Three years ago, whether you wrote a text yourself or had it written was a question of time. Today it is not. A blog article takes twenty minutes, a product image thirty seconds, a landing page one evening. Anyone who can operate an AI tool can produce marketing.
We see it in the agency every day. Our clients do more themselves than they used to, and they do it faster. That is a good development. People who know their own business often write about their industry more accurately than any outsider, and AI removes the hurdle that used to stop them: the blank page.
What we want to look at in this article is a different challenge, one that is rarely discussed: production has become easier, coordination has not.
What AI tools have really changed in marketing
It is not quality. Good AI results still need someone who knows what good looks like. What has changed is the barrier to entry, and it has changed in every discipline at once:
- Text. Blog articles, newsletters, product descriptions, job ads. The first draft no longer takes an act of will.
- Images. Mood images, illustrations, variants of a motif. No shoot, no stock photo subscription.
- Translations. A second language version is no longer a budget question but a five-minute decision.
- Campaigns. Google Ads suggests ad copy, keywords and budgets by itself, and on request it applies some of them automatically.
- Websites. AI site builders generate whole pages from a description.
For a medium-sized company this is a real gain in capability. What used to be “we will have to commission that” is now often “I will try that tonight”. And that is how it should be.
Where artificial intelligence actually helps in marketing today
When artificial intelligence in marketing comes up, most people think of text first. That is the most visible application, but not the most valuable one. In our experience these five fields deliver the most for mid-sized companies, and for each of them there are examples from our own daily work.
Content creation: from the blank page to a draft
The most obvious use of AI. Tools like ChatGPT or Copilot deliver a draft for a blog article, a newsletter or a job ad within minutes. The gain is not the finished text, it is the removal of the starting hurdle. “I really should write something” turns into a draft you can work on.
The same now applies to video and audio: subtitles, transcripts, editing suggestions and voice-overs are produced automatically. An interview that used to take two days of post-production is finished in a morning.
Data analysis: reading customer behaviour instead of guessing
This is where the underrated leverage sits. AI applications find patterns in your own numbers: which audiences actually buy, where visitors drop out, which products are ordered together. For e-commerce companies this is often the point where using AI pays off first.
Automating recurring processes
Following up on quotes, confirming appointments, collecting reviews, scheduling social media posts. These are not glamorous tasks, but they eat time. If you write down your marketing processes once and then check which steps a tool can take over, you usually gain several hours a week without a single customer noticing.
Chatbots and customer service
A chatbot built on your own content answers the same recurring questions about opening hours, delivery times and availability. What matters is the handover: as soon as a question becomes individual, the route to a human has to be obvious. A bot that sends customers in circles costs loyalty instead of building it.
Personalisation in campaigns
Advertising based on observed behaviour works better than advertising aimed at everyone. The technology for it is built into every ad platform today. The limiting factor is rarely the tool, it is the data you have and the question of what you are allowed to do with it under data protection law.
In concrete terms, personalisation with artificial intelligence means this: data analysis turns into customer experiences that follow the individual customer instead of the average. The start page shows different products, the newsletter a different topic, the offer a different moment.
It is particularly effective in customer retention. Personalised loyalty programmes that reward what the individual customer actually buys keep customers measurably longer than a discount for everyone. In programmatic advertising, automation applies the same principle to ad budgets: they are placed where the probability of a deal is highest.
Evaluating customer feedback systematically
One of the least conspicuous and most useful AI applications: sorting reviews, support requests and survey answers automatically by topic and sentiment. Instead of two hundred individual messages you see five patterns.
The value lies in reaction time. If customer feedback shows that complaints about a particular delivery time are piling up, you can solve the problem before it becomes visible in reviews. That is proactive customer service, and it costs almost no effort.
The five places where it goes wrong in practice
Not because anyone is doing something wrong, but because AI tools carry things very far very fast.
The text that sounds like nobody
AI texts are linguistically clean and usually correct in substance. They just sound like nobody. If three pages of a website come out of three different prompts, the website no longer has one voice, it has three. For a trade business whose strength is closeness to its customers, that costs more than it brings in.
Incidentally, Google does not judge content by how it was created but by whether it is useful to people. An AI text with substance ranks. Twenty AI texts without any experience of your own are volume without use, and there are grades below that.
Facts that sound good and are wrong
This is the point where it can get genuinely expensive. AI models invent details when they are missing, and they do it in a convincing tone. An invented certification, a wrong delivery period, a price with no basis: on a website those are not writing mistakes, they are statements your company is liable for.
The rule we recommend for this is unspectacular: every number, every name and every commitment in an AI draft is checked against the source by a human being. AI may deliver the language, not the facts.
Images with unclear rights
Here the situation is murky, and it is moving. The terms of use of the image generators differ considerably, some allow commercial use without restriction, others do not. On top of that, a purely machine-generated image probably enjoys no copyright protection in Germany, because the personal intellectual creation is missing. In practice that means your competitor may use the same image.
We are not lawyers and this is not legal advice. What we can say from practice: for mood images the risk is manageable, for a logo, a campaign key visual or anything that is meant to carry your brand, a specialist lawyer is worth the money.
Google Ads that optimises itself
Google Ads makes suggestions, and with one setting it can apply them automatically. That sounds convenient and it is a problem in a managed account: an automatically broadened keyword match or an automatically raised budget changes the campaign overnight. The numbers in the next report can then no longer be explained, and a running test is worthless.
The same applies to the heavily automated campaign types. They work, but they are a strategic decision, not a setting you switch on in passing.
The website that looks different overnight
A rather frustrating scenario from our everyday work: we spend two weeks on the loading time of a page while the client uploads five AI-generated images at full resolution. The measurement afterwards says: worse than before. Nobody made a mistake. One piece of information was missing.
With AI tools this happens more often than it used to, because more people in the company can now contribute something. The colleague from sales now writes a blog post as well. That is good. It just needs an agreement that was not needed before.
Shadow AI: the problem nobody reports
There is one point companies almost always underestimate. Your employees have been using artificial intelligence for a long time, including without approval. The technical term for it is shadow AI: the use of AI systems in everyday work that management knows nothing about.
This is not a trust problem, it is an information problem. When there are no rules, everyone decides by gut feeling, and then a customer list ends up in a free online tool because it was convenient at that moment. For data protection and IT security this is the most critical point in the whole subject.
Bans are unfortunately not always the answer here, because they only move the usage to a private phone. The answer is a short, honest set of rules: these tools are approved, this data does not go into them, if you have questions talk to this person (your data protection officer, for example). Good employees stick to rules they understand. And since you have good people in your company, that settles the problem. 🙂
What artificial intelligence cannot do in marketing
Honesty includes the other side. There are tasks where AI systems still do not deliver, and it is worth knowing them before expectations are built up.
AI does not know your market. It knows the internet about your market. What you learn in a conversation with a customer is written down nowhere, and that is exactly the part that separates your content from your competitor’s.
AI does not decide. A model can formulate three positionings, but it cannot know which one fits your business, your people and your next five years. That decision stays yours, and it is the most valuable one in all of marketing.
And AI takes no responsibility. If a statement is wrong, the company that published it is liable. No provider will stand in for your content.
AI does not replace collaboration, it shifts it
The honest answer to the question of what an agency still contributes when AI writes the texts: less production, more judgement.
What has shifted is the effort of producing. What has stayed: the decision about what should be written about in the first place. The check of whether a draft fits the brand and the facts. The technology behind the page. The question of whether a campaign is pursuing the right goal. None of that is done by a prompt.
That is why we see not less collaboration with our clients but a different kind: they deliver more raw material and more expertise, we take care more of strategy, quality assurance and technology. That is the better deal for both sides.
An AI policy for companies: the seven points that are really needed
As soon as several people in the company use AI tools, you need a handful of rules. Not as a list of bans, but so that nobody has to guess. A good AI policy fits on two pages and answers seven questions. If you look for a template online, check it against these points:
- Which tools are approved? A short positive list. Everything else only after asking. That prevents company data from ending up in a service nobody has checked.
- Which data may go in? The most important rule of all: no personal customer data, no quotes, no contracts, no credentials in a public AI tool. For the data protection assessment of a given service, get professional support.
- What may be created with AI and what may not? First draft, summary, translation, image idea: unproblematic. Legal texts, figures for accounting, statements about certifications and guarantees: not without professional review.
- Who checks before publication? A name, not “the team”. Every AI draft that goes out has exactly one person who approves it.
- How are facts verified? Numbers, names and commitments against the source. Anyone without a source deletes the statement.
- What has to be disclosed? The transparency obligations of the EU AI Act are still developing, and different requirements apply depending on the use. Have your specific case clarified legally, and decide internally how you want to handle it instead of leaving it open.
- Who is the contact person? One person where questions about AI topics come together. Otherwise everyone decides for themselves.
One sentence belongs in every policy: responsibility for a result stays with the person who publishes it. A tool is not an excuse.
For each of these points you will find a template AI policy to download online. As a starting point that is usable, as a result it is not: a template does not know your tools, your data situation or your processes. Take the template as an outline and fill it with your own answers.
A good source for this are the chambers of industry and commerce: many of them provide a free template AI policy for companies, which works considerably better as an outline than most vendor downloads, because it is not trying to sell anyone a tool.
A note on the legal situation, because the question always comes up. The European AI regulation, better known as the EU AI Act, follows a risk-based approach and enters into force in stages. Since 2 February 2025 the bans on certain AI practices apply, as does the obligation to ensure sufficient AI literacy among employees. From 2 August 2026 the transparency obligations apply that are particularly relevant for marketing: labelling wherever people interact with an AI system or see AI-generated content.
Independently of that, the General Data Protection Regulation remains fully applicable. An AI use that is not permissible under data protection law does not become permissible through the AI regulation.
We are not lawyers, and this overview does not replace a review of your specific case. What belongs in the policy is therefore less a date than a name: who in your company watches the development and updates the rules when something changes.
How to build up your use of AI step by step
The most common mistake is starting with the tool. This order works better:
- Task first. Write down three tasks that regularly cost time and give nobody any joy. Those are your candidates.
- Two tools, two weeks. Test them on a real task, not on an example. After two weeks you know more than after ten comparison articles.
- One person who learns it. Someone in the company tries things out systematically and passes on what works. Learning in passing does not work with this subject.
- Rules before many people use it. The policy comes before the tenth colleague starts, not after.
One final point that makes the difference in practice: measure the effect. Efficiency that nobody quantifies is a feeling. Two numbers per application are enough, for example hours saved and the quality of the result on a scale of one to five.
And what about working with your agency?
There you also need clarity about who may work on which project and who has to be informed when something changes. We work with a responsibility matrix: a table in which every project has exactly one person responsible, covering the website, the Google Ads account, the social media profiles, plus who may make changes and who is informed. On top of that, three steps with every change: announce, document, report back.
With AI tools this matrix has become more important, because more people can contribute in more places. By now it has a row of its own with us: who may publish AI content, and who checks it beforehand?
Frequently asked questions about AI in marketing
Does AI-generated content harm my Google ranking?
Not automatically. Google judges whether content is helpful to people, not how it came about. What gets penalised is mass-produced content with no contribution of its own. An AI draft that an expert reworks, enriches with their own experience and checks for facts is good content.
Do we still need an agency at all if we have AI?
For the pure production of standard texts, less and less. For strategy, quality assurance, technology and everything that has consequences, more than before. The sensible division of labour is usually this: you deliver expertise and raw material, the agency makes sure something correct and effective comes out of it.
Which AI tools do you recommend for mid-sized companies?
That depends on the use case, and the market changes too fast for a list that still holds tomorrow. More important than the tool is the order: first decide which task you want to solve, then try two tools for two weeks, then decide. Whoever starts with the tool collects subscriptions.
How do we get started with an AI policy?
Take the seven questions above, sit down for an hour with two or three people and write two sentences on each question. The result is better than any template from the internet, because it fits your business. You can refine it later.
What is shadow AI and how dangerous is it?
Shadow AI means the use of AI tools by employees without approval and without management knowing. It becomes dangerous where confidential data reaches an unchecked service. A short positive list of approved tools solves the problem in most companies.
May we put customer data into an AI tool?
Without a verified legal basis and a suitable agreement with the provider: no. That is the rule that keeps you right when in doubt. Whether and under which conditions it is possible after all is something your data protection officer clarifies for the specific service, not a blog article.
Conclusion
AI tools have democratised the production of marketing. That is a good thing, and we expressly encourage our clients to do more themselves. What has not grown along with it are the agreements.
The reflex to talk about tools first when new technology arrives is understandable. More useful is the question of who in the company makes which decision. The tools change every year, the responsibilities should not.
Two things solve most of it: a short AI policy inside your own company and a clear responsibility matrix with everyone working on the same project. Both fit on a few pages and cost an afternoon.
If you like, we can go through it together once. One hour, your projects, your tools, and afterwards everyone knows where they stand.
