How to find people asking for your product

Start with the question, not the keyword

A keyword alert answers “does this post contain my words?” A buyer is asking “who can solve this problem, and is it worth switching now?” Those are different questions, and they produce different lists.

Compare two posts that contain the same words. One is a listicle called “10 best databases for startups.” The other is a comment: “We are moving off X because the connection pooler keeps dropping. Is there a Postgres host that just works?” Both match. Only the second has a decision behind it.

Where the questions actually are

Start with two sources you can check daily. A broad crawl of “the open web” sounds better than it is: most results are SEO pages, not people. Reddit and Hacker News are small enough to read and specific enough to score.

Query patterns that find questions

Search for the language of a decision, not the language of your category. These phrases keep showing up in threads where someone is about to choose:

• "looking for" and "any recommendations" — an explicit ask.
• "alternative to" and "moving off" — a switch is already being considered.
• "has anyone used" and "worth it" — evaluation stage.
• "frustrated with" and "keeps breaking" — a pain with a workaround.
• "budget", "pricing", "self-host" — a constraint that narrows the options.

Add the buyer’s context, not your product name. “Postgres for a side project” finds better threads than your brand name ever will.

site:reddit.com/r/selfhosted "alternative to" postgres

Score the thread before you write a reply

Read the whole post and the first few comments. Score five things, 0–20 each. The total is a prioritization tool, not a prophecy.

A score of 60 or more is worth a reply. Below that, save the search and move on. LaunchEngine uses the same shape, keeps the reasons, and defaults to a threshold of 60.

Reply like a helpful stranger

The best reply answers the question first. Mention your product only when it is directly relevant, and say that you work on it. That is the whole etiquette.

• Answer in the first sentence. No preamble, no “great question.”
• Disclose your affiliation in the same sentence: “I work on X, so take this with that in mind.”
• Do not paste a link unless someone asks for one. A link in a first reply reads as an ad.
• Do not revive a thread older than a few days. The buyer has moved on.
• Do not send an unsolicited DM. It is the fastest way to get reported.

LaunchEngine’s drafting prompt follows these rules: it forbids naming any product, including yours, keeps drafts under 120 words, and marks every draft requires_human_post. The product cannot post for you.

A 45-minute weekly workflow

1. Monday: run the saved searches for the last seven days.
2. Triage: score each thread and discard anything below 60.
3. Draft: write the answer without your product name, then decide whether the disclosure belongs.
4. Post: one reply, from your own account, after reading the community’s rules.
5. Log: thread URL, date, score, reply, and outcome.

After a month, the log tells you which sources produce conversations rather than clicks. Frequency matters more than volume: five thoughtful replies a week beat fifty automated ones, and they do not put your account at risk.

What to automate, and what not to

Automate discovery, deduplication, scoring, and drafting. Do not automate posting. Reddit’s rules, Hacker News’s norms, and the reader’s trust all assume a human decided to answer.

LaunchEngine scans Reddit, Hacker News, and the open web, scores each post 1–100, and drafts a reply you review and post. The free plan includes 100 credits, one project, and one scan a day. The same signals are available to agents over REST and MCP.