Turn your sales call transcript into a proposal (prompt inside)
Writing proposals shouldn't drain your energy. It's a structuring task, not a writing task. Here's how I automate it.
We’ve all been there: you wrap up a great sales call, your notes are solid, and then you open a blank document. Immediately, all the energy from the conversation vanishes into the exhausting chore of writing it up.
For years, writing those offers took my team half a day per deal. Now? It takes under 30 minutes. We didn’t figure out a way to type faster or work harder. We just built a better system.
It’s not a writing task
I used to get this completely wrong. I treated every new proposal like a blank canvas, which meant my team essentially rewrote the same document from scratch after every single call.
But I realized that winning proposals always share the exact same anatomy:
The client’s needs and vision
Their current situation
The recommended solution
That structure never changes. The only variable is what the client actually said.
And since everything they said is already recorded in the call transcript, the job isn’t writing. It’s extracting and structuring.
Humans burn out doing that by hand. AI does it in seconds.
How the prompt actually works
First, I record the sales meeting (always asking for consent on the first call, of course).
Afterward, I dump the transcript into a prompt that is already loaded with our company’s “brain”: our product catalog, qualification criteria, brand voice, and sales process.
In one pass, the AI spits back the structured core of the offer:
Needs and vision: What the client wants to achieve in their own words, plus any hard deadlines or driving events they mentioned.
Current situation: Where they’re stuck right now, the tools they use, and their specific constraints.
Recommended solution: Built strictly from our actual product catalog. It maps their problems to specific items, explains why it’s a fit in one sentence, and flags if they don’t quite qualify for something they asked about.
The most important part? It doesn’t hallucinate. If a detail is missing from the transcript, the prompt drops in a [CONFIRM THIS] marker instead of making something up.
From there, I do a quick quality-assurance read to ensure nothing got lost in translation. Once approved, it drops straight into our proposal template, automatically pulls in the client’s details from the CRM, and it’s ready to send.
Sales Offer Agent prompt
Here is the example that you can use in your GPT Agent or Gemini Gem (derived from my Sales Offer Agent running in Gemini Gem):
ROLE
You are a sales offer assistant for [COMPANY NAME]. You take the raw
transcript of a sales meeting and produce the structured core of a
sales offer. You do not write the full proposal. You produce the three
sections below, accurately, using only what the transcript and the
loaded knowledge base support.
KNOWLEDGE YOU HAVE
You have access to the following loaded documents (see Part B):
1. Company overview
2. Brand voice guide
3. Products and services with qualification criteria
4. Sales process
Use them. Do not invent products, prices, timelines, or claims that are
not in these documents or stated in the transcript.
INPUT
A meeting transcript (or meeting notes) from one or more sales calls
with a single prospective client.
OUTPUT
Produce exactly three sections, in this order, with these headers:
1. CLIENT NEEDS AND VISION
What the client is trying to achieve and why it matters to them.
State the outcome they want, in their words where possible. Include
the underlying goal behind the stated request when the transcript
reveals it. Do not pad. If they named a deadline or a driving event,
include it.
2. CURRENT CLIENT SITUATION
Where the client is now: their current process, the tools they use,
the specific problem or bottleneck, and any constraints they stated
(budget signals, team size, technical limits, timing). Pull only
what the transcript supports. Mark anything the rep should confirm.
3. RECOMMENDED SOLUTION
The solution you recommend, built only from the products and services
in the loaded knowledge base. For each recommended item:
- name it exactly as it appears in the products document
- state in one line why it fits this client's situation
- note any qualification criterion the client meets or does not
yet meet for that item
If the client does not qualify for something they asked about, say so
plainly and recommend the closest item they do qualify for.
RULES
- Use only what the transcript and the knowledge base support. If a fact
is missing, write [CONFIRM: ...] rather than guessing.
- Match the brand voice guide for tone. No hype. No superlatives.
- Name tools and products exactly. Never use a category where a name exists.
- Every recommendation states the situation it fits. No unconditional claims.
- Do not write pricing unless it is in the products document.
- Do not write the cover letter, the closing, or the legal terms. Only
the three sections above.
COMPLETENESS SELF-CHECK
Before you finish, confirm each of these. List any that fail as
[GAP: ...] at the end of your output:
- Section 1 names a concrete outcome the client wants.
- Section 2 names the current tool or process and the specific problem.
- Every item in Section 3 maps to a named product or service.
- Every recommended item has a one-line fit reason.
- Every [CONFIRM] and [GAP] the rep needs to resolve is listed at the end.The part that most people miss
The prompt is only as good as what you load behind it. These are the four documents the Sales Offer Agent needs. Write them once for your company. Keep them short and specific. A page each is usually enough.
1. Company overview
What the assistant needs to know about who you are so it frames the offer correctly. Include:
what your company does, in two or three sentences
the markets and company sizes you serve
the outcomes clients typically come to you for
one or two proof points you are comfortable referencing in an offer
Keep it factual. This is context, not marketing copy.
2. Brand voice guide
How an offer from you should read. Include:
the tone you use with clients (for me: direct, specific, calm, no hype)
words and phrases you never use
how you talk about results (for me: always with a number and a constraint, never a vague promise)
whether you write in first person, team voice, or company voice
If you already have a voice guide for your content, reuse it. The offer should sound like you.
3. Products and services with qualification criteria
The most important document. The assistant builds the recommended solution only from this list, so it has to be complete and exact. For each product or service:
the exact name you want used in offers
a one-line description of what it is
who it is for and who it is not for
the qualification criteria a client must meet for it to be a fit (company size, current stack, budget signal, technical prerequisite, whatever is real for you)
pricing, only if you want the assistant to include it
If a service has tiers, list each tier as its own entry with its own criteria. Vague entries here produce vague offers.
4. Sales process
How your offer fits into the deal flow, so the assistant produces the right artifact at the right stage. Include:
where in your pipeline the offer is created
what the sales rep collects on the call that this prompt depends on
what happens to the offer core after the assistant produces it (which template it goes into, who reviews it, what triggers the send)
This document is what lets the prompt slot into a real process instead of producing a document that sits in a folder.
Where this usually falls apart
This system is amazing, but it’s not magic. It’s only as good as what sits behind it. Here are the three things that will break it:
A thin product catalog: If your internal product descriptions are vague, the AI will recommend things the client doesn’t qualify for. The fix: Write crystal-clear criteria for your offerings before you run anything.
Weak discovery calls: If you didn’t actually uncover the client’s real problems on the call, the AI can’t invent them. It will just hand you a bunch of gap markers. The fix: Use a solid call script designed to extract the info those three sections need.
Skipping the human review: AI catches the structure, but it misses nuance. It might miss that subtle joke or specific hesitation the client mentioned. The fix: A human reads the proposal before it goes out. Every single time. No exceptions.
🔧 Tools & Resources
Claude/Gemini/ChatGPT: Honestly, either works perfectly. The real work is writing your knowledge base once, not stressing over which model to pick.
Pipedrive Smart Docs: We use this to auto-fill company names, contacts, and dates straight from the CRM so the boilerplate text assembles itself.
Fireflies.ai: It handles the call recording and transcription seamlessly, plus it has robust API access, making it incredibly easy to port your transcripts directly into your external applications and automated workflows. Just ensure you have client consent. No transcript, no pipeline.
There’s also one step upstream of the call that I highly recommend automating, and that’s the lead qualification.
Before my team spends a single minute on a prospect, an AI qualifies the inbound lead via email, SMS, or voice, and only books a meeting if they are a genuine fit.
I use Ninjabot’s AI qualification for this. It protects a salesperson’s most valuable asset: hours spent in front of qualified people.
Just keep in mind, this only works if your qualification rules are genuinely strict.
Here is my challenge to you: Open your last recorded call. Paste the transcript into an AI. Read the core offer it gives you back. It will take you three minutes.
The exhausting part of sales was never the part that mattered.
You are irreplaceable on the call itself.
Everything after that is just structure, and structure is the one thing you never have to do by hand again.
Build a calmer business,
– Yuri
P.S. I am preparing the full nine-step sales offer automation guide, from initial inquiry to sent offer, with the clonable prompt and the knowledge-base spec you need to load behind it. I will publish it on my LinkedIn next week.
Yuri Vonchitzki
LinkedIn · YouTube · My services
Whenever you’re ready, there are 2 ways I can help you:
1. Join FutureBrief Plus, stop building alone. Get instant access to our private group chat for troubleshooting, Q&As with me, and a living library of proven automations, SOPs, and mental models to build a business that runs without you.
2. Let us build the machine for you. If you are an operator scaling past $500k and want to remove yourself from the daily grind, Ninjabot can audit your ops and build this entire ecosystem for you.
Reach 21,000+ business founders and operators implementing AI: sponsorship details are here.