Every post or message found by your keywords or by AI search is read by the AI. It does not just say "looks like it / doesn't" - it describes the post in facts and applies your campaign's rules to them. You do not need to add an analyst for this: the AI checks every match against the campaign description.
What the AI sees in a post
- Who is writing - a buyer, a seller (agency, specialist), an employer, a candidate, a private person, someone in a discussion, spam.
- What they are doing - asking for advice or a contractor, complaining about their own problem, choosing between options, offering services, hiring, looking for work, mentioning a brand, sharing experience.
- Is it on topic - exactly your topic, next to it, or not.
The conclusion is on the card in the AI assessment block: who wrote, what they are doing and whether it is your topic. Below it is a quote from the post that backs it up.
The AI takes the campaign's rules from what you told it: the description on the What to look for step ("Describe who you are looking for", "Who is definitely not a lead"), your answers under Here is how we read it, the Who to look for choice, and in campaigns with your own words - Micro-context for AI. On the campaign page this description is shown in the What counts as a lead block.
Three shelves
| Verdict | What it is | What happens |
|---|---|---|
| Lead (the card says "AI-approved") | A buyer asks directly for what you sell; someone mentioned your brand; a company is hiring in your area, if you turned vacancies on. | The card is in AI-selected. In Autopilot mode the reply goes out by itself. In Copilot mode there is no draft at first: the "Generate a reply: 1 credit" button writes it, and you send it. |
| Doubtful ("Make a decision: AI is unsure") | Complains about their own pain but asks for nothing; a neighbouring topic; unclear whether a buyer or a seller. | The card waits in the feed for your decision: Mark as lead or Reject lead. No reply goes out by itself. |
| Rejected (the Rejected by AI view) | Not your topic; ads and service offers; expert posts by colleagues and competitors; spam; chatter. | The card sits in Rejected by AI with the reason. If the AI got it wrong - the Mark as lead button. |
ℹ️ Note: if you press Mark as lead yourself and the campaign has an AI responder, a reply draft is written right away - also for 1 credit. You still send it yourself.
A credit is charged for the AI reading the post, not for the verdict: a rejected post costs as much as a lead, which is why it is not hidden. On Telegram, Threads, Bluesky and Hacker News whatever is dropped BEFORE the AI is free: a stop word, another language, no keyword, a post that is too short - such cards say so instead of a score. On LinkedIn, Reddit, X and Jobs every delivered post costs 1 credit, even if a filter drops it before the AI: we pay the platform for those posts.
"Why not a lead?"
When you press Reject lead, the card is replaced by the question Why not a lead? with buttons: Colleague / competitor, Offering, not looking (if the campaign looks for providers - Looking, not offering), Not my market, Off topic, Spam / bot, Other…. Answering is optional - there is Skip.
Three identical reasons within 30 days - "Colleague / competitor", "wrong side" or "Off topic" - and the AI stops showing such posts in this campaign. "Not my market", "Spam / bot" and "Other…" never become a rule. The rule is recalculated every night over the last 30 days.
Why an expert post is not a lead
A post like "how we grew a client's sales by 40%" is written by a seller. For a campaign looking for buyers it is a competitor's ad - it goes to the rejected ones, even if it contains all your keywords. If your customers are agencies and specialists themselves, choose People who offer it or Both in Who to look for (Edit → What to look for). The AI also learns from your decisions: mark such cards as leads a few times and it starts letting them through by itself.
Need discussions of the topic, not just requests? Under Here is how we read it, in the What else counts as a lead? block, tick Discussions of the topic - people talking about it without asking. Vacancies and resumes on your topic are switched on there too.
The confidence threshold
Every AI decision has a confidence. Below the threshold the AI does not call a lead by itself: the card comes to you marked "Make a decision: AI is unsure", and in Autopilot mode no reply to it goes out by itself. The campaign already has a threshold set. If an analyst is attached to it, the threshold is changed in the analyst's settings - the Score threshold (0-100) field.
What this means in practice
- Too much noise - describe who you are NOT looking for, answer "Why not a lead?" and add Stop words: such a post is cut off before the AI.
- Too few leads - see "Too many or too few leads".
- Autopilot sends too much - switch the responder to Copilot: drafts will wait for your confirmation.
Related articles
- "What to look for: description, keywords and stop words"
- "Feed: views and filters"
- "Replying to a lead"