What it does
The spam scorer adds up points for common spam signals: links and URL shorteners, link-in-bio pages, heavy capitals, emoji floods, hashtag stuffing, mass mentions, scam and follower-selling phrases, phone numbers, crypto wallet addresses, look-alike letters and near-duplicate copies within the batch. Scores at or above your threshold are marked likely spam; half the threshold or more are marked for review.
How is the score calculated?
Each warning sign adds a fixed number of points, the total is capped at 100, and the list under each post shows every signal that fired.
| Signal | Points | Why it matters |
|---|---|---|
| Exact copy of another post in the batch | 25 | Copy-paste campaigns |
| Near-identical copy (70%+ shared word triples) | 18 | Same campaign with small edits |
| Hashtag stuffing (7+) | 18 | Riding trending tags for reach |
| Mass mentions (8+ accounts) | 20 | Tag spam to force notifications |
| Mostly capital letters | 15 | Shouting is common in scams |
| Crypto wallet address | 15 | Payment requests, fake refunds |
| Mixed alphabets in one word | 14 | Look-alike letters that dodge filters |
| Emoji flood (10+) | 15 | Attention bait |
| URL shortener | 12 | Hides the destination |
| Little text besides the link | 12 | Drive-by link drops |
| Spam and scam wording | up to 36 | "free followers", "seed phrase", "DM me" and about 60 more |
| Income per day or week | 10 | Get-rich offers |
| Links | 5 for one, +10 each after | Most spam wants a click |
| Phone number, link-in-bio page, money symbols, !!!, stretched words | 3 to 6 each | Weaker signals that add up |
The default threshold is 50. Anything from 25 to 49 is marked for review.
How should I use the score?
Use it to sort, not to sentence: auto-hide only the very high scores and let a person read the middle band.
A sensible setup for a busy account or a social wall:
- Paste the latest batch of replies or queued posts.
- Anything above 70 can usually be rejected without reading closely.
- Read everything between 25 and 70. That's where false positives live: excited fans with lots of emoji, partners posting several links.
- Download the CSV and keep it with your notes. Patterns across batches tell you which phrases to add to your block list.
What does it miss?
It misses anything that doesn't show in the text: account age, posting speed, images, and spam written in plain, polite language.
Newer spam often reads like a normal comment ("This is so true, I wrote about the same thing") with the payload in the profile instead of the post. A text scorer can't see that. Neither can it read images or video. Treat a low score as "nothing obvious", never as "safe". Our 2026 spam patterns guide covers the kinds that slip through.
Does it work for languages other than English?
Partly: the structural signals work in any language, but the spam phrase list is English only.
Links, shorteners, capitals, emoji, hashtags, mentions, duplicates and mixed alphabets are counted the same way in any language that uses them. The phrase list, around 60 entries like "work from home" and "claim your", is English. For other languages, build your own phrase list in the rule builder.
Common questions
Is a high score proof that a post is spam?
No. The score counts warning signs in the text. A product launch post with three links and five hashtags can score high and still be legitimate. Use the score to sort a queue, not to make final decisions about people.
Why are duplicates scored so high?
Copy-paste campaigns are one of the most reliable spam signals. When the same message, or one with a word or two changed, arrives from several accounts, it's rarely a coincidence. The scorer compares every post with every other post in the batch.
Can it check account age, follower counts or bot behavior?
No. It only reads the text you paste. Account signals such as a brand-new profile, a default avatar or a burst of replies in a minute are strong spam indicators, but they need platform data that a browser tool doesn't have.
How many posts can I paste?
A few hundred works smoothly. Duplicate detection compares every pair, so very large batches take longer; split them up if the page slows down.