How do I introduce AI in my communications department without a big IT project?
Three use cases that start without a project budget, the Article 50 labelling duty in daily work, and why the approval rule matters more than the tool. Including a four-week starting order.
Start with the tasks whose output already passes through an approval step: research summaries, first drafts for recurring formats, and translations. These three need no interface, no project budget and no IT department. What they do need is a written rule on which data may go in and when a result has to be labelled.
What you’ll take away:
- Communications is the department where the Article 50 labelling duty bites first, because its output goes to the outside world.
- Three use cases start without integration. Anything requiring an interface belongs in a later stage.
- The approval rule decides the outcome, not the choice of tool.
- A first draft with human final editing is the workable cut. Fully automated publishing is not.
- Four weeks are enough to start, provided tool approval is settled on day one.
How do I introduce AI in my communications department without a big IT project?
Pick two or three tasks from live operations whose output is editorially approved anyway, and write a rule for those tasks. Use the AI features already licensed in your existing Microsoft 365 or Adobe environment. A project only starts when systems are to be connected or content published automatically.
The reason for this cut is practical. Communications work already has a checking step built in, because nothing goes out without a second read. That checking step is exactly the safeguard other departments have to build first. This makes communications one of the few areas where AI can start without new control machinery.
What destroys that advantage: when the draft looks so good that nobody reads it again. That is the one point where introduction fails in practice, and it is a matter of agreement, not technology.
Which three use cases start without a project?
Research summaries, first drafts of recurring formats, and translation. All three run inside the tool window, without a data connection, and produce something on day one that a person can check.
Research summaries save time most reliably and carry the largest error risk at the same time, because invented sources look plausible. The rule for this is short: every figure, every quotation and every source is verified against the primary source before use.
Recurring formats are the least spectacular and economically the best case. Job adverts, trade fair write-ups, product announcements and social posts follow a pattern that can be described. Translation, in turn, pays off where nothing was translated before for lack of time.
What applies under the Article 50 labelling duty?
The transparency obligations under Article 50 of the AI Regulation have been applicable and subject to penalties since 2 August 2026. For systems already running before that date, the transition period ends on 2 December 2026. Affected are mainly synthetic images, audio and video content, and chatbots in customer contact.
For day-to-day text work this means less than is usually feared. A text edited and owned by a person is a different thing from machine-generated content published unchanged. Where the line sits in an individual case depends on the specific format, which is why the department needs a written decision rather than a gut feeling.
For images the case is clear. An AI-generated or substantially altered image that goes out should be labelled. The detailed treatment is in our article on Article 50 in practice.
Who decides this when there is no IT department?
Management decides which tools are approved and which data may go in. The head of communications decides which formats use AI and how approval works. Both have to be written down, otherwise every person effectively decides it at their own desk.
That allocation is the real core of the question. Without an IT department it is not technical competence that is missing, because the tools are usable. What is missing is the place that checks default settings and notices changes. How this is organised without in-house IT is covered in our article on responsibility for AI without an IT department.
For the communications department, one page with four points is enough to begin with: approved tools, data boundary, labelling rule, and who to ask in case of doubt.
Which data must not go in?
Personal data from applications and customer records, unpublished financial figures, contract content, and anything covered by a confidentiality agreement. This boundary applies regardless of the tool in use, and it is the point where a policy shows effect fastest.
The practical everyday test is simple enough for anyone to apply: would I send this text to an external service provider without a contract? If not, it does not belong in a tool whose processing has not been clarified.
For image material the rights question is added. Prompts for image generation must not use protected third-party works, and where people are depicted, personality rights remain untouched.
What does a four-week start look like?
| Week | What happens | Who |
|---|---|---|
| 1 | Put tool approval and data boundary in writing | Management |
| 2 | Select two formats, define the approval path | Head of communications |
| 3 | Train the team, discuss the labelling rule | Everyone in the department |
| 4 | First formats live, review after two weeks | Head of communications |
Week 1 is the only one that can genuinely block. If tool approval is not decided, the team carries on with private accounts, and then company material sits in accounts nobody has access to.
After eight weeks a sober interim review pays off: which formats really run with AI, how much time was saved, and how often final editing had to correct substance. The third figure is the most interesting, because it shows whether the cut is right.
Frequently asked questions (FAQ)
Do we need Microsoft 365 Copilot for this? No. For the three use cases named, an approved language model in the browser is enough. Integration into the Office environment only pays off once content regularly has to be built from internal documents.
Does everyone in the department need training? Yes, and the scope is manageable. Two to three hours for the shared basics plus a short part on the labelling duty, because that risk is concrete in communications.
What about the brand if texts are machine-generated? Tone of voice remains the job of final editing. In practice a short style sheet with example sentences helps more than any tool setting.
May we use AI images on social media? Technically yes, with labelling and cleared rights to the source material. Check first whether it fits the brand. For references and customer situations, real photography is the better choice.
How do we stop people from skipping the second read? By giving approval a named person rather than a department. A responsibility without a name gets skipped in daily business.
Further reading
- How does an SME use AI in accounting without its own IT department? for the same cut in a department with a very different risk profile.
- Who is responsible for AI when there is no IT department? for the responsibility question beyond one department.
- What content does an AI training for the workforce need? for the training part in week 3.
Want tool approval and the labelling rule settled in one meeting? Book a conversation.
Sources: AI Regulation Art. 50, applicable since 2 August 2026, transition period for existing systems until 2 December 2026 · Regulation (EU) 2026/1744 (Digital Omnibus Regulation on AI), in force since 27 July 2026. Practical guidance, not legal advice. As of 18 August 2026.