Scalegram's AI assistant answers your Telegram profile with whatever instructions you give it. Vague instructions produce a polite receptionist. A real playbook produces a closer. The structure below is the one that works, and it maps directly onto the assistant's configuration.
1. Identity and Tone
Tell it who it is and how it speaks. "You are my assistant, answering as my profile. Tone: confident, warm, short messages, no corporate language, emojis sparingly. Match the lead's language; you speak English, Spanish and Italian." Tone leaks into everything, so nail it first.
2. The Offer, Stated Like You State It
Write your pitch the way you actually send it, not like a brochure. What you offer, what it costs, what the lead gets, what the next step is. Include the exact links to send. The assistant will reuse your phrasing, which is precisely what you want: leads should not feel a register change between the AI and you.
3. Qualification Questions
Give it two or three questions to ask early: experience level, budget range, country. Answers land in the contact record, so by the time you open the chat the lead is already profiled. This is where the assistant quietly does your pipeline hygiene for you.
4. Objections: Your Ten Best Answers
List the objections you hear weekly and your strongest response to each. "Is this a scam", "I lost money before", "can I start smaller", "why do I need to verify". Write your real answers, the ones that work in your voice. This section earns more money than any other paragraph in the playbook.
5. The Close and Its Limits
Tell it when to push: "Once the lead confirms budget and interest, send the sign-up link and ask them to complete the deposit today. Follow up once if they go quiet. Never promise profits, never quote returns, never offer anything not listed above." The limits matter as much as the push. You are the sender of record for every message the assistant writes, so bake your compliance into the prompt itself.
6. Handover Rules
Define when a human takes over: an angry lead, a big-ticket question, a keyword like "call me", or plain uncertainty. The assistant tells the lead you will personally follow up, flags the chat, and stops. And whenever you simply start typing, it yields instantly.
Iterate from the scoreboard. Read the assistant's conversations weekly in the mirror, and check the analytics: where do leads drop, which objection stalls it, how fast does it reply at your peak hours? Tighten one section of the playbook at a time and watch the close rate move.
"Do not prompt the assistant like a chatbot. Prompt it like you are onboarding your best-ever sales hire on their first day: offer, objections, lines, limits."
— Roman Onta, Executive Director, SINGUARD
Key Takeaways
- Structure the playbook in six parts: identity and tone, offer, qualification, objections, close with limits, handover rules.
- Write your real phrasing, not brochure copy: leads should feel no register change between the AI and you.
- The objections section earns the most: ten honest answers to the questions you hear weekly.
- Put compliance in the prompt: no promised profits, no invented offers. You are the sender of record.
Frequently Asked Questions
How long should the playbook be?
As long as your real knowledge, not longer. A page or two of dense, honest instruction beats ten pages of filler. Start with the six sections, run it a week, then expand the parts the analytics show are weak.
Can the assistant run different playbooks for different audiences?
Personalisation is deep: tone, languages, quiet hours and rules are all configurable, and leads arriving through tracking links come pre-attributed with their source, which the assistant can use for context.
What must never go in a playbook?
Anything you could not lawfully say yourself: promised returns, invented scarcity, advice you are not licensed to give, or pressure tactics aimed at vulnerable people. The Legal Center's AI Assistant Notice spells out the responsibility: the messages are yours.