The process for realistic synthetic or manipulated media must also include visible labelling.

Step four: measure the outcome without promising success in advance. Our conversion rate optimisation guide shows you how to design such tests properly.

The honest final question is this: if AI can do so much, why do you still need an agency? The answer lies in everything above.

AI can assist with production steps. Strategy, positioning, brand voice, local market knowledge and responsibility remain tasks for qualified people.

This is how we work: AI assists with selected research, variant development and analysis tasks. People verify sources, edit the copy and decide whether to publish. We judge whether this improves the process by examining the specific result.

If you want to know where AI could save time and money in your marketing and where experience remains the better choice, talk to us. In a free strategy session, we assess your situation and give you an honest view of what may be worthwhile—even when the answer is fewer tools and more focus.