Client
Job
Link
The AI editorial assistant for better texts and optimized content workflows
MISSION
TASK
The idea was to develop a digital product that could automate content production processes in publishing houses, media companies and editorial offices thanks to the use of large language models. When creating texts, this product should preserve the style, tonality and quality of the sender through good prompt engineering and be tailored to the target group and target channels.
Challenge
Content production in large publishing and media houses, for example, is faced with the challenge of constantly producing high-quality texts that meet the specific requirements and style of the respective publication. The texts not only have to be produced quickly, but also in large quantities and distributed to the various target groups. Traditionally, this requires a lot of time and resources, as each text has to be created, checked and optimized individually. In addition, the process is often inconsistent, as different editors apply different styles and quality standards.
GOAL
We set ourselves the goal of developing a tool that could standardize and automate this process while maintaining high quality standards. The aim was to create a product that uses the latest technological innovations and gives prompt engineers the opportunity to create prompts and prompt chains to make them available internally to their own editors – always in the style and tone of the company, always tailored to the target group and channels.
APPROACH
UX AND UI-DESIGN
By means of user research, wireframing, design processes, prototyping, the development of user stories and user journeys, the implementation of UX tests and surveys of beta users, a tool was developed that can already be used in large editorial offices.
Together in a large team from the areas of UX design, UI design, backend, frontend, QA and specialists from the fields of computational linguistics, natural language processing and machine learning, a product was developed that includes the following functions and features:
CHOICE OF THE LLM
Various large language models, such as those from OpenAI, are used to completely automate text creation and rewriting for the various target channels of texts.
INTUITIVE USER INTERFACE
A user-friendly interface design ensures that the product is easy and intuitive to use. Regular surveys and interviews with users help to adapt the product to changing needs. The original text can be compared with the generated output at any time.
CUSTOMIZED PROMPTS
Prompt engineers can quickly and efficiently create customized, company-specific prompts and prompt chains, categorize them, version them and manage them in their own library. This means that the prompts and prompt chains are always available to editors for their own work.
API
Rewrite can be integrated into content workflows and editorial systems via API. Repetitive creation processes can be automated and the style and quality of texts can be standardized.
SEO AND SOCIAL MEDIA OPTIMIZATION
The generated texts can be optimized for search engines and social media.
LANGUAGES
Regardless of the language in which the source text is available, it can be translated into all common languages. Cultural nuances are taken into account.
CITATION PRESERVATION
Original quotations can be retained on request. If the source text contains original quotations or passages that need to be copied one-to-one into the final text, these can be retained even after rewriting.
TEXT LENGTHS ACCORDING TO REQUEST
Exact text lengths can be defined, allowing source texts to be shortened or stretched as required.
CHAT-HISTORY
The individual chats are saved automatically and can be called up again at any time.
RIGHTS AND ROLE MANAGEMENT
Role management enables the targeted administration of prompt engineers, editors and admins and assigns them specific rights.
CONCLUSION
SUMMARY
With the new digital product AI Content Assistant rewrite from retresco, texts can be rewritten, adapted and output in the style, quality and tonality of the sender in a completely automated process thanks to the use of large language models and prompt engineering.