Acknowledgments

Why we publish this note

This website is about disclosing AI use in scholarly work. It would be odd of us to ask researchers to be transparent about their AI assistance and stay quiet about ours. This page records how we use AI tools when we write and edit content for ai-cards.org.

What AI helps us with

Large language models support several parts of our content workflow. The most common uses are listed below.

• Drafting. We use models such as Claude (Anthropic) and GPT (OpenAI) to turn outlines and source notes into first drafts. A draft is a starting point, never the final text.
• Editing and rewriting. We ask models to tighten sentences, smooth transitions, fix grammar, and check tone. Human authors approve every change.
• Structure and outlines. We sometimes brainstorm section structure and headings with a model before writing.
• Research support. Models occasionally help us summarise a long policy document or surface candidate references. We then read the primary source ourselves and verify any claim before it enters a published page.
• Code and tooling. Parts of the site itself were built with AI coding assistance. The repository is maintained by human engineers who review every change.

What AI does not do

We do not publish unreviewed model output. We do not let a model decide what a journal policy says, what a paper claims, or what a tool is designed to do. Every factual statement, citation, link, and recommendation on this site is checked by a human author before publication.

We do not list AI systems as authors. The named people behind this project carry full responsibility for what appears here, including any errors.

How we review AI output

Our review process for AI assisted text follows the same logic we recommend in our own AI Usage Cards.

• We read every sentence and rewrite anything that misrepresents a source or sounds wrong.
• We verify quotes, statistics, dates, and named policies against primary sources.
• We check that links go to the resources we intend and that external claims are still accurate.
• We remove repetitive phrasing, vague language, and tells of AI authored prose.

Confidentiality

We do not send confidential research data, unpublished manuscripts submitted to us, or private correspondence through third party AI services. Tools that process user content stay on devices controlled by the authors.

Tools we currently use

• Claude by Anthropic
• GPT by OpenAI
• Gemini by Google
• GitHub Copilot for code suggestions

This list changes as tools improve or fall out of use. When a specific article relied on a tool in a meaningful way beyond the general workflow above, we note it in the article itself.

Feedback and corrections

If you find an error or a passage that reads as inaccurate or unclear, please email wahle@uni-goettingen.de. We update content when readers flag mistakes.

Last updated 15 May 2026.