How to write an AI sales playbook that sounds human

Document your sales logic, tone, and objection handling so an AI SDR can replicate your best LinkedIn outreach without ever sounding like a bot.

Get started

Short answer

To write a sales playbook for an AI SDR, write down the goal for each touch, the rule for when the next one sends, and the tone limits the AI has to stay inside. Then add your five most common objections with the reply you would give. The AI follows what you wrote, so vague rules produce vague messages.

Define the goal for each step in the sequence

A playbook that says 'book a meeting' gives the AI nothing to work with. An AI SDR reads your instructions literally and applies them to every lead in the campaign, so each step needs one job and one definition of success.

Split the sequence into four moves: the connection request, the first message, the qualification question and the ask. Write down what counts as a win at each one. A win on the connection request is an accept. A win on the first message is any reply, including 'not interested', because a no tells you where the lead stands and silence tells you nothing.

Keep each win small enough that the AI can reach it. If the only win you define is a booked call, the AI will push for a call in message one and get ignored.

Connection request

Step
They accept
What a win looks like
One line on why they came up. No pitch.

First message

Step
Any reply, including a no
What a win looks like
One observation about their setup, one question

Qualification

Step
They describe how they do it today
What a win looks like
A question about their current process

The ask

Step
A time on the calendar
What a win looks like
Two specific slots, not 'let me know'

Write the follow-up logic before you write the messages

Most follow-ups fail because they restate the first message in different words. Write the rule first: what triggers the next touch, what it has to add, and when the sequence ends.

Use time as the trigger. Waiting for the lead to show interest stalls most sequences at touch one. Pick a number of business days, write it into the playbook, and let the AI move on its own.

The content rule matters more than the timing. Every follow-up has to carry something the previous touch did not: a different problem, a short example from a company that looks like theirs, or a question about how they handle the thing today. 'Just checking in' is the message that gets you removed from the thread.

  1. 1

    Set the gap

    Pick a number of business days between touches and write the number down. Three is common. Whatever you choose, the AI needs a figure, not 'a few days'.

  2. 2

    Name the angle for every touch

    Touch two covers a second problem. Touch three asks about their process. Touch four offers one thing to read. Write each angle into the playbook so the AI cannot repeat itself.

  3. 3

    Write the stop rule

    Decide how many unanswered touches end the sequence and where the lead goes afterwards. Without a stop rule the AI keeps going and your account carries the risk.

Set tone rules the AI can actually follow

Instructions like 'be friendly and professional' produce nothing useful. The AI has no reference point for those words, so it defaults to the safest, most formal English it knows, which reads like a template.

Give it constraints instead: how long a message can be, which words are banned, how it opens, whether it can use the lead's first name, whether it apologises for slow replies.

The rule I use most: write like you are messaging someone you met once at a conference. That one line sets distance, confidence and length better than a page of adjectives.

What the AI writes by default

  • I hope this message finds you well.
  • I wanted to reach out because we help companies like yours improve their workflow.
  • Would you be open to a brief call to discuss further?

What the rule produces

  • Saw you are hiring two SDRs. That usually means the manual research is getting heavy.
  • We take the list building off the SDR's plate. Worth a look?
  • Free Thursday at 2?

Build an objection library with a reply for each one

Replies are where AI outreach falls apart. The AI treats an objection as something to push through, so it argues. A person acknowledges the objection, asks one question, and lets the answer decide the next move.

Write down the five objections you hear most. For each one, write the acknowledgement, the question and the condition for handing the conversation to you. Keep the acknowledgement short. 'Makes sense' does more work than a paragraph of empathy.

Watch for the objection that is not one. 'Send me some info' usually means the person is being polite. A playbook that treats it as buying interest will send a deck and follow up three times on a dead thread.

Example

  1. Prospect

    We already use a tool for this.

  2. Suggested reply

    Makes sense. Is the reporting side handled, or is that still manual?

  3. Prospect

    Reporting is fine. The problem is the data going in is stale.

  4. You

    That is the part we fix. Want me to send over how we refresh the list? No call needed.

Put the whole playbook in one page the AI can read

You do not need a twenty page document. One page that the AI reads at the start of every campaign beats a wiki nobody updates, and it is short enough that you will actually revise it after the first ten leads.

Here is the shape. Copy it, replace the specifics, keep it under a page.

Leave out the company history, the product tour and the pricing table. The AI does not need them to write a first message, and every extra paragraph dilutes the rules that matter.

Paste this job

Goal: book a 15 minute call with heads of sales at B2B software companies with 20 to 200 staff.

Steps and wins
1. Connection request. Win is an accept. One line on why they came up. No pitch.
2. First message. Win is any reply. One observation about their team, one question.
3. Qualification. Win is them describing how they do it today.
4. Ask. Win is a time on the calendar. Offer two specific slots.

Follow-up rules
- 3 business days between touches.
- Every touch adds a new angle. Never repeat the previous message.
- Stop after 4 unanswered touches and move the lead to nurture.

Tone
- Under 45 words per message.
- No opening line about hoping they are well.
- Banned words: solution, synergy, circle back, touch base, reach out.
- Write like you are messaging someone you met once.

Objections
- 'We already have a tool' -> ask what is still manual today.
- 'No budget' -> ask when the budget resets, then stop.
- 'Send me info' -> send one link, ask one question, no call request.

Hand off to a human when
- They ask about price.
- They mention a deadline or a contract ending.
- They ask something the website does not answer.

Test the playbook on ten leads before you run it wide

Point the campaign at ten leads first and read every message the AI drafts. You are checking three things: whether the opening line is specific to that person, whether the follow-ups repeat themselves, and whether the objection replies argue.

Fix the playbook rather than the individual messages. If you edit a draft by hand, ask what rule would have produced that edit, then add the rule. Three edits to the same sentence means the playbook is missing something.

Then check the timing. A message that lands at 3am in the lead's time zone reads as automated even when the words are good. Set the send window to the lead's business hours and watch where the replies come from.

If you find yourself rewriting the same sentence in three drafts, that is a missing rule, not a bad draft.

Doing this with Kinlead

Kinlead takes the playbook as a campaign setting. You paste your sales script into the campaign, set the goal and the tone, and the AI writes the messages and follow-ups from it. Queued messages show up on the Actions page before they send, so you can read and edit them while the playbook is still settling.

Each campaign also decides how replies are handled: hand the conversation to you at the first reply, let the AI reply until there is real interest, or let it book a meeting with your scheduling link. If you want to read every reply before anything goes out, use hand-off mode, because the AI reply modes send on their own.

Kinlead runs outreach from your own LinkedIn account through Unipile and sends at most one action every 5 minutes per account, inside the daily limits you set and inside the send window for the lead's time zone. The objection examples above are worth reading in full at objection handling examples.

Frequently asked questions

Related

Get started with Kinlead

Prefer markdown? How to write an AI sales playbook that sounds human.md