Find people on X asking for what your product does
Review a short list of recent buying signals every four hours, with direct post links, fit reasons, and helpful reply drafts.
What you’ll produce
A ranked shortlist of posts from the last 24 hours, with author handles, intent labels, relevance scores, and reply drafts. Each report ends with “What needs my attention?” and up to three strongest opportunities.
Inputs & connections
- Product description, target customer profile, and clear exclusions
- Competitor names, customer problem phrases, and examples of relevant and irrelevant posts
- A persistent log of previously reported post IDs and handles, a report destination, and a timezone for the four-hour schedule
Requires a configured Grok Bot environment with authorized X search access, a persistent deduplication log, and a report destination. Set up the recurring four-hour schedule separately after reviewing a manual sample. This directory does not schedule runs or connect your X account.
Your starting prompt
Replace the bracketed fields. This is our suggested prompt, not a quote from the source.
Act as a lead discovery agent for [PRODUCT]. Product description: [WHAT YOUR PRODUCT DOES] Target customers: [WHO YOUR PRODUCT HELPS] Competitors and exclusions: [COMPETITORS, GEOGRAPHY, AND EXCLUSIONS] For each run of a separately configured four-hour schedule, search X for posts published within the last 24 hours. Record the run time and timezone. Focus on five types of posts: 1. Requests for recommendations for products like mine. 2. Complaints about problems my product solves. 3. Frustration with competitors or requests for alternatives. 4. Descriptions of workflows my product could improve. 5. Explicit requests for help with something my product does. For every relevant post, provide the original post text available to you and its direct link, the author's X handle, publication time, a specific fit reason tied to the product description, intent (actively looking, exploring alternatives, or expressing frustration), a relevance score from 1–10 with a one-line reason, and a suggested natural reply that helps the person. If no intent label fits, say unclear rather than forcing one. Reply drafts must make only supported product claims and must not imply that I am an unaffiliated customer. Prioritize explicit requests for a solution over general conversation. Use 8–10 for an explicit need and strong customer fit, 5–7 for a relevant problem with uncertain purchase intent, and 1–4 for weak fit; scores are judgment calls, not purchase probabilities. Exclude promotions, apparent bots, irrelevant mentions, and posts older than 24 hours. Bot detection is uncertain; flag ambiguous cases rather than claiming certainty. Never invent posts, links, user details, or quoted text. If the original cannot be verified, leave it out of the lead shortlist and note the access limitation. Check [PERSISTENT REPORTED-LEAD LOG] before reporting. Do not repeat previously reported posts or leads. Group multiple posts from the same author into one opportunity. Record post IDs, handles, source times, and report times for leads included in a successfully delivered report. If the log cannot be read or written, disclose that and do not claim deduplication worked. Keep the report short, with a maximum of [10] new opportunities, highest intent first. A small or empty report is acceptable; do not pad it. End with “What needs my attention?” and highlight up to three most promising opportunities, or state that no qualifying new opportunities were found. Report incomplete searches or failed runs separately from zero matches. Do not reply, follow, DM, or otherwise contact anyone. I will review each opportunity and decide whether to respond.
How to run it
- 01
Define your product, customer profile, exclusions, and search phrases for all five signal types; choose a small report limit.
- 02
Run one manual search and check each post, timestamp, fit reason, and reply draft against the original context.
- 03
Set up a persistent reported-lead log and confirm that a repeated sample run does not resend the same posts or people.
- 04
Configure the four-hour schedule and report destination in your own environment. Review the report before any response and inspect failures or empty runs.
Review before you act
Open each original post before acting. Check that the request is current, the product actually fits, and the draft offers useful help without unsupported claims or disguised affiliation. A relevance score does not prove buying intent. All replies, follows, and DMs remain manual decisions.
Limitations & success checks
- The excerpt’s author, original URL, and publication date are unverified. It describes an experiment and a suggested prompt, not measured lead or revenue results.
- X search coverage and account access can be incomplete. Freshness depends on checking post timestamps, not when a search engine indexed the page.
- A prompt alone does not create a scheduler or durable memory. Reliable deduplication requires a persistent log across runs.
- Intent scoring and bot detection are fallible. A complaint or workflow description is not necessarily a purchase request.
Measure: New opportunities accepted after review, duplicate and stale-post rates, unsupported fit claims, and conversations that progress after a human chooses to respond.
Sources & attribution
The sources below support the reported example or a tool capability. They do not validate our prompt or prove the outcome in your environment.
The supplied excerpt proposes four-hour X searches across five buying-signal types, a 24-hour freshness window, relevance scoring, deduplication, helpful reply drafts, and human review before any contact. The original author, URL, and date are not verified. This guide adapts the prompt and adds implementation checks; no conversion or revenue outcome has been established.
Editorial review: 2026-10-13. How we label evidence.