How to Find Buying-Intent Leads with an AI Agent Skill

People announce their buying intent in public every day: "looking for a tool that tracks brand mentions", "is anyone else fed up with [vendor]'s new pricing", "how do you all handle this?" A lead scan is the discipline of finding those posts while the intent is still live, and it is a job an AI agent can now run end to end, from query building through a prioritized report.

This tutorial sets that up with lead-gen-scan, an open-source agent skill from our team at Xpoz. The skill encodes the scan workflow; the data comes from the Xpoz index of Twitter/X, Instagram, Reddit, and TikTok (billions of posts, free tier available). Finding people with buying intent is one of the most common workflows Xpoz users run, and the skill packages what the heaviest users converged on.

What One Scan Produces

A scan returns a report, not a firehose. The top section, "This Week's Picks", holds 3-5 leads sized to what a person can act on today. Below it, leads land in three buckets: P1, act now (a clear ask, strong fit, thread still live), P2, worth engaging (real intent, weaker on one axis), and P3, watch (signal without an ask yet).

Every lead carries the quoted ask, a one-line reason it qualifies, and a suggested angle for a useful reply. Two more sections make the scan compound: Named Prospects lists individual users whose posts show strong conversion signals, and Demand Signals collects the recurring pains and phrasings across all candidates, which double as content and product research.

The Three Search Motions

The skill runs three distinct searches every scan, because each finds leads the others miss:

1. Product relevance. People asking for what the product does, phrased the way buyers phrase it: "looking for a tool that", "any recommendations for", "worth paying for", combined with the category terms.
2. Competitor disappointment. People frustrated with the alternatives: "[competitor] alternative", "[competitor] pricing increase", "[competitor] vs". "Alternatives" here includes non-tools people use instead, like manual spreadsheets or an official platform API.
3. The product's own name. People already comparing it ("[product] vs", "[product] worth it") are the warmest leads of all, and neither motion above finds them.

The skill also encodes the unglamorous query craft: keeping queries under the 250-character cap, preferring short quoted phrases OR-joined together over long exact phrases, and anchoring brand names that double as common English words ("mention", "plausible") with a category term so the results don't drown in false positives.

Different Platforms, Different Lead Shapes

Not every platform produces the same kind of lead, and the skill classifies accordingly. On Reddit, the lead is usually a thread: a live ask where one disclosed, genuinely useful comment reaches the asker plus everyone who finds the thread later. On X, Instagram, and TikTok, leads split into likely converters (individual users with a quantified need or budget pain, tracked as named prospects) and high-engagement comment spots (a viral complaint about a competitor, where a reply reaches a large relevant audience).

One shape gets separated on purpose: an existing customer publicly struggling is a retention save, not a sales lead. The skill reports it with a support-shaped angle and never inflates it into a P1.

Setup (Two Commands Plus Data Access)

Install the skill into your agent's skills directory. For Claude Code:

git clone https://github.com/XPOZpublic/xpoz-agent-skills.git
cp -r xpoz-agent-skills/skills/lead-gen-scan ~/.claude/skills/

Then connect the data layer, either the remote MCP server (OAuth sign-in, no keys):

claude mcp add --transport http --scope user xpoz https://mcp.xpoz.ai/mcp

or the Python SDK with a free API key from xpoz.ai/get-token (up to 75,000 results free, no credit card). The skill works identically through either path, and through any SKILL.md-compatible agent, not just Claude Code. The full install pattern is covered in How to Add Social Media Skills to Your AI Agent.

Run It

The skill activates on plain requests. Give it the product, the competitors, and optionally a window:

"Find this week's leads for my social listening SaaS. Competitors:
Brand24, Mention. People also just use Google Alerts."

"Who's complaining about [competitor]'s pricing right now?"

"Scan Reddit and X for people asking for CRM recommendations
in the last 7 days"

The agent builds the query book, searches the platforms (Reddit posts and comments carry most of the signal, and the skill spends its call budget there first), chases replies up to their parent threads, qualifies, and writes the report. Vendors promoting competing tools, keyword hits in unrelated contexts, and spam are hard disqualifiers that never reach you.

The Ground Rules Are the Point

The skill never engages anyone: no posts, no comments, no DMs, under any instruction. It suggests angles, never full drafts, and every suggested engagement assumes in-text disclosure of affiliation. That constraint is what keeps agent-assisted lead generation on the right side of both platform norms and buyer trust: the research scales, the human relationship doesn't pretend to.

FAQ

Can an AI agent find sales leads on social media?

Yes, for the discovery half of the job. An agent with a lead-scan skill and social data access can find people actively asking for what a product does, qualify the real asks, and rank them by intent, fit, and freshness. Engagement stays human.

What are buying-intent keywords?

Phrases people use close to a purchase decision: asking phrasings ("looking for a tool that", "any recommendations for"), budgeting phrasings ("worth paying for"), and switching phrasings ("[competitor] alternative", "moving away from"). Searching these families surfaces buyers instead of bystanders.

Will the agent post, comment, or DM people?

No. The skill finds and reports; humans do all engagement. It suggests an angle for a useful reply but never drafts full responses or contacts anyone.

How fresh are the leads a scan finds?

The default window is the last 7 days, with the last 72 hours weighted heaviest. Anything older than about three weeks is backlog regardless of quality.

Next Steps

• The manual version of this loop, with full phrase taxonomies per platform, is in How to Find Leads on Social Media by Monitoring Buying-Intent Keywords.
• Choosing a monitoring stack more broadly? See Best Social Listening Tools for Lead Generation.
• For a recurring weekly loop with lead dedup and a self-tuning query book, the open-source lead-gen-agent builds on this skill.