LLM Coding

Upload data, configure codebook variables, and code with one or more LLMs.

Please remember to cite our methodological paper if you use this tool:

Baranski, A., Cooper, D. J., & Lee, J. K. (2026). Are LLMs reliable coders of communicationcontent in economic experiments? NYUAD Division of Social Science Working Paper, #0115. View paper

Drop a CSV or Excel file here, or click to browse

Communication episode: the unit of analysis — a combination of messages exchanged through the same channel, or a collection of messages sent by one sender. Rows that share your chosen identifier(s) are merged into one episode.

Fully empty episodes are not sent to a model. Their original rows remain in the primary CSV with blank code cells.

No variables yet — click to define your codebook.

Click to edit codebook →

Each variable has its own aggregation method, category definition, and coded-value guidance. Per episode = one value per episode; per sender = one value per verified sender.

Download Codebook

Provide context about what the data represents and the research goals. Have a PDF with figures or tables? Use Import from PDF to convert it to text first. The placeholder is a constructed example.

Browser and package execution differ.Run Coding uses all models and runs configured below and requires their API keys. For package generation, CAT does not save an API key or the configured tuning settings; the downloaded package records only the first selected provider and model and makes one call per episode after the local script obtains a key at runtime.
1OpenAI — GPT-4.1 Mini
0.20
02
1.00
01
1024
648192
1 model × 1 run = 1 calls/episode

Learn CAT

What each tool does, what every field expects, and short demos to watch it in action.

Demo Video

This walkthrough demonstrates the published CAT coding workflow.