How to Automate LinkedIn Messages with AI (Prompts and Examples)

To automate LinkedIn messages with AI, connect your LinkedIn account to an outreach tool, build a short sequence (invitation, a message once it's accepted, one or two follow-ups) and replace fixed templates with an AI instruction that writes each message from the prospect's profile and your offer. The tool sends at a human pace and stops as soon as someone replies.

The AI model is rarely the problem. What separates messages that get answers from messages that read like a bot is the input you give the AI and the rules you set. Below: the sequence to build, prompts you can copy, worked examples and the guardrails we use.

Three ways to write LinkedIn messages with AI

LinkedIn's Message Assist is still labeled a public beta and only covers InMail in Sales Navigator. Copy-pasting into a chatbot works for 10 high-value accounts, not 300. Past that, you need an outreach tool, and a choice about risk: LinkedIn's User Agreement prohibits "bots or other unauthorized automated methods" to send messages, so any automation tool carries some account risk. We cover what that means, tool by tool, in our comparison of LinkedIn automation tools.

Step 1: Build the sequence before you write a word

AI writes better when each message has one job. A sequence that works for most B2B offers:

The sequence should stop the moment a lead replies, on LinkedIn or by email.

About invitation notes: LinkedIn's help center says free members can add a personalized note to only three connection requests per month, with 200 characters max; Premium members can personalize all of them. On a free account, send invitations without a note and put the personalization in message 1. If you want a note on every invitation, you need Premium.

Step 2: Give the AI the right inputs

An AI can only personalize with what you feed it. Two blocks of context matter.

About you (written once per campaign):

• Your offer in one sentence, and who it's for.
• One proof point you can back up (a client, a number from your own data).
• The call to action: a question, a 15-minute call, a resource.
• Tone: how you'd write to a peer, words you never use.

About each prospect (pulled automatically):

For company-level facts, use a research step rather than trusting the model. In useArtemis, AI enrichment adds AI-researched columns to a lead list (for example "Is this company hiring sales reps right now?"), and the message prompt can use those columns.

Step 3: Prompts you can copy

The examples below use a fictional prospect: Maya Chen, Head of Sales at Northwind Freight, 40 people, who posted last week about onboarding three new SDRs. The sender sells sales-onboarding software.

Prompt for the invitation note

Write a LinkedIn connection note to {first_name}, {job_title} at {company}.
Context about them: {headline}, {recent_post_summary}.
Rules:
- Max 180 characters (LinkedIn's limit is 200).
- Mention one specific, true detail from the context. If there is none, mention their role.
- No pitch, no link, no question about meetings.
- Sound like a peer, not a salesperson. No "I came across your profile".
- Output only the note.

Example output: "Hi Maya, saw your post on bringing three new SDRs up to speed at Northwind. I work on sales onboarding too, would be glad to connect."

Prompt for the first message after acceptance

You write the first LinkedIn message after {first_name} accepted my invitation.
About me: {my_offer_one_sentence}. Proof: {proof_point}.
About them: {job_title} at {company} ({company_size} people). {recent_post_summary}. {ai_enrichment_column}.
Rules:
- 3 short sentences, under 350 characters.
- Sentence 1: one observation tied to their situation, using only the facts above.
- Sentence 2: why it matters for someone in their role (no feature list).
- Sentence 3: one open question they can answer in a line.
- No link, no calendar request, no "hope you're well".
- Write in {language}. Output only the message.

Example output: "Thanks for connecting, Maya. Onboarding three SDRs at once usually means managers lose a few weeks to shadowing and call reviews. How are you getting new reps to their first booked meeting today?"

Prompt for a follow-up

Write a follow-up to my previous LinkedIn message (below), sent 4 days ago with no reply.
Previous message: {previous_message}
Rules:
- Max 2 sentences. Don't repeat the previous message or apologize for following up.
- Add one new, useful angle: a short tip, a pattern you've seen, or a resource described in words (no link).
- End with a yes/no question.

Example output: "One thing that helps teams in your spot: recording the first 20 calls of your best rep and turning them into a review checklist. Would it be useful if I shared the one we use?"

Prompt for replies

A prospect replied to my LinkedIn outreach. Draft my answer.
Conversation so far: {thread}
About me: {my_offer_one_sentence}. Pricing and availability: {facts_you_allow}.
Rules:
- Answer their question first, in 1-2 sentences, using only the facts provided.
- If they ask something you have no fact for, say you'll check and come back.
- If they're interested, propose two time slots or ask for their preferred time.
- If they say no, thank them and stop. Never insist twice.

Automatic AI replies are useful for simple cases (a "yes, tell me more", a request for times). Keep a human on anything involving price negotiation, objections or legal questions. In useArtemis, automatic AI replies are available on the Scale plan and up; on every plan, replies land in a unified LinkedIn + email inbox.

Step 4: Guardrails that stop AI messages from sounding like AI

Add these rules to every prompt:

• Use only the facts provided. Models happily invent a "recent funding round" or a mutual connection. Tell it explicitly not to.
• Give it a fallback. "If there's no recent post, use their role and company size" avoids messages built on nothing.
• Set hard limits. Character count, one question, no links in the first two messages.
• Ban the tells. "I hope this message finds you well", "I came across your impressive profile", "I'd love to pick your brain", exclamation marks, emojis.
• Match the prospect's language. Write in French to someone whose profile is in French.
• No fake familiarity. Don't pretend you read a post you didn't; don't invent shared interests.

Step 5: Test on 20 prospects before scaling

1. Generate messages for 20 leads from different sub-segments (job titles, company sizes).
2. Read every one. If more than two need fixing, change the prompt, not the messages.
3. Run two prompt versions on similar segments for a week or two.
4. Track three numbers per version: acceptance rate, reply rate and positive reply rate. A message that gets replies but no meetings is too vague about what you do.
5. Keep the winner, then test one change at a time.

Our LinkedIn connection message templates are a good baseline to compare your AI version against.

Step 6: Send at a pace LinkedIn tolerates

AI doesn't change LinkedIn's limits. LinkedIn doesn't publish its weekly invitation cap, and when you hit it the restriction typically lasts a week. Most users run into it around 100 invitations per week; see how many LinkedIn invites you can send per week. Spread actions over working hours with random delays, warm up new accounts, and never run two automation tools on the same profile.

How to set this up in useArtemis

1. Build your list. Search the Discover lead database, export a LinkedIn or Sales Navigator search with the LinkedIn scraper, or import a CSV.
2. Add context. Find and verify emails if you want an email fallback, and add AI enrichment columns for company-level facts.
3. Create a campaign mixing LinkedIn steps (profile visits, invitations, messages, likes) and email steps, with a stop when the lead replies.
4. Describe your offer. The AI writes each message from your pitch and the prospect's profile instead of a {firstName} template.
5. Connect your LinkedIn account(s) through the secure connection and launch. Sending follows safe limits and a human-like pace.
6. Handle replies from the unified inbox, with automatic AI replies on Scale and Agency.

Plans start at $69 per month (Starter, 2 LinkedIn accounts) as of October 2026, with a 7-day free trial on Starter (card required, one LinkedIn account during the trial). Details on the pricing page.

Mistakes to avoid

• Pitching in the invitation note. Its only job is to get accepted.
• Personalization that isn't. "I see you work in sales" in front of a sales leader is worse than no personalization.
• Long first messages. If it needs scrolling on a phone, it's too long.
• Letting AI answer objections alone. Draft, then review.
• Treating AI as a volume multiplier. Better messages to a tighter list beat more messages to everyone. Our guide on how to automate your LinkedIn outreach covers targeting and the rest of the workflow.

FAQ

Can you automate messages on LinkedIn?

Yes, with a third-party outreach tool: it sends invitations, messages and follow-ups for you on a schedule and stops when someone replies. With an AI writing step, each message is written from the prospect's profile instead of a template.

Is it against LinkedIn's rules to send automated messages?

Yes. LinkedIn's User Agreement prohibits using bots or unauthorized automated methods to send messages or add contacts, and accounts that do can be restricted. Cloud tools that keep volumes low and human-paced reduce the risk; none removes it.

Can ChatGPT write LinkedIn messages?

Yes, and the prompts above work in ChatGPT, Claude or Gemini. The limit is volume: you copy each profile in and each message out. To send AI-written messages to hundreds of prospects, use an outreach tool that runs the prompt for every lead automatically.

Does LinkedIn have a built-in AI message writer?

Sales Navigator offers Message Assist, labeled as a public beta, which drafts first-touch InMails using lead and account data. It writes one draft at a time that you review and send yourself; it doesn't automate a sequence.

Should AI reply to prospects automatically?

For simple answers (confirming interest, sharing times, answering a factual question you've provided), yes. For pricing negotiations, objections or anything sensitive, let the AI draft and a human send. For more on handling replies at scale, see our LinkedIn inbox management tips.