AI in the writing process

AI at TUD

The TUD is open to the use of (text-generating) AI systems as of May 2024. Specific guidelines can be found on the University Executive Board's website.

The term "artificial intelligence" covers a broad field of research. In scientific writing, the focus is on large language models (LLM) such as ChatGPT. In common parlance, "AI" has become established, which is why we also use this term to refer to LLMs.

The implementation of AI systems in the writing process offers numerous advantages, but also harbors the potential for unreflected dependency. In order for AI to support learning, reading and writing processes profitably, a competent and critical approach to this technology is a basic prerequisite.

The TUD Writing Center is the point of contact for all questions relating to the use of AI in academic writing. We are also happy to provide individual advice. This service is aimed at both students and teaching staff.

How generative AI can support academic reading and writing

AI tools can be used in every phase of the writing process. Although they do not completely replace human actions, they are suitable, for example ...

for inspiration and brainstorming
for the creation of outline drafts
for help in selecting methods
as an additional research option
for the development of literature networks and summaries
for summaries, translations and help with understanding texts
for data evaluation and visualization
for generating texts (and other media), e.g. from key points or as different versions of the same text section
for text feedback on structure and writing style according to predefined criteria
for correcting spelling and grammar

For more information, see the handout "Use of AI writing tools by students" from the Writing Center at Goethe University Frankfurt am Main.

AI tool tips

Which AI tools can usefully support the academic writing process? What can they do and what can't they do? How do you use them and what do you need to bear in mind with regard to terms of use, copyright and data protection?

Here is the most important information about our (current) favorites:

Large Language Models

Brief description	A pre-trained large language model (LLM) generates texts (and other media such as images or videos) after input (prompts) from users based on its training data. During generation, new, unique content is created by the models calculating the probabilities of the word sequence - they do not use wikis or databases in the training data to look up facts. As a result, the models sometimes hallucinate information and output false facts. This is not malicious or intentional, the models have no understanding of our world - their aim is to generate content and this content is factually correct with a certain probability. Depending on which model is used, different multimodal functions are available. The different models can, for example, also research their answers by accessing the Internet or generate even better texts using the newer reasoning models. A large language model can be used web-based (via a browser) or locally (‘installed’ on your own computer), and they are divided into proprietary models (such as ChatGPT) or open source (such as Llama) - this results in different functionalities and concerns in use.
Strengths	Generate texts Above all, the models can generate very good texts. They are good conversation partners with strong multilingual skills. The exchange can enable a change of perspective and the model can be used as a “sparring partner” to look at problems from different angles. It never gets tired of generating answers and is always available. Depending on the provider, the models are also particularly helpful for programming, research or other specific aspects.
Weaknesses	Hallucinations The models hallucinate information and have no understanding of our world. The output they generate is always based on the training data and they reproduce everything they have been given (e.g. false information, biases, racism and discrimination). Resources The powerful models consume a lot of resources (water and electricity), but the users (usually) have no exact insight into the consumption. Biases All models are subject to certain biases that must be taken into account when using them, e.g. social biases such as discrimination or political biases.
Data protection	The situation with regard to data protection and copyright is currently difficult - legal issues have not yet been conclusively clarified. Data protection Depending on the provider, different standards are set for data protection and data processing. Please always check how and where the application you are using processes and stores data. Most providers store and process user input, which is why personal data or research data as well as internal company data may not be passed on under any circumstances. An overview and further information on this can also be found on the page on the use of AI at the TUD. Copyright Large language models may use copyrighted content in their training data, but depending on the model, users have varying degrees of insight into the sources on the basis of which the new texts were generated. Copyright infringement cannot therefore be ruled out, and all content must be checked by the users. Under German law, large language models have no copyright on the outputs they generate, as they are not persons. There are different approaches to how the legal situation in Germany can currently be viewed and interpreted, some ideas and insights can be found here: Article and podcast (in German) by legalnerd “ChatGPT and copyright” Lecture (in german) by Jan Hansen 14.11.2024 “Can the law tame learning machines?”
Helpful for ...	Inspiration and brainstorming, creating draft outlines, summaries, translations and help with understanding texts, generating texts (and other media), text feedback on structure and writing style according to predefined criteria, correcting spelling and grammar
Access	An account is most times required, chat histories are saved, depending on the provider.
Further information	There is a growing number of providers of large language models, e.g. web-based proprietary LLMs such as ChatGPT (OpenAI), Claude (Anthropic) or Gemini (Google) web-based open source LLMs such as Mistral (Mixtral), DeepSeek or Llama (Meta) local LLMs such as GPT4all (nomic) In addition, various access platforms have developed that bundle access to different LLMs, e.g. Academiccloud.de (access with TUD email address possible) Chats of the KI Campus (KI Campus login required) You.com (pilot phase of the TUD with you.com)

ScienceOS

Brief description	Application with a focus on literature research and processing answers questions in natural language using scientific literature (AI Science Chat). Also provides support in understanding the literature (AI PDF Chat).
Strengths	Simple and clear The tool from the Dresden-based start-up company is intuitive and easy to use and the interface is clearly laid out. The tool was explicitly designed for use in science and research and helps to efficiently analyze specialist literature and data. The tool is constantly being further developed with the involvement of users. Open Access Almost all texts displayed are freely accessible and directly linked. ScienceOS obtains the scientific texts from Semantic Scholar, another AI-based research tool with a current database of over 200 million scientific sources, which is updated weekly.
Weaknesses	Limited source base The scientific sources that Semantic Scholar, and therefore ScienceOS, draws on are largely limited to texts from English-speaking countries. As almost all the texts displayed are published as open access, there are hardly any sources from licensed journals. A complete literature search is therefore not possible with this tool. Furthermore, not all scientific disciplines are equally represented.
Data protection	Neither the input nor the output is used to train the models further. The data is mainly processed on European servers.
Helpful for ...	Inspiration and idea generation, research processes, development of literature networks, summaries and help with understanding texts
Access and website	A free account is required, the chat histories are saved. The free version only works with the abstracts of the articles. Website ScienceOS
Further information	Tutorials on the ScienceOS Website (Englisch)

DeepL Write

Brief description	The AI-supported writing tool enables the optimization of texts in the areas of grammar, spelling, style and expression.
Strengths	User-friendliness The tool is extremely easy to use. Simply open it and get started. The limited selection of functions ensures a clear interface. Functions Texts entered can be edited for simple, business, academic or casual writing styles. You can also select a "tone" (e.g. enthusiastic, diplomatic) and search for synonyms for individual words. Spelling, grammar and punctuation errors are corrected automatically.
Weaknesses	Limited language selection The tool only offers work on German or English texts.
Data protection	Only the DeepL Write Pro version guarantees, among other things, "maximum data security". This means that texts are deleted immediately after processing, are never shared with third parties and are not used to train the models.
Helpful for ...	Revision, correction
Access and website	https://www.deepl.com/de/write The number of possible text optimizations is limited in the free access. The paid Pro version for 10 euros per month (as of June 2024) offers unlimited text optimizations, use of writing styles and alternative suggestions in addition to the above-mentioned guaranteed data protection.

Perplexity AI

Brief description	Perplexity AI is a research tool that, like GPT, is operated via chatbot, but bases its answers on a list of linked sources. It can also be asked about uploaded files (texts, images, diagrams).
Strengths	Source information All answers (usually given as theses) are generated on the basis of researched sources. Each answer is accompanied by a list of references with direct links to the sources. The individual theses are linked to specific sources from the list. Up-to-dateness The tool can search the Internet on a daily basis and can therefore also answer very current questions. Although this is not always necessary for academic research, it is certainly a big plus for certain questions, such as those relating to current media discussions. Search filter The tool offers the option of setting search filters, for example to search exclusively for academic sources or to generate text only without referring to sources. However, the filters can only be used if the Pro version of the tool is used. Attachment Files can also be uploaded and queried. Perplexity can also use researched sources for this purpose.
Weaknesses	For academic research, this tool is only worthwhile in the Pro version. In free access mode, you get a colorful mix of sources that are available on the Internet. In the Pro version, however, many of the non-academic results can be filtered out using the Academic search filter, so that it is possible to find useful academic sources. In the compiled sources (usually 10-15) you can find what you are looking for. However, the results are far from a comprehensive overview of research - and here too, of course, you need to check whether each type of source is suitable for your own work (e.g. if you are quoting from a popular science blog).
Data protection	Under Settings, you can determine whether or not uploaded data can be used for further training of the tool. You can exclude your own data from further use with just one click.
Helpful for ...	Research processes, development of literature networks, summaries and help in understanding texts, data evaluation
Access and website	https://www.perplexity.ai/ There is a free and a Pro access for currently 20,- USD per month (as of June 2024). In free access mode, you use the standard Perplexity AI model, while the Pro version allows you to choose between the more powerful AI models such as GPT-4o, Claude-3, Sonar Large and others. In addition, the search filters, which are not insignificant for academic research, are only available in the Pro version.
Further information	YouTube tutorial

SCISPACE

Brief description	SCISPACE is an AI tool for literature research and text comprehension. In addition to researching scientific sources, PDFs can be uploaded and questioned in a targeted manner.
Strengths (selection)	Data basis In response to a question entered, the tool lists a series of texts from (currently) around 280 million scientific sources and presents (among other things) an "insight" that relates the text to the question asked. Unlike Consensus , for example, Scispace also includes sources that are not published as open access, which expands the search results. Copilot The tool offers a range of application options for the researched texts in the so-called "Copilot". This can be asked in a targeted manner about the results or methodology of the texts and write short summaries. In addition, any question can be asked via the chatbot, allowing specific aspects of the text to be highlighted. In addition, the bibliography of the respective source is provided, via which all cited sources are directly linked. The results of the copilot can be saved in the "Notebook". My library The tool also allows you to upload PDFs from your own device. The Copilot can also be used here. This means that even uploaded texts can be sorted and questioned according to results, methodology or other levels of comparison.
Weaknesses	A search with Scispace does not replace a comprehensive database search, nor do the abstracts and insights replace the reading of some texts. As a rule, subject-specific databases are (even) better when it comes to completeness. However, the tool is very well suited for an initial literature overview on a specific topic due to the data basis, the diverse application possibilities for texts and the clarity.
Data protection	As PDFs are uploaded from your own hard disk, the issue of data protection is not insignificant here. Scispace itself therefore describes its high level of commitment to data protection: https://typeset.io/t/security-commitment/
Helpful for ...	Inspiration and idea generation, help in selecting methods, research processes, development of literature networks, summaries and help in understanding texts, data evaluation
Access and website	https://typeset.io/ In the free version, all of the above-mentioned fields of application can be used, albeit to a limited extent. In the premium version, which currently costs USD 12 per month (as at June 2014), all features can be used without limitation and insights are generated from more sources. The Copilot is also based on a better model than the free version.
Further information	Tutorial on the SCISPACE YouTube channel External YouTube tutorial

Textshine

Spelling correction	Textshine corrects spelling, grammar, punctuation and typography of German texts. Texts are not rewritten, i.e. it should not be seen as a substitute for proofreading.
Strengths	In contrast to general purpose models such as ChatGPT, Textshine is optimized for spelling and grammar correction and the associated processes.
Weaknesses	Currently only available for German texts, and only for Word files (.docx format). Textshine is not (yet) optimized for academic texts.
Data protection	The company is based in Austria. The documents are transmitted in encrypted form, processed exclusively for the purpose of order fulfillment and irrevocably deleted immediately afterwards.
Helpful for ...	Revision, correction
Access and website	Website Textshine User account required, the first 10,000 characters can be revised free of charge; 1,000 characters can be entered again and again free of charge on the homepage. Price packages in the paid version are charged according to the number of characters.
Further information	Introduction by Alexander Seifert (founder)

Offers for students

The AI workshops for students regularly take place once per semester. If you are interested in further dates and offers, please sign up for our list of interested parties on Opal.

Workshop "Talk to the Machines: Prompt Engineering" (English)
12.12.2025 | 10:00-12:00 am | Fritz-Foerster-Bau 148, registration via OPAL
This workshop takes place as part of the ESF-funded STUDY SMART project.
Input session "HOW TO ... AI Tools for Academic Writing" (English)
22.01.2026 | 9:20-10:50 am | Fritz-Foerster-Bau 148, registration via OPAL
This input session takes place as part of the ESF-funded STUDY SMART project.
Workshop "AI Tools for Academic Writing: Basics and Challenges" (English)
27.01.2026 | 15:00-17:00 pm | tba, registration via OPAL
This workshop takes place as part of the ESF-funded STUDY SMART project.
Workshop "AI Tools for Academic Writing & Studying: Practical Approach" (English)
19.02.2026 | 15:00-18:00 pm | tba, registration via OPAL
This workshop takes place as part of the ESF-funded STUDY SMART project.

Offers for lecturers

Jump-Start your Writing Didactics

In the brochure "Starthilfe Schreibdidaktik" (Jump-Start your Writing Didactics) published by the SZD, you will find a chapter on the use of AI applications in teaching and the evaluation of their use by students.

You can download the brochure "Starthilfe Schreibdidaktik" (in German) here.

Information on hot to get started

On our information website "Academic Writing with AI Tools. Information and ressources for teaching" we report on how to successfully approach the complex topic of artificial intelligence in the university context and in scientific writing.

Workshops for lecturers

You can find all current continuing education courses offered by the Writing Center, including various workshops on “AI tools and academic writing”, in OPAL-Kurs "Weiterbildungen des SZD".

If you are interested in additional dates or further training requests, please send us an e-mail: .

Further offers

Offers from other TUD service centers

ZILL: Information website on the didactic use of AI
Studium Generale: Lecture series "Work in progress": AI - Opportunities and risks for employees and their interest groups
ScaDS.AI Teaching and Training: courses and self-study materials

Offers of the SLUB

Makerspace: workshops, experiments and exchange from May to July 2024
Courses, workshops and events: SLUB event catalog

Offers from other universities and competence centers

Göttingen Society for Scientific Data Processing (GWDG): Portal with free access to various OpenSource LLMs with TUD login. academiccloud.de
KI-Campus: Learning platform with courses, videos, podcasts and blog entries that can be used after free registration. KI-Campus.org
VK:KIWA: Virtual Competence Center on AI and Research Associates with publications and an overview of AI applications at various stages of the writing process. VK:KIWA.de
Hochschulforum Digitalisierung: Dossier "Generative AI", various contributions
Writing Center of the Goethe University Frankfurt a. M.: Handout"Use of AI writing tools by students"
University of Basel : Guide "Quoting from AI"
Leipzig University Library: YouTube playlist "How to: Wissenschaftliches Arbeiten mit KI". Brief presentation of various AI tools and their potential applications in the academic writing process.

If you have any further questions, need support or advice, or would like to share examples of best practice, please do not hesitate to contact us.

Paulina Hösl steht vor einem verschwommenen hellgrünen Hintergrund. Im Bild ist die obere Körperhälfte. Sie hat offene, dunkelbraune Haare. Sie trägt einen dunklen Pullover.

Consultant for Writing Didactics

Paulina Hösl M.A.

Project sTUDy smart, expert for AI & writing

Send email

Contact Information

Writing Center of TUD

work

Visiting address:

Fritz-Foerster-Bau, Floor 5, Room 556 Mommsenstr. 6

01069 Dresden

Show map of this location.

work Tel.