AutoGEO
Research code that learns generative-engine preferences and rewrites content
What it is
Code for an ICLR 2026 paper from CMU. It mines content-preference rules from generative search engines, then rewrites documents to raise their visibility in generated answers while preserving accuracy, either through a prompt-based API model or small released fine-tuned checkpoints. For researchers and technical SEO people testing GEO claims empirically.
Good for
- Reproduce GEO experiments from the paper
- Extract engine preference rules for a domain
- Benchmark a rewriting approach
Install or run
git clone --recursive https://github.com/cxcscmu/AutoGEOCommand as given in the project README. Read the repository before you run anything on your machine.
What you need
Conda environment; model API keys in keys.env
Sources
- https://github.com/cxcscmu/AutoGEO (project README, read 10 Nov 2026)
- https://api.github.com/repos/cxcscmu/AutoGEO (GitHub API, read 13 Nov 2026)
The summary is ours, written from the README. Spotted an error? See how we work.