Relevance and labels
How FishOut decides whether a post is about you, what the post is doing, and whether it is worth a reply.
Every post that matches a keyword is read by the AI once, with your project's Brief beside it. What comes back is a handful of plain judgments rather than one number: how relevant the post is and why, what kind of post it is, how the author feels, what language it is in, and whether it is worth answering.
This is AI triage. It runs on every plan, including Free, and is never metered.
The Brief
The Brief is two to six sentences on the project's settings page: what you do, who it is for, and what to ignore. It is drafted from your website when you sign up, and you should edit it — it is sent with every post the AI judges, so it is the single biggest lever on what reaches your inbox.
Acme Mail is an email client for small support teams. Our buyers are support leads at companies of 5–50 people. Ignore posts about the Acme cartoon brand and general email deliverability news.
A keyword's Means… line does the same job for one term. See Keywords.
Relevance
Each post gets a relevance from 0 to 100, shown as a bar. It answers one question: is this actually about your product, your competitor or your topic, or is it a coincidence of words?
Beside the bar, Why chips say what the AI saw:
| Chip | Means |
|---|---|
| Matches your brief | The post is about what your Brief describes |
| Means the product, not the word | The keyword is used in your sense, not another one |
| Asking for help | The author wants an answer or a recommendation |
| Not answered yet | Nobody has given them one |
| Fresh | Recent for its platform — hours on X, a day on Reddit, a few days on YouTube |
| High reach | Seen by more people than most posts on that platform |
A chip only shows when the AI is confident, so a post with no chips is not a post with bad news — it is one where nothing stood out.
Filtered
Posts below the relevance threshold do not reach the inbox. They go to the Filtered view instead, each with the reason:
| Reason | Means |
|---|---|
| Different meaning | The post uses the word, but means something else |
| Spam | Spam, bot output or an advertisement |
| Off-brief | About the term, but outside what your Brief asks for |
| Below threshold | Related, but not relevant enough to show by default |
The threshold is 40, and it is not a setting. If too much or too little gets through, the fix is the Brief, a keyword's Means line, or teaching — not a slider.
Filtered posts never count toward your plan's relevant mentions. A broad keyword that matches a lot of junk costs you nothing. See Plans.
Open Filtered now and then. It is where you check the AI's work, and Relevant after all moves a post back to the inbox. If you would rather see everything, switch off Hide low-relevance posts in the project's settings.
Labels
A label says what a post is doing. A post can carry several, and the AI applies them to every new post.
Built-in labels
Nine, on for every project and never counted toward your plan:
| Label | The author… |
|---|---|
| Asking for recommendations | asks others to recommend a product, tool or service |
| Comparison | compares two or more products or weighs alternatives |
| Has a problem | describes a pain or problem they have not solved yet |
| Question | asks a question that expects an answer |
| Complaint | complains about a product, company or experience |
| Praise | praises or recommends a product, company or experience |
| Announcement | posts a launch, release, funding round, news item or official update |
| Promotion | self-promotes, advertises or posts affiliate content |
| Discussion | discusses something that fits none of the above |
Switch one off on the Labels page and new posts stop getting it; posts that already have it keep it.
Your labels
Your own labels cover what only your project cares about — Pricing objection, Asks about SSO, Mentions a migration. Give one a name and a one-sentence description, the way you would explain it to a new teammate, and the AI applies it. You can add a Don't apply when… line and up to three examples, and Test on recent posts tries the description on your last 50 relevant posts before you save it.
Leave the description out and it is a manual-only label: the AI never applies it, you do, and it does not count toward your plan.
Paid plans
Labels the AI applies need a paid plan: 3 on Plus, 10 on Pro and 30 on Business. Manual-only labels are unlimited on every plan.
Compare plans.Sentiment and language
Sentiment is Positive, Neutral or Negative. Language is a chip on each post and a filter in the inbox. FishOut shows posts in every language; the language chip is how you narrow to the ones you can answer.
Reply call
On Reddit, Hacker News and YouTube, each relevant post also gets a Reply call:
| Call | Means |
|---|---|
| Reply now | Someone is asking for help, the thread is fresh, and nobody has answered yet |
| Reply, carefully | A reply could help, but read the room first |
| Watch | Already answered, posted by a vendor, or getting old. Worth knowing about, not answering |
| Skip | Promotion, bot output, or a thread where a reply would read as self-promotion |
A call to reply comes with a hint — Answer the question directly, Lead with help, don't pitch or Don't include a link. The call ages with the post, so a Reply now from this morning can be a Watch by tomorrow.
Turning features off
Sentiment, Language, Reply call and Hide low-relevance posts can each be switched off per project, under the project's settings. Turning one off stops the AI asking about it and hides it everywhere in the app; views that depend on it hide too.
Teaching
When the AI gets one wrong, tell it. Teach on any post offers:
| Action | Does |
|---|---|
| Not for us | Archives the post. Similar posts will rank lower |
| Relevant after all | Moves a filtered post to the inbox. Similar posts will rank higher |
| Fix | Removes a wrong label, adds a missing one, or corrects the sentiment |
| Mute | Drops new posts from this author or this source before the AI sees them |
Your team's most recent 40 corrections are sent to the AI as examples with every post, so teaching changes what happens next, not just the one post. Review what I've taught, on the project's settings page, lists them with undo.
When a judgment looks wrong
Read the Why chips first — they usually explain the gap. Then fix the cause, not the post:
- Wrong sense of the word. Add a Means line to the keyword, or an exclude word for the collisions you can name. See Keywords.
- Right topic, wrong audience. Say what to ignore in the Brief.
- One-off misreads. Teach. A handful of corrections moves more than you would expect.