How to write a sales playbook for AI LinkedIn outreach
Document your sales logic and tone so AI agents send LinkedIn messages that sound human and avoid corporate jargon. Practical steps for B2B teams.
Get startedShort answer
To write a sales playbook for AI, document the logic and tonal markers of your best human conversations instead of using generic templates. Give the AI clear guardrails and a specific goal to prevent it from using corporate jargon or pushing for a meeting too early.
Define the conversation goal
Most people tell an AI to get a demo. That's why AI outreach feels like a pitch. When the goal is too big, the AI tries to shortcut the relationship by pushing for a meeting in the first message. Prospects read that as automation.
Instead, define a binary outcome for each step. A binary outcome is a yes or no answer to a specific question. This keeps the conversation focused on the prospect's problem rather than your calendar.
If you want a meeting eventually, make your first goal confirming a pain point. Once the prospect agrees that a problem exists, the transition to a call feels like a solution. The goal also tells the AI what to do when they say yes or no. If they confirm the problem, the AI asks for a short call or shares a relevant example. If they say no, the AI asks a different qualifying question or closes the loop.
Generic goals
- Get the prospect to book a demo
- Introduce the product and its features
- Start a conversation about our services
Binary goals
- Confirm they are currently using a manual spreadsheet for X
- Get them to agree that Y is a bottleneck in their current workflow
- Confirm they are the person responsible for Z
Map the message sequence
A natural conversation follows a logical progression. You don't start with a proposal; you start with a reason for reaching out. Document this flow as a series of stages so the AI knows which phase of the relationship it's in.
Keep the friction low. Every message should ask for a very small amount of effort from the prospect. Asking for a 30 minute call in message one is high friction. Asking if they have a specific problem is low friction.
Include a closing loop for people who stop responding. The note gives you a clean exit and gives them permission to ignore you. It also moves the lead out of your active pipeline without pretending the conversation is still alive.
Write each stage as if you were handing it to a new SDR. The AI needs to know what the message is for, what it should ask, and what it should never do at that stage. A vague sequence turns into a pile of similar messages with no clear reason for each one.
- 1
The trigger message
Acknowledge a specific event like a new job or a recent post. Ask a low-friction question about their current process.
- 2
The value bridge
Share a one-sentence example of how you solved that specific problem for a similar company. Don't list features.
- 3
The soft ask
Ask for a short call or a specific piece of feedback on your approach to the problem.
- 4
The loop closer
Send a brief note stating you'll stop reaching out for now so you don't clutter their inbox.
Set tonal guardrails
AI defaults to a corporate, overly polite style that feels fake. To stop this, give the AI a list of forbidden words and structural rules. This forces the AI to write like a person typing a quick message.
Human LinkedIn messages are usually short and slightly imperfect. They use contractions and avoid complex sentences. If a message looks like a polished marketing email, the prospect will ignore it.
Set a hard limit on length. Long messages on LinkedIn are perceived as spam. If the AI can't make the point in three short paragraphs, it's trying to sell too much too soon.
Give the AI examples of what good looks like, not just adjectives. Tell it to copy the rhythm of a message you would actually send to a peer. A list of banned words helps, but a short example does more work than the word "casual" ever will.
Vocabulary
- Dimension
- synergy, game-changer, excited to connect
- Avoid
- use, total, helpful, glad to see
Structure
- Dimension
- Long paragraphs, formal greetings
- Avoid
- Short sentences, contractions (I'm, don't)
Formatting
- Dimension
- Bullet points, bold text, emojis
- Avoid
- Plain text, line breaks for readability
Document your handling logic
The most dangerous part of AI outreach is the reply. If the AI handles a nuance poorly, it can kill a deal. Document exactly how the AI should respond to common objections or unexpected questions.
Create a logic map for the AI. For example, if a prospect mentions price before they understand value, the AI shouldn't just give a number. It should give a rough frame and then ask about the problem they're trying to solve.
Know when the AI should stop. Certain triggers, like deep technical questions or complex pricing negotiations, should hand the conversation to you. Write those triggers down so the AI doesn't try to improvise through a conversation it should not own.
For every common reply, write one line: if the prospect says X, do Y, then Z. That format is easy for an AI to follow and easy for you to edit when a new objection shows up.
Example
- Prospect
I'm not the right person for this, maybe talk to Sarah.
- Suggested reply
Thanks for pointing me to Sarah. Do you mind if I mention you suggested I reach out to her, or should I just send a cold message?
- Prospect
How much does this cost per month?
- Suggested reply
It usually lands around X for smaller teams and Y for larger ones. The exact number depends on your setup, so I'd rather not guess. Can I ask what you're trying to fix first?
Test the playbook on real profiles before you send
Before you let the AI send anything, run the playbook against profiles you can already see. Pick ten people who match your customer profile and ask the AI to draft the first message for each one. Read the drafts as if they landed in your inbox.
You are looking for two things: does the message make sense for that specific person, and does it sound like something you would send? If every draft opens with the same compliment about their company, the playbook is too loose. If every draft sounds like a press release, the tone rules are not working.
Edit the playbook after the test, not after a week of live sending. Small changes to the goal and tone rules are easier to judge when you can compare drafts side by side.
- The first line names a real reason for the message.
- The goal is a yes or no question.
- The message has no banned words and no long paragraphs.
- The next step is clear and small.
If a draft needs a paragraph of explanation before it makes sense, the goal is too vague.
Doing this with Kinlead
You can save your sales script as an optional sales playbook in Kinlead. The AI uses that playbook, your campaign goal and your chosen tone to draft messages that follow your rules.
Kinlead sends at most one action every 5 minutes, inside daily invite and message limits you set and inside a send window in the lead's time zone. You can read and edit queued campaign messages on the Actions page before they send.
If you use Pro, any week after a two-week warm-up with fewer than three qualifying bookings, with at least one agent active all week, can be claimed as a refund minus transaction charges. See the minimum booking guarantee.
Paste this job
Playbook Prompt Example: Goal: Confirm the prospect is using manual CSVs for reporting. Tone: Direct, casual, no corporate jargon. Constraints: Max 3 sentences. No emojis. Use contractions. Logic: If they mention a competitor, acknowledge the competitor's strength in X but ask if they struggle with Y.
Frequently asked questions
Related
Get started with Kinlead
Prefer markdown? How to write a sales playbook for AI LinkedIn outreach.md