INTELLIGENCE
Sentiment that admits when it isn't sure.
Every label carries a confidence score, sarcasm is flagged rather than guessed at, and one click tells us we got it wrong. Corrections retrain your project, not somebody else's.
How it works
-
Every mention gets a label
Positive, negative, neutral, question or buying intent — as a word, a glyph and a colour, never a colour on its own.
-
And a confidence score
A 0.55 and a 0.92 are not the same claim. Low-confidence mentions sort to the top of a review queue instead of quietly counting as fact.
-
You correct it in one click
The correction is stored against your project and moves that project's threshold. Nobody else's numbers change because of your call.
In detail
-
Sarcasm is flagged
When the model reads irony it says so rather than picking a side and being confidently wrong.
-
Buying intent as its own class
"Anyone found an alternative to…" is not positive and it is not negative. It is the one you answer first.
-
Questions separated out
A question is a support opportunity, not a sentiment reading.
-
Per-project thresholds
A gaming community and a compliance product do not share a tone baseline. Yours moves with your corrections.
What a multi-channel tool cannot do here
General-purpose sentiment models are trained on reviews and tweets, where the register is flat and the sarcasm is signposted. Reddit is neither. We classify with the thread as context — the parent comment, the subreddit, and what the conversation was already about — so "yeah, great, another analytics tool" lands where it belongs.
What this does not do
- We publish no accuracy percentage. Any number we printed would be measured on our corpus and would not predict yours; the confidence distribution in your own analytics is the honest version.
- Sentiment is scored per mention and per subreddit, never per redditor. Profiling individuals is prohibited under Reddit's terms and we do not build the capability.
- Non-English mentions are labelled, but confidence is lower and we mark them as such.
Which plans include it
| Plan | Keywords | Check frequency |
|---|---|---|
| Free | 3 | 6 h |
| Pro | 25 | 60 min |
| Business | 100 | 15 min |
| Agency | 300 | 5 min |
| Enterprise | Custom | 5 min |
Related
Questions
- How accurate is it?
- We do not publish an accuracy figure, because the honest answer depends on your corpus. What we do publish, inside your account, is the confidence distribution across your own mentions and how often you corrected us.
- What happens when I correct a label?
- The correction is recorded against your project and shifts that project's threshold. It never leaks into another customer's account.
- Do you track sentiment per Reddit user?
- No. That is individual profiling and it is prohibited under Reddit's API terms. Sentiment exists per mention and per subreddit.
- Is Reddit content used to train your models?
- No. We use language models to classify and to draft; Reddit content is never used to train one, and our AI provider is contractually zero-retention.
Your brand is being discussed right now.
Four minutes to set up. Your first mentions land before you finish reading this page.