How to use AI in writing – and still sound like Bayer
The use of AI is on the rise among Bayer employees. And this applies not just to our proprietary AI engines for specific purposes, but also to the general-purpose AI tools that are available publicly.
We already published an article on AI image generation – but we see just as much interest in large language models such as ChatGPT, Claude, Gemini, or Perplexity. These are styled as ‘answer engines’, meaning they can fulfill any text query you give them – but one of their most common uses is for creating written works.
Today, we have more written communication to produce than ever, from emails and chat messages to social media posts, internal documents and longer-form creative or technical works. So, it’s natural that AI-powered writing tools will be of interest. At the same time, there are ongoing concerns about the accuracy, copyright status and environmental impact of AI – and, especially when it comes to written works, a concern that AI-generated content is fake, inauthentic, or otherwise ‘not good enough’.
So, can you use AI writing tools in the course of your work as a Bayer employee?
Yes – we’re a forward-thinking company and pride ourselves in enabling colleagues to use the latest technologies where appropriate.
BUT – take care in how you use them. AI isn’t a ‘zero-effort shortcut’ to perfect content; you must still put in the time to ensure your output meets high standards for safety, accuracy and creative execution.
How can you achieve this every time? To get you started, there are some overall rules in the GenAI Guidance and Rules sheet – such as being transparent about the use of AI, respecting sensitive or personal data, and refraining from prompting with copyrighted material. Beyond this, there are more subjective matters regarding the tone, style and approach that people have come to expect from Bayer. This article covers these creative considerations, giving you advice and examples that you can hopefully apply to your own writing tasks.
Don’t replace people with AI – empower them
How much should you use AI to help with written content? Using it smartly can mean using it more and less at the same time. More because its potential goes beyond pure creative writing tasks like articles or social posts; it’s worth experimenting with it in more functional, administrative or technical communication contexts too. Its impact doesn’t have to be revolutionary; even if all it does is save you some rewriting or repetition, it can still be worth it.
You also use AI less because you don’t replace your entire workflow with AI, instead using it smartly where it will improve your results without taking away what makes your work memorable. LLMs can be used for a variety of tasks in writing workflows apart from the writing itself – such as research, giving ideas or examples of formats, modifying initial drafts you have written, or acting as a reviewer of your writing and suggesting ways to improve tone and clarity.
It is difficult to give rigid rules as each use case will be slightly different, but you could summarize the overall principle as:
The human always leads the machine, not the other way round.
We use AI to enable and accentuate human creativity, not to overrule or replace it. The finished content isn’t so much AI-generated as AI-assisted with human oversight.
Bayer is a people-oriented company, and we are driven by the human impacts of our work; our communication processes should reflect that. Furthermore, people respond better to content that they perceive as coming from another person, not from a faceless entity or computer program. That’s why carefully adjusting written output for tonality is crucial.
Our personality is visionary, optimistic and passionate, and we embody this through all our communication. But public AI engines are general-purpose, and not specifically trained on our brand voice. So, their default output may sound a little off-brand. You may also have seen discussions online of so-called ‘AI speak’, which can sound flat and generic.
This is certainly a pitfall you need to avoid. But these identical-sounding responses are not inevitable – AI tools can offer a surprising amount of variation. And when it comes to prompting, you get out what you put in: a basic, formulaic response may be the result of a similarly basic prompt.
When you prompt an LLM for a certain written task, you can include more than the deliverable itself – you can add pointers on tone, format, structure, or anything else. When aiming for the Bayer tone of voice, you could consider adding any of the following (or your own ideas) to the prompt:
Avoid typical corporate and business speak, repetition, and filler phrases that do not add meaning.
Write as if this was an interpersonal conversation, not a corporate page.
Tone: empathetic, personable, respectful, understandable.
Assume an optimistic personality while remaining realistic and avoiding unsubstantiated claims.
Language: personal and accessible. Minimize scientific language and abbreviations unless they are necessary.
Limit sentence and paragraph length.
Use the active voice.
Write in the first person where possible.
Focus on human goals, impacts and stories, rather than scientific or technical statements.
Write for this audience: (Specify your audience, their needs, emotional state, and what this content needs to deliver to them).
Include a call to action.
There is virtually no limit to the features or constraints you can add – so you can even experiment with more ambitious, challenging or unconventional conditions, such as:
Write this content as if you are a c-level executive / head of communications / technical expert / farmer.
Imagine that I am a cynic about Bayer and its products. Your job is to overcome my negativity and inspire me.
Tell me what information you would need to make this draft even more impactful.
Critique the latest draft and give me three changes that would make it resonate more with its audience.
Check the full brand tonality page for more guidance you can add.
You can add these to every prompt individually. But for consistency and time-saving (especially if you’ll be producing a high volume of content, such as if you’re a marketing team incorporating AI into your strategy), it’s worthwhile setting up an AI assistant with the tone requirements in its system prompt. That way, our tone of voice will be in everything you create, right from the start.
For examples of the difference smart prompting can make, view the dropdown below. And if you have found your own prompting tricks that reliably produce engaging writing – please share them with us!
AI or not, quality always wins
However you produce written content, what matters is the quality of your output. And our brand guidelines apply just as much to AI-assisted content as to anything else. In fact, since AI assistance can scale up our output and enable us to produce more content faster, it’s even more important to ensure approved terminology and a consistent tone. And that’s easier if you include this from the beginning, rather than trying to edit it in later on.
More content means more complexity, but it also means more touchpoints and more opportunities to build a brand identity and strengthen relationships with our customers. So let’s harness this opportunity and let’s get it right.
A final disclaimer: LLMs can get things wrong! It’s not the focus of this article, but accuracy is just as important as tonality, and it’s imperative to factually check all claims. Sometimes, what sounds good isn’t the same as what’s factually correct. This is even more important when it comes to technical subjects or highly regulated environments. AI is impressive but it is not a replacement for the human expertise we have in Bayer.
These examples show what happens when we give AI a generic prompt, which just asks the model to do a writing task with no information on the required tone; and then the same prompt but with a comprehensive list of Bayer tone-of-voice aspects added.
These are just high-level examples and do not reflect real use cases. And the more advanced prompts still don’t produce ‘perfect’ responses – a real user would continue to edit these, or even incorporate parts from both versions of the response to reach their desired result. Still, you can see how different the two versions are, and how some work fine-tuning your prompt can pay off.
| Prompt | Basic response (no tone guidance) | Advanced response (with tone guidance) |
Write a social post about the benefits of the biological crop protection product Serenade for farmers | 🛡️ Protect your crops — naturally. | Growing with confidence – naturally. |
Write a summary of the products in the Bepanthen range and where they are available | 🌿 Bepanthen Range Overview
💧 Bepanthen Sensiderm Cream
🌞 Bepanthen Derma (Adult Skincare Range)
🛍️ Where to Buy
| Bepanthen: gentle support when skin needs it most 👶 Bepanthen Nappy Care Ointment Where to buy:
|
Write an internal email asking colleagues to attend a staff barbecue | Subject: You’re Invited: Staff Barbecue
| Subject: Come along to our summer barbecue ☀️🍔 |
Bayer’s AI language model, myGenAssist, is available now for all colleagues to use.
Not only can it handle all general queries, it also has built-in assistants for a wide range of business tasks, from writing emails to generating podcast audio to building website front-ends. It supports file upload and tool integration, and has access to some Bayer-specific data.
If you have any further questions about this or any other section of Bayer Identity Net, please contact: