How to use AI to sort and draft LinkedIn replies

Learn how to use AI to classify LinkedIn lead intent and draft responses using a sales playbook to save time and keep a human touch in high-value chats.

Get started

Short answer

Use AI to sort LinkedIn replies into intent buckets, then draft responses from a short sales playbook. Keep a human in the loop for high-intent conversations, and review the categories and prompts until the sorting is accurate.

Define your intent categories

AI cannot guess what a good conversation looks like if you do not define the categories first. If you tell it to reply to everything, it defaults to a generic tone that people ignore.

Group every incoming message into a small set of buckets. Each bucket gets its own rule for what the AI should do. A person asking for a price needs a different answer than a recruiter offering a job.

Keep the list short. Four or five categories are enough. If you have twelve, the AI will split hairs and you will not know which bucket to trust. Start from your sent folder. Open twenty replies you handled in the last month and label each one. The labels you use without thinking are your real categories.

Write one sentence definition for each category. 'Interested' is not enough. 'Interested means asks about price, a demo, a trial, or the next step' is a rule the AI can follow.

Interested

Intent
Asks about pricing, a demo, a trial, or the next step
What it looks like
Stop drafting and notify you, or send a booking reply if you use a scheduling link

Question

Intent
Asks how a feature works, asks about fit, or asks for a detail
What it looks like
Answer from the playbook and ask one small next step

Not interested

Intent
Says no budget, no need, bad timing, or already has a vendor
What it looks like
Send a short close and mark for a later follow-up

Out of scope

Intent
Recruiter, student, vendor pitch, or wrong market
What it looks like
Archive and do not reply

Referral

Intent
Points you to a colleague or says 'talk to my boss'
What it looks like
Ask for an intro and log the new name

Create a response playbook

A playbook is a set of rules for how the AI speaks. It is not a full script. Without one, AI writes long paragraphs with fake politeness and too many adjectives.

Write rules you would give a new sales rep. Answer the question first. Use one question per message. Keep messages under 80 words. Do not list features unless asked. Stop when the lead asks for a call.

Add a short section for common objections: price, timing, security, and 'we already use something'. Give the AI one plain answer for each and a next step.

Keep the playbook in a shared doc. When a rep writes a good reply, add it as an example. When a draft sounds wrong, add a line that says what not to do. Tone rules should be concrete. 'Friendly' is vague. 'Write like a busy rep: short sentences, no exclamation marks, one question at the end' is usable.

Weak draft

  • I hope this message finds you well. I would be honored to connect and discuss how our comprehensive platform can support your growth.
  • Our platform offers a suite of tools including analytics, integrations, and a modern dashboard to improve your workflow.

Strong draft

  • Thanks for asking, Sarah. It syncs your calendar and drafts replies for you. Do you have 15 minutes on Thursday to see it?
  • We do not have a native integration for that yet, but we can export the data to CSV. Would that work for your setup?

Set the AI reply logic

Do not automate every message. The more a lead is worth, the more human the interaction needs to be. Use automation for sorting and early qualification.

Set a workflow where the AI handles the noise and drafts the medium-value questions. It stops when a lead shows high intent. If you use a tool with reply modes, choose hand-off for enterprise leads and AI-first for low-ticket leads. The same rule can change by campaign.

Log every hand-off. If the same question causes a hand-off three times, add it to the playbook. The goal is to shrink the review queue, not to make it permanent.

  1. 1

    Sort every reply first

    Classify the message before drafting anything. If the intent is unclear, put it in a review queue instead of guessing.

  2. 2

    Draft low-risk answers

    Let the AI answer questions about features, fit, and process from the playbook. Keep pricing, security, and legal answers for a human.

  3. 3

    Stop at buying signals

    When a lead asks for a demo, a proposal, or a call, stop the AI. Notify yourself and take over the thread.

  4. 4

    Review the queue daily

    Spend ten minutes on misclassified replies. Add the best examples to your category definitions and prompt.

Review and refine the drafts

Even with a good playbook, AI can invent features or use words that sound unnatural. A quick manual review of drafts prevents mistakes that can kill a deal.

Check that the AI references the lead's profile or their specific comment. If the reply is too generic, it will be ignored. Keep a list of bad drafts and add the pattern to your playbook.

Read the first line of every draft. If it starts with 'I hope' or 'Thanks for reaching out', cut it. The lead knows why you are there. Check names and company details too. AI will sometimes swap them or use the wrong case.

Example

  1. Prospect

    I saw your post about lead scoring. Does this work for agencies with more than 50 people?

  2. Suggested reply

    Yes, it does. It works for agencies with more than 50 people. Would you be open to a quick chat?

  3. You

    Yes. The setup is the same at 50 people, you just add more lead groups. I can show you how it works on Thursday at 10 if that works.

If a draft contains a word you would not say out loud to a client, rewrite the sentence. Plain words beat polished ones.

Write the classification prompt

A prompt is not a magic spell. It needs examples. Put two or three real replies under each category. Remove names and company details before you paste them.

Ask for confidence and a reason. That gives you a way to spot weak sorting without reading every message twice. If confidence is low, send the reply to review.

Keep the prompt short enough to read in one pass. If it runs to two pages, the model will miss rules. Put the categories first, then examples, then the output format.

Paste this job

You are sorting LinkedIn replies for a B2B software company.
Read the reply and the original message.
Choose one intent: Interested, Question, Not interested, Out of scope, Referral.
Rules:
Interested means asks about price, demo, trial, or next step.
Question means asks for a detail about the product or fit.
Not interested means says no, no budget, no need, or has a vendor.
Out of scope means recruiter, student, vendor pitch, or wrong market.
Referral means points to another person.
Return JSON with intent, confidence from 0 to 100, reason in one sentence, and suggested next action.
If confidence is below 70, set intent to Review.
Do not write a reply.

Test with real replies before you turn it on

Before you let AI reply on its own, run a test batch. Pick replies you already handled. Hide your answer. See what the AI does with the same message.

Check the intent, the answer, and the tone. If it misses more than a couple in a row, fix the categories or add examples. Do not blame the model until you have read your own prompt.

Run the test again after any change. A prompt that works on five messages can fail on the next five because the language changes. Use a simple scorecard: intent correct, answer correct, tone correct. If two of three fail, rewrite the prompt. If one fails, add an example.

Doing this with Kinlead

Kinlead puts replies from every campaign in one inbox, classified by intent, with interested replies first. You get an email when someone replies, so you do not have to watch the inbox.

Each campaign chooses how replies are handled: hand the conversation to you at the first reply, let the AI reply until there is real interest, or let the AI reply until a meeting is booked using your scheduling link. Messages are written from the campaign goal, tone, and an optional sales playbook. To review every reply yourself, use hand-off mode. Replies land in the unified inbox.

Frequently asked questions

Related

Get started with Kinlead

Prefer markdown? How to use AI to sort and draft LinkedIn replies.md