🗓️ Lecture 02: AI tools in practice
🎯 Content and goals
- You can choose the right kind of AI system for a given research task.
- You can write prompts that are specifications, not wishes.
- You can verify AI output to the standard your term paper is held to.
- You can disclose your use of AI so that it cannot be read as cheating.
Explicitly not a goal: knowing which app is currently best. The tools change faster than a semester; the reasoning does not. The list of tools I currently recommend is kept below, where it can be corrected quickly.
🖥️ Lecture Slides
🧰 Templates and resources
Three companion resources support this lecture. Each is available on the website and as a download (PDF, MS Word, and Markdown source):
- Documenting and disclosing the use of AI — a fillable template plus a worked example for declaring AI use in your term papers and coursework.
- Writing effective AI prompts — a practical guide with a quick prompt skeleton and a more advanced meta-prompt.
- Checking AI-generated citations — the in-class exercise: one paragraph, three references, one of which does not exist.
🧭 Tools I currently recommend
Products change faster than a semester, so the slides teach the four capability classes and this page names the products. Reorganised from the 2025 edition of the lecture by class.
Last reviewed: August 2026. If you are reading this much later, treat every name below as a starting point for your own search, not as an endorsement.
A — Conversational assistants
Generating, restructuring, explaining and critiquing text and ideas. Ungrounded by default.
- Claude, ChatGPT, Gemini — the general-purpose assistants. Any one of them is enough; there is little point in subscribing to several.
- DeepSeek — an open-weight alternative. Note that where the hosted service runs matters for the data-protection questions discussed in the lecture.
B — Grounded search
Retrieving real sources and citing them. Always use a proper database (Web of Science, Scopus) as the foundation and treat these as complements — see Lecture 08.
- Semantic tools (“finders”) — start from a question or topic description: Elicit, Consensus, Undermind, SciSpace.
- Citation-based tools (“connectors”) — start from one or more seed publications and follow citation networks: Connected Papers, ResearchRabbit, Inciteful.
- Perplexity — general grounded search rather than an academic tool; useful for orientation, weaker for a systematic review.
A practical order that works: database first, then a semantic tool to widen the net, then a citation-based tool once you have identified key publications.
C — Document and data analysis
You supply the material — PDFs, transcripts, datasets — and the system reads what you give it.
- The general assistants in class A now handle uploaded PDFs and datasets well, and have largely absorbed the single-purpose tools of a few years ago.
- NotebookLM — works only from sources you upload, which makes it easier to check what a claim rests on.
- MaxQDA — qualitative content analysis, with AI features added; relevant for Lecture 06 if you have access to a licence.
- Remember the confidentiality rule from the lecture: interview transcripts and unpublished firm data do not go into any of these without thinking first.
D — Agentic systems
Planning and executing multi-step work — browsing, running code, producing files. You review the trace, not only the answer.
- The agent modes of the class-A assistants are the realistic entry point.
- Useful for mechanical work with a checkable result. Not useful for anything where you could not tell a good result from a bad one.
Not AI, but do not skip it
- Zotero — reference manager. Nothing on this page removes the need for one, and it is what keeps a fabricated reference from silently entering your bibliography.
Also worth knowing
- The Virtuelles Kompetenzzentrum KI maintains a curated list of AI tools that is updated more often than this page.
📚 Mandatory Reading
No mandatory reading for this lecture.
🏆 Further Readings
✍️ Exercises and coursework
Suggested exercises (to help you remember the material; not graded):
- Work through the citation-checking exercise if you were not in class, or repeat it on a paragraph you generate yourself.
- Take a prompt you have actually used and rewrite it as a specification — context, task, format, constraints — then compare the two outputs.
- Fill in the disclosure template for a piece of work you have already handed in. It is quicker than you expect, and much quicker than reconstructing it later.
If you have questions or problems, please post them in the Moodle forum.