Letter Writing & Communications
Fundraising letters, thank-yous, and tone-matching prompts.
On this page
- What the group settled on
- The three workable patterns
- 1. Upload examples; let it copy your voice
- 2. System-prompt rules
- 3. Dia / writing-aware browser
- Specific letter use cases
- Fundraising letters at a 6th-grade reading level
- Press releases
- Grant proposals
- Eulogies / hespedim
- Condolence / yahrtzeit notes
- "Sounds too AI": fighting the tells
- Honest-feedback prompts
- Pre-response clarification
- ZeroGPT check
- Related
Drafting letters, emails, and short writing in a voice that sounds human, and ideally like you. A foundational use case from the chat's first weeks.
What the group settled on
| Use case | Model |
|---|---|
| Pastoral / emotional / "knowing your baalei batim" | Claude |
| Personality / fun / casual tone | Grok |
| Default workhorse | ChatGPT |
| Style-matching past work (uploaded examples) | Claude with examples in the chat, a Project, or a Custom GPT |
"Claude is better in my experience than ChatGPT. Provide a good outline of what you want the letter to say."
"I find that Grok writes with more personality and a more genuine tone than ChatGPT."
The three workable patterns
1. Upload examples; let it copy your voice
A eulogy workflow:
- Upload 20+ past eulogies you've written.
- Add a system message: "Copy this style for future writing."
- When a new eulogy is needed, take notes with the family, paste them into the same chat, and get a draft that matches your prior style (similar openings, similar closings).
This generalizes to thank-you letters, sponsor letters, condolence notes, and press releases.
2. System-prompt rules
The most-shared writing prompt in the chat, an anti-AI style sheet. The key constraints:
- No em-dashes ever (use commas, semicolons, or two sentences).
- No "leverage", "synergy", "tapestry", "delve", "in conclusion".
- Active voice. Short paragraphs. Vary sentence length.
- A Flesch reading score of 80 or higher.
- Avoid adverbs. No clichés. Plain English.
- Acknowledge with "Got it" and then respond to the actual request.
See the full prompt in Master Prompts.
3. Dia / writing-aware browser
Dia has a built-in AI assistant where you pre-load your writing style and sample authors once, and then everything you write gets ghostwritten in that voice without re-prompting each time.
Specific letter use cases
Fundraising letters at a 6th-grade reading level
The Jeff Brooks technique, adapted:
- Write your own draft.
- Ask AI: "Rewrite this at a 6th-grade reading level while keeping my voice and main points."
- The result is more accessible and historically gets higher response rates.
Press releases
Generating press releases plus media-list customization is a chat specialty. The pattern: write one well-targeted release; generate ten variations per outlet; check before sending.
Grant proposals
Paste the grant questions into AI with your Chabad House context; AI fills out the answers. Then you edit. It saves hours.
Eulogies / hespedim
The pattern above. Also use AI to research the deceased, the names, dates, family members, and public mentions, before the family meeting.
Condolence / yahrtzeit notes
Use the recipient's name and one specific memory; AI writes the connective tissue. Don't trust AI to find the memory, that is your job.
"Sounds too AI": fighting the tells
Common AI tells in letters:
- Em-dashes everywhere. Strip them.
- "In today's world, ..." / "In conclusion, ..." Cut these.
- Triadic sentences. "Faster, smarter, kinder." Three is the AI default.
- Repeated framings. "Not just X, but Y." "It's not about A, it's about B."
- Verb-heavy openings. "Imagine." "Picture." "Consider."
The 2026-04 TechCrunch article "AI Writing: It's not just this, it's that" catalogs the patterns at length.
Honest-feedback prompts
When the AI is sycophantic ("That's a great idea! You're brilliant!"):
"Be direct and brutally honest. Stress-test my ideas hard. If something is weak, say so plainly and explain why. Think from first principles, use logic over sentiment, and push until the argument is solid."
"Recently I asked GPT to criticize my output and be 'merciless' about it. It worked."
"After AI response: 'Rate your response out of 10.' If not satisfied, give your own rating, e.g., 'I'd rate that a 6/10.' AI recalibrates its understanding of what you need."
Pre-response clarification
"Ask me clarifying questions to make sure we're on the same page before responding or doing a task."
A small prompt addition that prevents misinterpretations on complicated writing tasks.
ZeroGPT check
"Always paste your text into zerogpt.com or similar before posting to see if it's too much AI."
Related
- Master Prompts, the anti-AI style sheet and other prompts
- Claude
- ChatGPT
- Grok
- Fundraising