AI in Marketing: From Buzzword to the User Perspective

As an online marketing agency, we use AI in marketing in certain areas of our projects, but we also critically assess its use. In this article, we share our own user perspective and try to bring some depth to the subject of AI in marketing.

AI as a buzzword: Less talk, more action!

The catchword “artificial intelligence” can be found ever more frequently in agency briefings.

On the one hand, it is commendable when companies and organizations seek to modernize by using AI in marketing or critically question the extent to which AI-generated content meets journalistic, quality and legal requirements.

On the other hand, it is alarming how poorly the topic of AI is sometimes treated from a strategic point of view. In the 2010s, opinion leaders in politics, the media and business were still throwing around buzzwords such as “going digital,” “big data,” and “4.0.” Lately, since ChatGPT came around, AI seems to be the new favorite concept of sloganeers and self-promoters.

AI is more than just a symbolic political tool. AI doesn’t always make everything cheaper or faster. We’re not all going to lose our jobs. And we can also just use AI where it makes our work easier or improves quality, instead of waiting for the end of a two-decade long meeting of the Ethics Council in their ivory tower.

When is it worth using AI in marketing?

Marketing professionals in particular are susceptible to be afflicted by the shiny object syndrome triggered by AI: They find it appealing because it’s new and on everybody’s lips. 

But the use of AI in marketing is not an end unto itself. As with every other tool we use in business, there must be an economic rationale for it.

We ask you: Why would you like to use AI in marketing?

Unfortunately, this question is often not even asked, or is answered with untested assumptions instead of facts.

Possible perspectives are:

1. AI saves costs
2. AI improves quality
3. AI creates new products or services

Of course, these perspectives cannot be considered in isolation from one another: If, for example, the use of an AI tool to translate content saves costs, but the quality is so poor that users abandon their user journey or have erroneous expectations of products or services, then something is amiss. Or if the AI tool can only enhance quality or create innovation at disproportionately high costs, then it is equally unprofitable.

AI in marketing: 4 Use cases from our day-to-day agency work

Germany’s economy, once a high-tech country and land of poets and thinkers, needs more and deserves better than empty phrases and endless discussions in ethics councils between supposed experts, some of whom have never even worked with AI themselves.

That’s why we would like to showcase a few specific applications of AI in marketing from our day-to-day agency work. We look at the extent to which we can save costs, improve quality or create new products. You will gain insights into our approach, the limits of AI and the compromises we have to make.

1. Image generation with Midjourney

Midjourney is a paid AI software for image generation. It is controlled via chat, with the Midjourney bot taking text-based prompts via Discord, a widely used chat platform.

We last used Midjourney for a website regionalization project: The international website of a manufacturer of plastic additives is to be expanded to include a new region for the Indian market. This also involves the use of regionally adapted images of plastic products with Indian rather than Western-looking people.

This is what a prompt for Midjourney looks like, for instance:

The /imagine command initiates the image generation. The keywords or half sentences that follow are the content briefing for the AI. The parameter –ar 2:1 defines the aspect ratio of the image, as ar is short for aspect ratio, in this case two times as wide as it is high.

Midjourney, trained with huge sets of image data and human feedback, then generates 4 images that pick-up keywords from the prompt, each containing designs that have a relatively high probability of matching the prompt well:

However, this is where the real work begins. Put yourself in the shoes of B2B decision-makers who manufacture such plastic products and take a closer look!

All 4 images contain some obvious errors that need to be corrected:

1. The diving goggles blend in with the eyebrows, a typical AI error.
2. The snap hook seems to attach the diver to the shelf.
3. The image contains products that make no sense and shouldn’t exist.
4. The mouthpiece is not connected to the oxygen tank.

Possible solutions are:

1. Try again with a different prompt.
2. Generate variants that are based on a scene but differ in the details, e.g., by clicking on V4 for variants of scene 4.
3. Extract one of the images in high resolution and edit it manually, e.g., by clicking on U4 to upscale image 4. Then use Photoshop to add a tube connecting the mouthpiece to the oxygen tank.

To summarize, Midjourney opens up new avenues in image generation, but the process is management-intensive and requires a critical examination of the image content, as B2B target groups are generally discerning and well informed. If the target groups get the idea that you’re trying to cut costs or time in the wrong place or that you’re taking them for fools, potential customers will be lost.

Why is the implementation worth it: Image generation with Midjourney?

Depending on how you approach it, AI saves costs, increases quality or creates new products in this scenario:

1. Costs: Image generation with Midjourney is cheaper, less time-consuming and more readily available than a photo shoot with Indian models and “real” products containing the plastic additives. Since the aim here is to present examples of applications for the chemical raw materials rather than to market specific diving equipment, a degree of technical inaccuracy is tolerable within certain limits.
2. Quality: Our research shows that Indian target groups are more likely to identify with Indian models than with Western ones. The use of AI increases the quality and effectiveness of the approach. 
3. New products: As the supply of suitable stock photos is limited, generating images with Midjourney bears to an extent the weight of the regionalization project, which we would otherwise not have been able to carry out.

However, the nuances are also important for all these advantages: Regionalized, AI-generated images enhance the quality of the approach. But a professional shoot, if the funds are available, would result in even better and more accurate photos.

In addition, while AI-generated images are cheaper than a photo shoot with real Indian models and products, they are nevertheless more expensive than purchasing stock photos from professional image databases when the manpower costs for operating the AI and post-processing are taken into account.

2. ChatGPT as an editorial assistant

ChatGPT is probably the best-known AI tool, receiving the most media attention. Its language skills are better than those of many laypeople and it is developing at a rapid pace.

Nevertheless, it has its limitations, because its textual output is based on frequency analysis, not on conceptual or editorial decisions. In addition, ChatGPT occasionally hallucinates and invents facts, something that doesn’t happen with human authors if they do good journalistic work.

For instance, if you ask ChatGPT to write an SEO blog post on a specific topic, copy the AI’s answer and publish it online, you will usually be disappointed: We have seen that content from ChatGPT performs significantly worse than content from conceptually competent copywriters or SEO specialists, measured in terms of Google visibility and conversions (e.g., in the form of business inquiries or purchases).

Various case studies also point to this conclusion, e.g., the experiment by Mark Williams-Cook, who published 10,000 pages of AI-generated content on a website.

Beyond copypasta: ChatGPT in day-to-day online editorial work

In order to leverage ChatGPT’s strengths and avoid its weaknesses, our editorial team uses ChatGPT for the following purposes:

I. As an interviewer

Our author introduces himself to ChatGPT as an expert on a certain topic, whereas ChatGPT is to take on the role of an interviewer (example prompt below). ChatGPT asks questions which should help our author to structure their text logically. The actual article is then compiled from the author’s answers to these questions. In this way, our author remains in the lead and can rule out the possibility of ChatGPT hallucinating facts or derailing the article from a conceptual standpoint. This process is of interest for the content creation of case studies, for example, while for SEO texts we use a different methodology not involving AI.

II. As a translator

Compared to other software, ChatGPT is pretty good at translating text, e.g., from English to German or vice versa. This saves time in comparison to a 100% manual translation. Nevertheless, the translations have to be carefully checked and, in some cases, extensively revised: It is possible, for instance, for some sections to sound “Denglish” with an unusual or complicated sentence structure, for technical terms to be translated incorrectly or inconsistently, or for ChatGPT to pick out the wrong meaning for a “kettle”.

III. As a social media copywriter

ChatGPT can generate suitable texts for social media posts from a blog post for the purpose of promoting it. The original input makes it unlikely for the output to be too generic. Along with an appropriate description of the desired tone of voice, ChatGPT can create a coherent style for your company, brand or organization. ChatGPT can also create alternatives to existing social media posts drawing on a broad vocabulary.

IV. For data analysis

ChatGPT can open and analyze certain file types, such as .csv, .xlsx, .docx and .pdf. Among other things, we supplied ChatGPT with web analytics data in .csv format and asked for a color-coded evaluation of which webpage had the most hits overall, as well as from Google searches in German and English.

V. Using voice control

ChatGPT 4o allows for natural conversations using voice control via the smartphone app. ChatGPT responds with voice messages, can be interrupted and can even simulate emotions. The voice chat is transcribed in ChatGPT and can be viewed again at a later time. ChatGPT will soon also be able to process video input from a smartphone camera. Some initial demos have already been made, but the feature is not yet publicly available.

VI. ChatGPT search

ChatGPT can be integrated into the browser via a Chrome extension and used as a search engine. As of August 2024, ChatGPT can browse the internet for ChatGPT Plus subscribers to access up-to-date information. Before August 2024, however, ChatGPT was completely unable to comment on current events and was limited to historical data up to around September 2021.

Beyond copypasta: ChatGPT in day-to-day online editorial work (cont’d)

If it can be justified from an economic point of view, ChatGPT supports our editorial team in various assistance functions that require no strategic or conceptual considerations.

Our editorial team keeps full control of strategy and conceptual aspects, because ChatGPT can make mistakes at any time and it is difficult to subsequently improve the conceptual content of an AI-generated text. To make a comparison: If you fail to cover the topic in your homework and get a failing grade at school, you can’t change it into a top passing grade just by replacing single words.

Moreover, we always try to research facts ourselves or, if we use ChatGPT Search to a limited extent, at least check the quoted sources very carefully. We do not wish to entrust ChatGPT with tasks where it could hallucinate facts.

3. Elevenlabs: AI audio

An important use case for AI in marketing concerns audio files or the soundtrack of video files: Elevenlabs is an AI audio platform known primarily for text-to-speech applications and voice generation. Among other things, it allows you to convert a text into an audio file and choose between different AI speaker voices.

Elevenlabs also allows users to turn their own voice into an AI voice to provide a recognizable voice without having to record themselves every single voiceover track. Just as with all other AI tools, the result is not 100% accurate, but may be sufficiently similar to be confused with the genuine thing, depending on application.

The closed caption generator from Elevenlabs can also automatically generate subtitles for videos. As the AI makes mistakes, careful checking and selective editing are necessary in this case too – especially in a professional context. But this method is less time-consuming than writing new subtitles from scratch and adding time stamps to each one.

At svaerm, we use AI to create subtitles for videos with subsequent post-processing and occasionally also for text-to-speech. In the future, we also have plans for the use of AI in other audio applications.

Practical example: Ambiente Trends Trailer 2025

The trailers for the Ambiente Messe 2025 trends were created as our first use case with an AI-based text-to-speech application. The prompts and the sound correction for 3 voices in 2 languages each took about 6 days overall. The implementation of the project using AI in this way is considerably more streamlined than voice casting, selection and studio production.

However, you can also hear the limitations of AI: Professional speakers could better express the appropriate emphasis, individuality and intonation of speech.

4. Infographics made with napkin.ai

napkin.ai is an AI tool for generating infographics from text-based prompts. Here is an example of a prompt consisting of labels for the planned infographic:

napkin.ai creates several design proposals based on it, from which you can then choose the one you prefer. One of the resulting infographics is already known from this article:

The design elements can be selected and edited or exported individually. For example, the labels can be translated directly in napkin.ai:

napkin.ai is currently in beta testing and is free of charge.

The agency used it for the first time for this blog post. Since infographics and illustrations are mostly used as a backdrop for the corporate design of our customers and the latter is created manually, we still consider napkin.ai to be an experimental tool. 

The limits of AI

The use of AI in marketing is promising, but the technology is by no means all-powerful. AI tools are trained to recognize patterns and probabilities and deliver results based on them. However, there are significant limitations:

1. Strategy and conceptual content: While AI-generated texts are stylistically better and AI-generated artwork looks better than that of some amateurs, their content is often shallow or with faulty details. People with the required expertise can develop approaches that are strategically more target-oriented and more suitable from an editorial and conceptual point of view.
2. Hallucinations: Both AI and humans can make mistakes, but AI runs the risk of hallucinating facts. This alone should be reason enough to realize that it’s not a good idea to indiscriminately copy and publish AI-generated content. Determining the right amount of human revision and post-processing requires a critical view and a great deal of subject-specific know-how. A constructive and transparent error culture is also required within the organization, because ultimately people have to take responsibility for mistakes caused by AI.
3. Cost-benefit ratio: AI can generate output quickly, but a combination of multiple feedback loops and time-consuming post-processing is often required in order to use the output effectively. In addition, many resources are being invested directly in the operation and (further) development of AI. Some marketing executives who don’t actually work with AI themselves succumb to the naive belief that the benefits of AI are that it makes everything faster, better and cheaper, without being at all aware of the costs and disadvantages of AI.
4. Ethical concerns: There is a lot of media coverage about copyright, deep fakes and optimization in connection with AI. These topics are all legitimate, but there are many more important issues: Who or rather which country will win the global race for AGI (Artificial General Intelligence)? What happens if user data stored by AI start-ups or even the AI algorithms themselves fall into the wrong hands? How can we meet in a sustainable way the enormous power consumption needed for the operation of AI? Will AI devalue text as a medium to such an extent that soon nobody will be reading it anymore?

Summary

The crucial factor for the strategic use of AI in marketing is to identify suitable use cases where the strengths of AI and business considerations come together in a sensible way (saving costs, improving quality, creating new products, etc.).

Human experts should continue to be entrusted with decisions on strategy and conceptual matters. Neither a tokenistic nor an uncritical approach to AI are appropriate.

Moreover, a reality check based on practical examples shows that AI is generally used to provide an additional level of support and usually can’t yet deliver a finished end product.

Contact options for companies

At svaerm, we handle customer projects of all kinds and always decide on a case-by-case basis whether or not to use AI in marketing. Our full-service agency advises and supports you at all levels of your online marketing projects – from the initial concept stage through to attaining your strategic goals.

We are always available for any inquiries about your project. After receiving your initial written request via the contact form below, we will also be happy to advise you personally by phone, video conference or on site.