Prompt
I’m unhappy with OpenAI Moderation because it misses some spammy posts
Latest observation
Nov 2, 2026
OpenAI APIWeb search: off
If you’re using OpenAI Moderation to catch spam, the problem is that moderation is mainly designed for policy/safety classification, not as a dedicated spam detector. So it can miss obvious “spammy” content, especially if it’s benign-looking, repetitive, promotional, or low-context.
A better approach is usually a layered anti-spam system:
-
Heuristics first
- repeated posts
- too many links
- keyword stuffing
- very short generic text
- excessive hashtags/emojis
- suspicious usernames / posting velocity
-
Behavioral signals
- account age
- IP/device reputation
- posting frequency
- copy-paste similarity across posts
-
LLM or classifier for borderline cases
- classify “spam / not spam / maybe”
- use your own examples to fine-tune prompting or train a model
-
Threshold tuning
- moderation APIs often need conservative thresholds
- combine scores from multiple checks rather than relying on one call
-
Human review for uncertain cases
- especially if false positives matter
If you want, I can help you design a more reliable spam detection pipeline or suggest a prompt/classifier setup that works better than moderation alone.
Brands