How to Turn LinkedIn Post Engagement Into Qualified Leads with an AI Agent

You already have warm prospects hiding in plain sight.

Every time the right person comments on or reacts to a relevant LinkedIn post, they tell you something useful: this topic has their attention right now. That is a much better reason to start a conversation than “they match a filter in a database.”

The mistake is treating every engager as a lead and blasting the same message to all of them. Don’t do that. A reaction is a signal, not permission to pitch.

Here’s the workflow I’d use with you instead: choose a post that attracts your real market, collect the people who engaged, qualify them against a clear ICP, review the conversation context, and only then prepare a small batch of messages worth sending.

By the end, you’ll have:

• a clean list of commenters and reactors from one LinkedIn post;
• a prioritized shortlist with a reason for each score;
• personalized connection and follow-up drafts;
• a clear approval checkpoint before anything is sent;
• verified SalesTouch states for every action that runs.

Friendly advice: run this manually once before you automate it. The first pass teaches you which signals actually predict a good conversation in your market. That insight is worth more than a large unreviewed list.

Before you start: decide what a good lead looks like

This is the part most people skip, then they blame the extraction when the list is bad.

Your agent needs a useful definition of “qualified.” Job titles alone aren’t enough. Give it enough context to understand who you help, what problem you solve, and which signals make someone worth your time.

Write down five things:

1. Who you help. Industry, company stage, team size, geography, role or buying responsibility.
2. The painful problem. The situation that makes your offer relevant now.
3. Positive signals. A thoughtful comment, a question about the problem, the right role, an active project, or recent posts about the topic.
4. Disqualifiers. Competitors, students, job seekers, service providers selling the same thing, irrelevant geographies, or companies that cannot buy.
5. The next conversation you want. A peer exchange, a diagnostic question, a resource offer, or a demo — not all four at once.

If you already have an offer document or ICP note, point your agent to it. If not, use this prompt and keep the answer short.

Help me define the qualification rules for a LinkedIn post-engager workflow.

My offer: [DESCRIBE THE OFFER]
Who it helps: [DESCRIBE THE IDEAL CUSTOMER]
The problem it solves: [DESCRIBE THE PAINFUL PROBLEM]
Geography or market constraints: [ADD CONSTRAINTS]

Return:
1. Five positive qualification signals.
2. Five clear disqualifiers.
3. A simple 0–100 scorecard split into fit, intent, and conversation relevance.
4. The minimum score a prospect should reach before I review a message.

Keep the scorecard practical. Do not invent missing information and do not suggest outreach yet.

Checkpoint: you should be able to explain in one sentence why someone belongs on the shortlist. If the answer is only “they liked a post,” your qualification rules are still too loose.

Step 1: choose a post with the right audience

Start with relevance, not vanity metrics.

A post with 40 reactions from your exact market is more useful than a viral post with 4,000 reactions from everyone. The source post determines the quality of everything downstream.

Good source posts usually have three characteristics:

• the topic is directly connected to a problem your offer solves;
• the comments contain opinions, questions, examples, or objections rather than applause only;
• the audience includes people who can experience, influence, or buy around that problem.

You can use your own post, a customer’s post, or a respected operator’s post. Be more careful with competitor posts: the context can be useful, but your message must never pretend the person engaged with you.

Give your agent two or three candidate URLs and ask it to compare them before you extract anything.

Compare these LinkedIn posts as potential sources for qualified prospects:

[POST URL 1]
[POST URL 2]
[POST URL 3]

Use my qualification rules from this conversation. For each post, assess:
- topic relevance to my offer;
- likely audience fit;
- quality of the visible discussion;
- whether a future message could reference the engagement honestly;
- the main risk of using this post.

Recommend one post and explain the decision in no more than five bullets. Do not extract people or contact anyone yet.

Checkpoint: choose one post and write down the honest reason it gives you to talk to its audience. You’ll reuse that reason when reviewing messages.

Step 2: give your agent live LinkedIn access

Up to this point, your agent only needed the context you gave it. Now it needs live access to LinkedIn.

This is where SalesTouch comes in. It connects your AI agent to LinkedIn so it can inspect the post, extract the people who engaged, read the profiles you select, check existing relationship context, and run only the actions you approve.

If SalesTouch is already connected, ask your agent to list the available LinkedIn accounts and continue.

If it isn’t, create your SalesTouch account and follow the Getting Started guide. Come back here once your agent can see the SalesTouch tools.

Check whether SalesTouch is connected and ready for this workflow.

List the available LinkedIn accounts and show the current SalesTouch status, including queue and action limits. Do not run any outreach action.

You want three confirmations before continuing:

• the correct LinkedIn account is connected;
• the agent can access the SalesTouch tools;
• there is no account or queue warning you need to handle first.

Don’t rush this: a connected tool is not the same as a blank cheque. The agent should still stop before every message or invitation until you approve the exact copy.

Step 3: extract commenters and reactors separately

Commenters and reactors are not the same signal.

A specific comment can tell you what the person believes, wants, or struggles with. A reaction only tells you the post caught their attention. Keep those sources separate so the agent doesn’t overstate the context later.

Start by resolving the post. Then extract comments and reactions as two read-only datasets. Keep the first preview small; large outputs are easier to review through stored result pages than inside one huge chat response.

Use SalesTouch to resolve this LinkedIn post and inspect its author and content:

[POST URL]

Then, as separate read-only operations:
1. Extract the people who commented on the post.
2. Extract the people who reacted to the post.
3. Preview no more than 20 people from each result.

Return the result ID for each extraction, the total count, and a compact table with name, LinkedIn URL, engagement type, and comment text when available.

Do not look up every profile yet. Do not send invitations or messages.

Now clean the raw audience before spending time on individual profiles:

• remove the post author;
• deduplicate people who both commented and reacted;
• remove obvious company pages, irrelevant profiles, and known competitors;
• label commenters separately from reaction-only prospects;
• preserve the exact comment text when available.

Clean the two SalesTouch extraction previews without contacting anyone.

Rules:
- remove the post author;
- deduplicate by LinkedIn profile URL;
- keep “commenter” as the stronger engagement source when a person appears in both lists;
- preserve exact comment text;
- flag obvious competitors or clearly irrelevant profiles instead of deleting them silently;
- do not infer job details that are not present.

Return three groups: likely relevant, uncertain, and excluded. Explain every exclusion in a short reason.

Checkpoint: you now have a manageable candidate pool, not a send list. Nothing in this step justifies outreach on its own.

Step 4: qualify for fit, intent, and conversation relevance

Now you spend the expensive attention where it matters.

Ask the agent to inspect profiles only for the likely relevant and uncertain candidates. If the extraction is large, begin with 20–30 people. You’re looking for evidence, not trying to enrich the entire internet.

Score each person across three dimensions:

• Fit: do their role, company and market resemble the customer you can help?
• Intent: did their engagement reveal a question, problem, project or strong interest?
• Conversation relevance: can you start a useful conversation without manufacturing familiarity?

Commenters often score higher on intent because they gave you language to work with. Reaction-only prospects can still be excellent fits, but the opening should be softer.

Qualify the likely relevant and uncertain candidates against the scorecard we defined earlier.

For each candidate:
1. Use SalesTouch to look up the LinkedIn profile.
2. Use only visible, verifiable information.
3. Score fit, intent, and conversation relevance.
4. Quote the exact evidence behind each score.
5. Mark missing information as unknown instead of guessing.

Return a ranked table with:
- name and LinkedIn URL;
- engagement source;
- fit score;
- intent score;
- conversation relevance score;
- total score;
- evidence;
- recommended next step: review, nurture, or skip.

Stop after the first 25 candidates. Do not draft or send messages yet.

Review the top ten yourself. The agent can organize evidence quickly, but you still own the judgment.

Ask these questions:

• Would I still want to know this person if they never bought?
• Can I explain why the conversation is relevant to them now?
• Is the source post close enough to my offer to make the transition natural?
• Am I using information they chose to make public, without getting creepy?

A useful rule: if you need a paragraph to justify why someone is relevant, they probably aren’t. Good qualification usually survives a one-sentence explanation.

Checkpoint: keep a shortlist of roughly 5–15 people. Small is a feature here. It gives you room to write messages that sound like you actually noticed the person.

Step 5: check relationship context before writing

Before drafting anything, check whether you already know the person.

An existing connection, an old unanswered message, a pending invitation, or an active thread completely changes the right next step. Sending a fresh invitation to someone already in your inbox is an easy way to look careless.

Use SalesTouch to check lead status and conversation history for the shortlist.

For each shortlisted prospect, use SalesTouch to check:
- current LinkedIn relationship status;
- pending or accepted invitation state;
- queued follow-up state;
- whether a conversation already exists;
- the recent message history when a conversation exists.

Return one recommended path per person:
1. Existing conversation — continue the thread.
2. Existing connection, no conversation — draft a direct message.
3. Not connected — draft an invitation and a follow-up after acceptance.
4. Pending or recently contacted — do nothing now.

Do not draft or send anything in this step. Flag any ambiguous state for my review.

This step protects you from duplicate outreach and gives the eventual message a truthful starting point.

Step 6: draft messages that continue the context

Now you can write.

The message should feel like the next sentence in a conversation, not a campaign that happened to scrape a name.

For commenters, reference the idea they shared — not the fact that you scraped the comment. For reactors, mention the topic softly and lead with a relevant question or observation. Never claim they have a problem because they clicked a reaction button.

A good first message usually does three things:

1. establishes the honest shared context;
2. shows why the conversation could be useful to them;
3. makes the next step easy and low-pressure.

Avoid links, calendar requests and product explanations in the opening unless the person explicitly asked for them.

Draft outreach for the approved shortlist. Do not send anything.

Use the qualification evidence, exact engagement source, and relationship context already collected.

Writing rules:
- sound like an experienced peer, not a sales automation;
- keep the invitation note under 250 characters;
- keep the first follow-up under 500 characters;
- never say “I saw you liked a post”;
- never invent familiarity or a pain point;
- for commenters, reference the substance of their comment;
- for reaction-only prospects, reference the topic without overstating intent;
- ask at most one simple question;
- no calendar link, pitch deck, or product paragraph;
- make every draft meaningfully different.

For each person, return:
1. Why they are worth contacting.
2. The exact context used.
3. Invitation note, if needed.
4. Follow-up message.
5. The risk or uncertainty I should review.

Wait for my explicit approval before any LinkedIn action.

Read every message out loud. If it sounds strange without the prospect’s first name, it’s probably a template wearing a personalized hat.

Edit until you can answer yes to all four:

• Is the context true?
• Is the message useful even if they never buy?
• Would I be comfortable receiving it?
• Does it sound like me?

Checkpoint: approve the exact copy person by person, or approve a clearly identified batch. “Looks good” should never be interpreted as permission to send unrelated future drafts.

Step 7: run a small approved batch

Start with five people.

That is enough to test the workflow and small enough to inspect every result. You can expand later once the qualification and copy prove themselves.

For existing connections, the agent can send the approved direct message. For people who aren’t connected, SalesTouch can send the invitation and save one approved follow-up to run after acceptance.

Execute only the five outreach actions I explicitly approved in the previous step.

Rules:
- reuse the exact approved copy with no rewriting;
- use the selected LinkedIn account;
- for an existing connection, send the approved direct message;
- for a non-connection, send the approved invitation and save the approved follow-up for after acceptance;
- if SalesTouch requires queueing, show me the reason and use queueing only when allowed;
- if relationship state changed, skip the action and report it;
- do not replace a failed prospect with another person;
- do not exceed five prospects.

Return a result table with person, action, exact copy, SalesTouch state, and action or queue identifier. Treat only the returned SalesTouch state as truth.

Do not accept “done” as the report. You want the state of each action:

• sent: the action completed;
• queued: SalesTouch accepted it for later execution;
• saved follow-up: the message is waiting for invitation acceptance;
• skipped: the agent correctly avoided an unsafe or duplicate action;
• failed: the provider rejected the action and nothing should be inferred.

Step 8: verify the workflow and learn from it

The first batch is not only outreach. It is calibration data.

Use lead_status to verify invitation and follow-up state. Then keep a lightweight record of what happened:

• which source post produced the lead;
• commenter or reactor;
• qualification score;
• message angle;
• sent or queued state;
• accepted, replied, skipped, or no response;
• what you learned.

Verify the five prospects from this playbook with SalesTouch lead status.

For each person, report:
- current relationship state;
- invitation state;
- saved or queued follow-up state;
- message state;
- any provider error or action identifier;
- the next safe action and the earliest sensible time to take it.

Then summarize what this batch taught us about the source post, qualification scorecard, and message angles. Do not send anything.

After a few days, compare the outcomes. You’re looking for patterns, not declaring victory after one reply.

Maybe commenters with specific questions outperform all reactors. Maybe a certain job function accepts but never replies. Maybe your best conversations come from smaller posts written by practitioners rather than viral creators.

That is the real advantage of running the workflow carefully: you don’t just collect names. You build a better model of where your market shows intent and how they prefer to start conversations.

What to automate after the first manual run

Once you trust the workflow, automate the boring parts — not the judgment you haven’t earned yet.

Good candidates for automation:

• extracting new engagers from approved posts;
• deduplication and source labeling;
• applying a proven scorecard;
• preparing a review queue;
• checking relationship state;
• drafting messages for approval;
• verifying sent and queued states.

Keep these approval-first:

• changing qualification criteria;
• choosing a new source-post category;
• contacting borderline prospects;
• approving message angles;
• increasing batch size;
• sending invitations and messages.

The goal isn’t to remove you from the workflow. It’s to remove the repetitive work so your attention lands on the decisions that change the result.

The one-shot prompt

If you’d rather hand the whole strategy to your agent, use the Copy for your AI agent button at the top or bottom of this page. It copies this playbook as clean Markdown with the safety rules included.

Give the agent your offer context, your ICP, and one or more candidate LinkedIn post URLs. It will guide you through the rest without pretending missing information is known.

Run the first batch small. Read the messages. Watch the returned states. Then improve the system with evidence from your own market.