Shiur & Curriculum Preparation
Using AI to research, outline, and present Torah content.
On this page
Using AI to turn a sicha, a maamar, a Gemara sugya, or an article into a polished shiur or curriculum.
Where the consensus has landed
Claude wins this category. Repeatedly tested against ChatGPT and Gemini for organizing sichos into 60-minute shiurim with discussion questions, teacher guides, and student handouts. The pattern:
"I tell it to organize into a 60-minute shiur with discussion questions, summaries, and a follow-along printout, teacher's guide and student handout. Does an amazing job."
"I tried with ChatGPT and Gemini, and what they came up with was very poor in comparison. Claude was beautifully formatted."
"Claude beautifully extracted the nekuda of the maamar and presented it in a way that a 13-year-old can understand. It was the biggest nachas to see a boy from my community chazering a maamar at his bar mitzvah."
The pattern
- Source the text. Pull plain text from a vetted corpus: Dach.dev, NotebookLM (for the Rebbe's Likkutei Sichos, Igros Kodesh, etc.), Sefaria, or upload your own PDFs.
- Hand the whole thing to Claude. Don't ask it to fetch the sicha; that is where hallucination starts. Paste the sicha.
- Specify the format.
"Create structured, detailed and comprehensive speech notes based on this talk. It should follow a logical structure. I'm not looking for a verbatim speech script, only content notes. Include all stories and anecdotes in full."
- Add the audience and use-case. "60-minute shiur, mixed-level adult class, English with Hebrew terms transliterated."
- Request the artifacts explicitly. Teacher's guide, student handout, discussion questions, exit-ticket questions, a printable PDF (Claude can use
pptxand HTML-to-PDF skills). - Add Q&A and stories. Claude will, if asked, embed niglah stories with proper context.
- Verify. Cross-check any cited mareh-mekomos against the source. The model will fabricate them otherwise.
Audio-first workflow
You recorded the shiur instead of writing it:
- Transcribe with Sofer.ai (Hebrew/Yiddish/English mix), or TurboScribe (English), or Whisper. See Transcription.
- Paste the transcript to Claude.
- "Summarize this shiur and produce a one-page printable PDF." This pattern runs Sofer.ai to Claude.
- Iterate: longer version, shorter version, focus on specific points.
NotebookLM as a corpus tool
Special-purpose use of NotebookLM: upload an entire shas or all of Likkutei Sichos as the sources. NotebookLM will cite chapter and verse, but it caps at about 12 sources per query, so structure prompts to demand specificity:
"I just searched for 100 insights into the yomtov of Pesach and it provided from 109 different sources."
A shared community notebook of the Rebbe's Torah (Likkutei Sichos, Igros Kodesh, Toras Menachem, Maamorim) has been circulated for exactly this.
Audio Overview & podcasted shiurim
NotebookLM's Audio Overview generates a podcast-style two-host discussion from any uploaded sources. One community example: a sicha converted into a structured Rosh Hashanah speech. Then Interrupt lets you join the conversation as a third voice.
Drosho-specific tips
- "Roast this!" After pasting your speech, ask the model to critique it mercilessly. Surprisingly useful.
- Read it aloud. ElevenLabs V3 handles Hebrew with ches/chet pronunciation if you write Ashkenazi-style with nikkud.
- Cross-language. AI can take a Yiddish sicha and produce an English shiur, if you give it the text. See Hebrew & Yiddish.
Custom GPTs / Gems for repeated use
Some shluchim build custom GPTs / Gems with their preferences baked in (preferred opening prayer, recurring stories, brand voice). See ChabadOne Integration for analogous patterns on the website side. As one member put it: hard-code your preferences, the flyer styles you like, past examples, and brand colors, and you will get more consistent results.
Open questions
- Reverse mafteach inyanim: given a page of a sefer, list every inyan discussed. One report noted that "Gemini failed miserably." The best path is to have AI write a Python script over indexed data, not run inference each time.
- An AI lecturer that reliably handles five or more acharonim without flattening them, an open question across AI and Torah Accuracy.