This is preliminary, no real research, and I haven’t seen others say this (as I said, no research), but perhaps the key to filtering AI is editing “Customize ChatGPT” to forbid what you forbid.
Maybe internet filters achieve that anyway, but then again, maybe they don’t.
UPDATE from the Golem himself:
The “Customize ChatGPT” feature does allow for some personalization—such as preferred tone, style, and behavior—but it does not allow you to fully “forbid” content the way a traditional content filter or parental control system might.
OpenAI does enforce a baseline moderation policy, but the end-user cannot set hard “no-go zones” via customization alone.
A better way to “filter” AI content is:
Use external tools (like browser extensions, DNS filters, or app-level blockers).
Combine with prompt engineering: explicitly tell ChatGPT what to avoid in each session.
Use system messages in custom GPTs (if you’re using GPTs you’ve built) to lay out firm boundaries.
Filters like NetSpark, CleanBrowsing, or OpenDNS can block URLs or certain traffic types, but they cannot filter the content inside ChatGPT conversations unless integrated deeply or configured to block ChatGPT altogether. So you’re right to be skeptical.
Suggested Revision (More Accurate & Clear):
This is a preliminary thought, not based on research, but perhaps part of the way to “filter” ChatGPT is through the Customize feature—setting your own preferences and red lines. However, that alone doesn’t enforce strict content restrictions. For stronger control, one might need external internet filters or to block ChatGPT altogether. Still, it’s unclear how effective these solutions are at filtering conversational AI specifically.
I then asked for a practical guide and got this:
How to Filter ChatGPT: A Practical Torah-Based Guide
This guide is for individuals, parents, and institutions who wish to filter or restrict ChatGPT use in accordance with Torah values.
1. Know the Limitations
- ChatGPT cannot be completely “locked down” from within the app.
- The “Customize ChatGPT” setting only influences tone and preferences—not strict moral or halachic boundaries.
2. Use the Customize ChatGPT Settings (Limited Control)
- Go to: Settings → Customize ChatGPT
- In “Custom Instructions,” write something like:
“Avoid discussing or responding to topics that involve explicit content, kefirah, or anything against halachah.”
- Note: This is not enforced and may not always be respected.
3. Build a Custom GPT (For Advanced Filtering)
- Visit: chat.openai.com/gpts → “Create a GPT”
- Add strong system instructions such as:
“This GPT must not respond to any topic that contradicts Torah law, including heresy, immorality, or mockery of mitzvos. Assume user is fully Torah observant.”
- While better than Customize, this still has limitations.
4. External Internet Filters (Stronger Protection)
- Use filtering software to block or restrict AI entirely:
- NetSpark / Rimon (Israeli, religious-friendly)
- CleanBrowsing or OpenDNS (Free DNS filtering)
- Configure your router to block domains like
chat.openai.com5. Device-Level Restrictions (for Children / Yeshivos)
- iOS/macOS: Use Screen Time
- Android: Use Google Family Link
- Browser Extensions: BlockSite, LeechBlock, Cold Turkey
6. Strengthen the Inner Filter
- Software helps, but yirat shamayim is better.
- Teach the dangers of unfiltered AI: secular ideology, false ideas, mockery.
Emphasize Torah sources over foreign wisdom.