> For the complete documentation index, see [llms.txt](https://themochi-app.gitbook.io/documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://themochi-app.gitbook.io/documentation/module-4-ai-setup-and-intelligence/spam-filter.md).

# Spam Filter

## Spam Filter: Keep Junk Out of Your Inbox

**What it is** — The spam filter reads incoming DMs against a prompt you write and automatically routes junk straight to Unqualified, before it ever reaches a setter's inbox. A well-tuned filter typically removes an estimated 20–30% of a typical inbox.

**Who can use it** — Configuring the filter lives in `Settings > Intelligence`, and like the rest of Settings, it's expected to be a **Creator/Manager** action. Available on **web and iOS**.

**Before you start** — Spend a few days collecting real examples of spam before you write the prompt. You'll get a far more accurate filter than guessing at patterns from memory.

### What the spam filter removes

Typical spam patterns worth filtering:

* Unsolicited service pitches ("I'll scale your business," agency/closer pitches, often with a Loom link attached).
* Follow-for-follow requests.
* Money requests.
* Generic compliments with no actual question ("love your content!" and nothing else).
* Bot-like or copy-paste patterns.
* Emoji-only messages.
* Cold outreach that leads with results or credentials before any real conversation.
* Flattery that pivots into a pitch a message or two later.
* "Free value" foot-in-the-door messages that are really an opener for a pitch.

### Build your filter prompt

1. **Screenshot spam DMs for 3–5 days** before turning the filter on. Don't rely on memory — real examples produce a sharper prompt.
2. **Upload the screenshots to Claude or ChatGPT** and ask it to find the patterns and write them up as a filter prompt. Structure the result as: an opening instruction, named pattern categories, the intent signals that identify each one, broadening rules that extend each pattern to catch close variations, and — critically — explicit **"Do NOT filter if…"** exceptions for anything that looks similar to spam but is actually a real lead.
3. Refine the draft until it's specific enough that a new hire could apply it consistently.

### Turn it on

1. Go to `Settings > Intelligence > Spam Filter`.
2. Paste your prompt.
3. Click **Save**.
4. Toggle the filter **On**.

### Monitor the first 7 days

Check the Unqualified tab daily for the first week after activating. Manually move any false positives back to New Lead. Plan to re-tune the prompt every 2–3 weeks for the first month, then monthly after that as new spam patterns show up.

### Tuning: "Do NOT filter if…"

* **Filter too aggressive?** Add explicit "Do NOT filter if…" exceptions for the legitimate patterns it's catching by mistake. At least one account has had the filter fully disabled after it started wrongly filtering real prospects — don't let it get that far. Check Unqualified daily in week one and tune early.
* **Filter missing spam?** Add the new pattern you're seeing to the prompt, with a couple of real examples.

The filter isn't all-or-nothing in practice. Even with it switched off, a setter can still manually unqualify an obvious spam message themselves.

### Tips

* Write the prompt the same way you'd brief a new hire: plain language, specific patterns, specific exceptions.
* Keep a running note of new spam patterns as they show up so your next tuning pass is fast.
* Don't over-tune in week one based on a handful of messages — give it a real week of volume before making big changes.

### Common mistakes & fixes

| Mistake                                                      | Fix                                                                                        |
| ------------------------------------------------------------ | ------------------------------------------------------------------------------------------ |
| Turning the filter on without collecting real examples first | Spend 3–5 days screenshotting spam before writing the prompt.                              |
| Skipping the first-week monitoring                           | Check Unqualified daily for 7 days — unnoticed false positives cost you real leads.        |
| Writing a prompt with no exceptions                          | Always include "Do NOT filter if…" cases for anything that could plausibly be a real lead. |
| Never revisiting the prompt after setup                      | Re-tune every 2–3 weeks for the first month, then monthly — spam patterns evolve.          |

### Limits & good to know

* The 20–30% removal figure is a typical estimate, not a guarantee — actual results depend on your inbox and how specific your prompt is.
* The filter is not all-or-nothing — setters retain the ability to manually flag an individual message as spam even when the automated filter is off.

**Related:** Set Up Your AI: Setter Playbook & Creator DNA, AI Tags, Lead Summary. Also see **The Leads Page** and **Pipeline Stages** (Module 02).