Subtext uses AI to classify every turn of Student–AI tutoring transcripts into educational moves - with confidence scores, session summaries, and exportable datasets.
0+
Turns classified
0+
Sessions analyzed
0.7%
Pipeline uptime
0-pass
AI pipeline
Features
From raw transcripts to publication-ready annotated datasets, Subtext handles the heavy lifting so you can focus on the research.
Every turn is classified into educational moves like Information Seeking, Epistemic Agency, Guiding, and Explaining. A 3-pass ensemble pipeline votes across multiple AI runs for mathematically grounded confidence scores.
Learn moreNot sure about a classification? Subtext automatically surfaces turns with low AI confidence so you can review and override them in one click.
Learn moreAfter analysis, Subtext generates a 3-sentence natural language summary of the educational dynamics and key themes - ready to paste into your paper.
Learn moreUpload .txt, .csv, .json files, paste raw text directly, or import from a ChatGPT share link. Subtext handles the parsing automatically.
Learn moreExport your annotated transcripts as CSV, JSON, or formatted PDF reports. Bulk export across multiple sessions for cross-dataset analysis.
Learn moreDisagree with the AI? Override any classification and your corrections are saved permanently - even across re-analysis runs.
Learn moreHow It Works
Drop a transcript file, paste raw text, or import from a ChatGPT share link.
Our 3-pass AI pipeline parses, classifies, and summarizes. Regex pre-processing skips the LLM when possible to save cost and time.
Browse classified turns, override labels, filter by confidence, and export your annotated dataset in CSV, JSON, or PDF.
The transcript analysis view gives you classification labels, confidence scores, session analytics, and AI summaries - all in one place.
The student demonstrates strong information-seeking behavior throughout, progressively building toward conceptual understanding. The AI effectively scaffolds learning through detailed explanations and targeted feedback. Overall epistemic agency remains low, suggesting the student is in an early exploration phase.
Classification Distribution
Metadata
Research Framework
Built on established educational discourse analysis frameworks, each conversation turn is classified into one of 8 domain-specific labels.
Information Seeking
Asking for facts or help
Conceptual Revision
Correcting their own understanding
Epistemic Agency
Taking control of their learning
Off-task
Unrelated chatting
Guiding
Asking guiding questions, scaffolding
Explaining
Providing direct answers
Feedback
Evaluating student responses
Clarifying
Asking for clarification
Pricing
Every plan includes the full AI pipeline. Upgrade when your research needs grow.
Perfect for trying Subtext on a single study.
For active researchers running multiple studies.
For labs and research teams collaborating at scale.
Join researchers at leading universities already using Subtext to analyze thousands of AI tutoring sessions.