How to use AI to turn your customer discovery call notes into a repeatable sales script your whole team can use without sales training

You just hung up one of the best sales calls you've ever had — the prospect got it immediately, the objections practically answered themselves, and they said yes. You have a vague memory of what you said. Your notes say "good rapport, talked about pricing, they liked the timeline." That's it.

This post walks you through how to create a sales script from customer call recordings using AI — pulling the gold out of your real discovery calls and building a structured script your whole team can run. No sales training required. If you've ever wondered how to write a sales script without sales training, this is the workflow.

It's more straightforward than it sounds, and you don't need any new tools you haven't heard of.

---

What you need before you start

Otter.ai{:target="_blank"} — a tool that joins your Zoom, Google Meet, or Teams calls and automatically writes out everything that was said, word for word. There's a free plan that covers 300 minutes of transcription per month, which is plenty to get started.

Fathom{:target="_blank"} — another transcription tool, and the free individual plan gives you unlimited recordings as of early 2026. It also auto-generates summaries and highlights right after the call ends, which saves a step. Either tool works for this workflow — pick one.

ChatGPT{:target="_blank"}, Claude{:target="_blank"}, or Google Gemini{:target="_blank"} — this is where the actual script gets built. The free versions can handle this task. If you have a paid plan (GPT-4o, Claude 3.7 Sonnet, Gemini 1.5 Pro), even better — they handle long transcripts without hiccups.

Time required: About 90 minutes the first time. Faster once you've done it once.

Skill level: If you can copy and paste text into a browser, you can do this.

---

Why your best calls are already a sales script from customer call recordings waiting to happen

Here's the thing most small business owners miss: the best sales script for your business isn't in a template you download from the internet. It's already in your closed-won calls — the language your customers used to describe their problem, the exact moment they shifted from skeptical to interested, the objection you handled smoothly without even thinking about it.

Gong's research on discovery calls{:target="_blank"} found that in successful calls, the prospect talks about 57% of the time. That's a lot of customer language sitting in your transcripts — their exact words, their exact fears, their exact reasons for buying. That's the raw material. AI is just the tool that helps you organize it into something your team can actually follow.

---

Step 1 — Record and transcribe your calls

1. Set up Otter.ai or Fathom by connecting it to your calendar. Both tools will ask to integrate with Google Calendar or Outlook so they can automatically join scheduled calls.

2. Let the bot join your next call. Otter or Fathom will appear as a participant. It records and transcribes in real time — you don't have to do anything during the call except sell.

3. After the call ends, grab the transcript. Otter gives you a searchable text document. Fathom gives you a summary plus the full transcript. Either way, you're looking for a long text file of everything that was said.

The Otter.ai sales script workflow starts here — once you have a handful of transcripts from your best closed-won calls, the rest of the process moves quickly.

One small heads-up before you do any of this — jump to the legal section at the bottom of this post. You need to tell people they're being recorded. Fathom and Otter both announce this automatically, but it's worth understanding why.

---

Step 2 — Select the right calls to build your AI sales script from discovery calls

Don't grab just any call. This part matters more than people think.

1. Go through your last six months of closed deals and identify five to ten calls where the prospect said yes. Not your fastest calls — your best ones. The ones where the conversation felt natural and the prospect seemed genuinely convinced.

2. Skip calls where the sale felt like a fluke — the client who was going to buy no matter what, or the one where a mutual friend vouched for you. Those don't contain the pattern you want to replicate.

3. Gather your transcripts into one place — a folder on your desktop, a Google Doc, wherever. You'll be pasting them into the AI tool in the next step.

Five to ten calls is the sweet spot. One call gives you one conversation. Five calls give you patterns — the objections that come up every time, the phrases that land, the questions that open people up.

---

Step 3 — The exact prompts to turn call transcripts into a sales script

Open ChatGPT, Claude, or Gemini. Start a new conversation. Paste in your first transcript (or as many as fit — modern AI tools can handle very long documents).

This first prompt sets up the AI to think like a sales coach, not a summarizer. That framing makes a real difference in the quality of what comes back.

You are an experienced sales coach reviewing transcripts of successful discovery calls for a small business. Your job is to identify the patterns that made these calls successful. From the transcript(s) I provide, extract the following: (1) the top 3–5 pain points the prospect described in their own words, (2) the objections that came up and how they were handled, (3) the questions that got the prospect talking most openly, and (4) the moments that seemed to shift the prospect toward wanting to move forward. Use the customer's actual language wherever possible, not paraphrased summaries.

Paste the transcript below the prompt and run it. Read what comes back. It'll probably surprise you — you'll see patterns you didn't consciously notice.

Now run this second prompt to turn those patterns into a draft script structure:

Using the patterns you just identified, create a modular sales script structured into four sections: (1) Opener, (2) Discovery Questions, (3) Objection Responses, (4) Transition to Close. Format this as a table with four columns: Stage | What to Say | Why It Works | Common Objection at This Stage. Use the actual customer language you found in the transcripts. Keep each response under three sentences.

The table format is deliberate. A narrative script is hard for a new team member to use under pressure. A table gives them the context — what to say and why — in a quick glance.

---

Step 4 — Format the script so your small business team can actually use it

1. Copy the table from the AI and paste it into a Google Doc{:target="_blank"} or Notion{:target="_blank"} page. Both tools let you maintain a living document your team can access.

2. Add a short intro note at the top explaining what the script is and isn't. Something like: "This is a guide, not a word-for-word recitation. Use the discovery questions as a starting point and let the conversation breathe."

3. Share it with your team and walk them through it once — even a 20-minute call where you explain the reasoning behind each section goes a long way. The table format helps here because you can explain the "why" column out loud.

Team members who've never sold before are much more confident when they understand the reason behind what they're saying, not just the words. That's the whole point of the table structure.

---

Step 5 — Test, refine, and keep the script current

1. Role-play the script with a team member before anyone uses it on a real call. Notice where it feels wooden or awkward.

2. Go back to the AI and give specific feedback. This step is where most people stop too early — don't.

The objection response in the pricing section sounds too defensive. Rewrite it to feel more empathetic — like we understand why price is a concern, not like we're justifying ourselves.

3. Update the script every quarter as new calls come in. Add a fresh batch of transcripts, run the extraction prompt again, and note anything new — new objections, new language, things that used to work but don't anymore.

A sales script that's six months old without updates is already drifting from what your customers are actually saying. Worth the hour to keep it current.

---

Legal basics: what you need to know before recording calls

Recording calls without consent can create real legal problems, so here's the short version.

In the US, federal law only requires that one person on the call knows it's being recorded — and that can be you. But eleven states, including California, Florida, and Illinois, require that everyone on the call consents. If you're in one of those states, or if you're calling customers who might be, you need to tell them at the start of the call.

The easy fix: both Otter.ai and Fathom automatically announce when they join a call, which handles disclosure in most cases. You can also just say it yourself: "Hope it's okay — I record my calls so I can focus on the conversation rather than notes." Most people appreciate that.

If you work with clients in other countries, check local rules. This is a "ask your lawyer for five minutes" situation, not a reason to skip the workflow.

---

When something goes wrong

The script sounds generic, like it could apply to any business. This usually means you fed the AI one transcript instead of several, or you skipped the role-framing in the prompt. Go back and add the "You are an experienced sales coach" setup, and add more closed-won transcripts.

The AI summary misses objections you know came up. Otter's AI Chat feature lets you ask direct questions about a transcript. Try typing "What objections did the prospect raise about cost?" directly into the chat. You can also re-prompt: "Review the transcript again specifically for any moment where the prospect expressed hesitation, doubt, or a reason not to move forward."

Your team isn't using the script. This is almost always a format problem, not a motivation problem. If the script is a wall of text, nobody reads it under pressure. Go back to the table format — stage, what to say, why, objection. Short. Scannable. Usable mid-call.

---

What to do next

Put this workflow on your calendar for a recurring quarterly session — 90 minutes, a fresh batch of call transcripts, a script update. That's the habit that makes this stick.

If you want to take this further and build out the rest of your sales process, check out the Off Prompt walkthrough on using AI to automate sales follow-up emails for small business teams.

---

FAQ

How do I create a sales script from customer call recordings if I only have a few calls? Honestly, even three solid closed-won calls will give you something useful. Five to ten is where patterns really start to show up. If you're just starting out and don't have many recorded calls yet, start recording now and run the workflow in a month or two once you have enough material.

Do I need a paid AI plan to build a sales script from call transcripts, or will the free versions work? The free versions of ChatGPT, Claude, and Gemini can all handle this workflow. The one limitation is document length — very long transcripts might get cut off on free plans. If that happens, split the transcript into two parts and run the prompts on each. Paid plans (usually around $20/month) remove that limitation and give you access to the most capable models, but they're not required to get started.

What if I don't do video calls — can I still record calls to build a sales script? A common question. Otter.ai works with in-person meetings and phone calls through its mobile app, which can record audio directly. You'd start the recording manually rather than having it auto-join a link. The transcript quality is slightly lower than a dedicated meeting link, but still very usable for this workflow.

Can I use lost-deal call recordings to figure out what went wrong in my sales process? Yes, and it's actually a great exercise. Run the same extraction prompt on a few lost deals and compare the patterns to your won deals. You'll often find the difference isn't what you think — it might be a question that was never asked, or an objection that was handled defensively instead of curiously. Ask the AI: "Compare these transcripts to the successful calls. What's different?"

Is it weird to tell clients I'm recording the call? It felt a little awkward the first time. It doesn't anymore, and most clients don't think twice about it — especially if you frame it as "so I can stay present in the conversation instead of writing notes the whole time." That framing is honest and most people genuinely appreciate it.