# Leads and replies > Find people who need your product but don't name it, reply with a drafted message, and see which replies get clicks A tracked problem surfaces a handful of genuine requests a month, not a daily feed. ## Leads A brand keyword finds posts that name your product. Leads come from **problems**: one line stating a problem your product solves, like "Recording product demos takes hours". Each problem is searched through two or three **phrasings**, the words people who have it write: a pain ("recording demos takes forever"), an intent ("storylane alternative") and a question ("how do you record product walkthroughs"). Every phrasing is a lead keyword. ### Problems On the **Keywords** tab, type a problem in plain words and click **Add problem**. Claude writes three phrasings and they start scanning at once, each at the default pace. Each problem shows its leads over 30 days, what it costs a day, and its phrasings with a small search-quality bar. **Phrasings prune themselves.** Once a phrasing has had 40 results checked in 30 days, it pauses when most results are sellers (under half not selling), mostly off topic (under 30% on topic) or under 10% of its scored posts reach your minimum fit. A paused phrasing is not scanned; the card says why and the daily digest lists it. **Resume** scans it again and it won't be paused again for 30 days. Once a week, a problem with fewer than two running phrasings gets one suggested replacement, written from its paused phrasings and the leads you kept or dismissed. **Add** tracks it; **Dismiss** waits for next week's. Suggestions are free; the phrasing costs what any lead keyword costs. Lead keywords from before problems are listed under **Other searches**. Move one into a problem with its **Move into** select. ### Lead profile Open **Leads → Lead profile** and describe your ideal customer, problems solved, alternatives, buying signals, and exclusions. Fields are optional. Website suggestions prefill the form for review; unsupported details should stay empty. New projects started from a domain include this profile in their review card. **Suggest problems from your profile** proposes problems with their phrasings. Click **Add** to track one, or add your own. New projects started from a domain offer problems on their review card. Each phrasing starts at every 6 hours on X, Reddit and LinkedIn and daily elsewhere, and adds scan costs. Suggestions alone do not start scans. Posts need not contain the exact phrase, so Overhear ranks each source's results by how on-topic they are, whether the author is asking for help and whether they are selling, and qualifies the top few. **Search quality** compares up to three tracked phrasings over the last 30 days, listed under their problems: how much of what each search returns is on topic, asking and not selling, and how many of the kept posts reached your minimum fit, showed active intent or were posted within a week. Phrasings with fewer than 10 results are faded. The same numbers pause noisy phrasings. Once you have dismissed, saved or replied to five leads, the card also compares the median fit and gate rank of leads you dismissed with those you saved or replied to over the last 90 days. Scoring tracks your judgment when the saved line is clearly higher. ### Fit, intent and evidence **Fit** measures customer suitability from 0 to 100, judged against the problem, not the search phrase: a post that shares the words but not the problem scores low, and the reason says how the author relates to the problem. It is not a purchase probability. Missing role, company or requirement details remain unknown. Sellers, job ads and explicit profile exclusions receive zero fit and stay hidden even when the minimum fit is zero. **Intent** is assessed separately: - **Actively looking:** asks for recommendations, alternatives, or a solution. - **Exploring a problem:** describes first-person pain without seeking a solution. - **No buying intent:** no buying signal in the post. - **Intent unclear:** insufficient evidence. Each new assessment shows a reason and a supporting quote checked against the source post. The source link and post age let you judge the evidence and timing. Missing dates are shown as unknown. **Author profile.** For a lead from X at or above your minimum fit, Overhear looks up the author's public profile once (1 credit) and shows the bio and follower count on the card; the reply draft uses the bio too. Other sources show no profile. Dismissing a lead deletes its bio. **Scoring learns from your clicks.** The project's last four kept (saved or replied) and eight dismissed leads, with the reason you gave, go into every later assessment as examples, across all its phrasings. Dismiss with a reason when you can: "Seller" and "Excluded" teach the flags, "Wrong audience" lowers fit for people like them, "Not asking" teaches intent. The **Leads** tab shows results at or above the project's **minimum fit** (60 by default, under **Lead fit** in **Settings**), ordered by fit, then intent, then date. Lower scores remain stored but hidden. New email alerts require active intent and fit of at least 80, or your minimum if higher. Profile changes apply to future scoring. Existing results retain their assessment; results from before separate intent scoring are labeled **Previous score**. > **Tip:** Specific phrasings find buyers; generic ones find noise. "arcade alternative" also matches music plugins and arcade games. Phrasings with a competitor's name or the problem in your customer's words work best. ## Drafting a reply Every lead and mention has **Draft reply**. Claude writes a reply in the voice of your team that: - answers what the person asked first, - says plainly that you work on the product, - includes a tracked link to your site once, if the project has a domain. Edit the draft, **Copy** it, **Open post** and paste it there. Overhear never posts for you. Then click **Mark replied**. **Redraft** writes a new version and keeps the same link. Each draft and each **Redraft** costs 15 credits. ## Tracked links and clicks The link in a draft looks like `https://overhear.dev/r/11e3c3ea8a`. It counts the click and sends the reader to your homepage with `utm_source=overhear`, `utm_medium` set to the source (`reddit`, `x`, …) and `utm_campaign=reply`. - Bots and link previews (X, Slack, Discord fetch every link they show) don't count. - A click marks the reply as sent, even if you forgot **Mark replied**. - Replies you post on X from an account listed under **Your X accounts** in Settings mark the post as replied too, within 6 hours and at no credit cost. Only a direct reply to the post counts. - Signups happen on your site, so find them in your own analytics under `utm_source=overhear`. The top of the **Leads** tab shows the funnel for the last 30 days: **Leads**, **Replied**, **Clicks · from replies**.