How to Write Sales Call Scripts With AI That Don't Sound Scripted

Don't ask AI for a word-for-word script. Give it transcripts or notes from your best real calls and ask for a call map: a one-line opener, five or six questions, short answers to your common objections and one clear next step, all as bullet points you glance at rather than read. Then rehearse it aloud, ideally in an AI voice roleplay.

Scripts sound scripted because they're written to be read. They use full, tidy sentences, leave no room for the other person, and imagine the prospect's replies in advance. People on the other end notice within a sentence or two, and the call turns into a polite exit. A call map keeps the structure you need, what to cover and in what order, while leaving the words to you in the moment. AI is good at building that structure from your own material; it's bad at sounding like you from nothing.

Follow me on Instagram@sagnikteaches

Why AI-written call scripts sound like scripts

Ask a chat assistant for "a sales call script for my business" and the result is fluent written English. That's the problem: speech is shorter, looser and full of questions. Here are typical lines from a generated script, next to what a person would actually say:

Connect on LinkedInSagnik Bhattacharya

The spoken versions are shorter, name something specific, and end with a question the other person can actually answer. Every rule in the rest of this tutorial is a way of getting AI to produce more of the right-hand column.

Subscribe on YouTube@codingliquids

Start from your own best calls, not a blank prompt

The example here is hypothetical: a café that wants to sell breakfast and lunch catering to offices of 20 to 60 people nearby. The owner has had a handful of good conversations with office managers who came in for coffee, and a few calls that went nowhere.

The best raw material is a recording or transcript of a call that went well. Recording rules differ between places and some require everyone's agreement, so always say at the start that you'd like to record for your notes, and stop if they'd rather you didn't. No recordings? Write up the last two good conversations from memory straight afterwards, including the phrases you remember using and what the other person said. Then ask:

Below are notes and a transcript from three of my best sales conversations
with office managers about catering. Pull out:
1. The exact phrases I used that got a positive reaction (quote them)
2. The questions I asked that got them talking
3. Their concerns, in their words
4. What the next step was each time
Don't rewrite anything into polished language. Keep my phrasing.

What came back (illustrative) included phrases the owner didn't realise she used: "we can have it on your desk by 8:45", "no minimum for the first order", and a question that had opened up two conversations, "what do you do for food when you've got visitors in?" Those phrases are the core of the call map. They're hers, they've worked, and they sound like a person because a person said them. If you want the same voice across calls, emails and proposals, a brand voice guide that AI can follow captures it once and reuses it.

Build a call map instead of a script

A call map fits on one screen. The template:

CALL MAP: [type of call]
Goal of this call: [one next step, e.g. book a tasting]
Opener (one line): [who you are, why this person, permission question]
Reason for calling (one line): [the specific trigger]
Questions (pick 3-4 of these):
  - ...
Proof points (only if relevant): [2-3 facts]
Objections: see cards
Next step to ask for: [exact ask]
If not now: [smaller ask]

And the café's filled-in version, built from her phrases and a second prompt asking the assistant to organise them into the template without adding new claims:

• Goal: book a free tasting box for their next team meeting.
• Opener: "Hi, is that [name]? It's [me] from [café], just round the corner on [street]. Have you got a minute, or is now bad?"
• Reason: "A few people from your office come in most mornings, and I've started doing breakfast platters for meetings."
• Questions: "What do you do for food when you've got visitors or a team meeting?" · "Who sorts it out, you or someone else?" · "What's annoying about how it works now?" · "How often does that happen, roughly?"
• Proof points: on your desk by 8:45 · no minimum for the first order · everything labelled for allergens.
• Next step: "Could I send over a tasting box for your next meeting, on us? Which day's best?"
• If not now: "Can I email you the menu so it's there when you need it?"

Notice what's missing: no paragraph about the café's story, no list of every item on the menu. Those go in a separate sheet to open only if asked.

Openers for three kinds of call

The opener does one job: earn the next 30 seconds. It should say who you are, why you're calling this person, and ask permission. It changes with the kind of call:

• Cold call: "Hi, it's [name] from [café] round the corner. I'm calling a few offices nearby about meeting catering. Is now a bad time?" Asking if it's a bad time sounds backwards, but it gives the person control, and people who say "go on then" listen properly.
• Calling back an enquiry: "Hi [name], it's [me] from [café], you filled in our catering form yesterday about a lunch on the 14th. Is now still good to talk it through?" Name the detail they gave you so they know you read it.
• Following up a quote: "Hi, it's [me] from [café]. I sent the quote for the Thursday breakfasts on Monday and wanted to check it made sense. Did anything look off?" "Did anything look off" invites the real objection; "any questions?" invites "no, it's fine".

The same shapes work in other trades. An illustrative bicycle repair shop calling a delivery company about servicing its riders' bikes opened with: "It's [name] from [shop], we're the ones near the depot. A few of your riders come to us when their brakes go, and I wondered who looks after the fleet at the moment." An illustrative bakery calling restaurants about wholesale bread: "I'm calling a few places that serve brunch nearby. Who decides on the bread you use?" Both are short, both name a real connection, and both end with a question. To build a list of the right people to call, see finding B2B prospects with AI research tools.

Questions that turn a pitch into a conversation

The questions are where a call stops being a pitch. Good ones are open (they can't be answered with yes or no), about something that has already happened, and specific to the person's situation. AI can produce a long list quickly; your job is to cut it down. The prompt the café owner used:

I sell meeting catering to office managers. Write 15 questions I could ask
on a first call to understand how they handle food for meetings now.
Mark each one: OPEN or CLOSED, PAST or HYPOTHETICAL.
Keep each under 15 words, in plain spoken English.

A selection from what came back, with her verdicts (illustrative):

• "What did you do for food at your last team meeting?" (open, past) Kept: easy to answer and always gets a story.
• "Would you be interested in higher-quality catering?" (closed, hypothetical) Cut: nobody says no, and the yes means nothing.
• "Who else gets involved when you order food?" (open, past) Kept: finds the real decision-maker early.
• "What's your catering budget per year?" (open, but awkward) Cut: too formal for a first call. "Roughly what do you spend on a meeting like that?" works better later.
• "What went wrong the last time food was a problem?" (open, past) Kept: the best one on the list, because it surfaces the pain in their own words.

Four questions made it onto the map. That's enough: a good first call usually only gets through two or three, because each answer leads somewhere you follow instead of moving on to the next line.

Objection cards: short answers you can find in two seconds

Objections are the moment most people revert to reading, because they panic. A card per common objection solves that. Give the assistant the concerns from your real calls and ask:

Here are the objections office managers raised in my calls, in their words.
For each, write an objection card with four lines:
1. What it usually means underneath (one line)
2. A reply I could say out loud (max 20 words, plain spoken English)
3. A question to hand the conversation back (max 12 words)
4. When to accept the no and move on
Use my phrases from the list below where they fit. No sales jargon.

Sample cards (illustrative), with the owner's edits:

The first AI draft of the price card read: "I understand budget is a key consideration for your organisation, and we pride ourselves on offering competitive value." It's 17 words of nothing. The rule "max 20 words, plain spoken English" helped, but the owner still rewrote it to include the actual price, because a number is more reassuring than a claim. Each card also needs the fourth line, when to accept the no: for "we have a contract until next year", the answer is to ask when it's up and put a reminder in the diary.

Rehearse with an AI roleplay

A call map only stops sounding scripted when you've said it aloud enough times that you don't need to read it. Voice modes in chat apps, such as ChatGPT's voice conversations or Gemini Live, make good practice partners because they answer back in real time. Start with a prompt like this, typed before you switch to voice:

Let's roleplay a sales call. You are an office manager at a 40-person
firm. You're busy, mildly interested, and currently order sandwiches
from a supermarket for meetings. Don't make it easy: raise one or two
of these objections naturally: price, "send me an email", allergies.
Keep your replies short, like a real phone call. When I say "stop",
tell me: where I sounded scripted, where I talked too long, and
whether I asked for a clear next step.

Run it three times with different moods: busy and curt, friendly but vague, and price-focused. Feedback from one practice run (illustrative): "Your opener was good. You spoke for about 40 seconds without a question after I mentioned allergies. You said 'comprehensive' twice. You didn't ask for the tasting; you offered to send a menu, which let me off the hook." That last point is the most common one in practice runs, and hearing it from a patient roleplay partner is kinder than hearing it from silence on a real call.

Treat the roleplay as practice, not prediction. The AI is a reasonable sparring partner, but real office managers will surprise you in ways it won't.

The café's first ten calls

With the map and cards on one sheet of paper and three practice runs done, the owner made her first batch of calls over two afternoons. The numbers are illustrative:

After call four she changed one thing on the map. Three people had asked "how far ahead do I need to order?" before she mentioned it, so "order by 3pm the day before" moved from the menu sheet to the proof points. The weekly order, a Monday breakfast for a 35-person team, was worth about $315 a week at $9 a head. Twenty-five calls, and roughly six hours including preparation, paid back many times over.

Tuning the map after every batch of calls

A call map is a draft that your calls keep correcting. After every five to ten calls, spend 20 minutes on it. If you recorded (with permission), transcribe the calls and ask an assistant to find:

• Where the other person talked most, and which question set it off.
• Objections that came up but aren't on your cards.
• Moments you talked for more than 30 seconds without a question.
• Whether you asked for the next step, and the exact words you used.

That review is exactly what turning sales call recordings into coaching notes covers in depth. It also helps to log what was agreed straight after each call, so follow-ups don't depend on memory; updating your CRM automatically after sales calls shows how to take that job off your hands, and automating follow-ups without being pushy handles what happens next.

Habits on the call that give a script away

• Reading instead of glancing. If your eyes are on the sheet for more than a second or two, the map has too many words on it.
• The rehearsed pleasantry. "How are you today?" to a stranger signals a sales call instantly. Get to the reason.
• Not pausing after a question. Count two seconds of silence before filling it. The answer usually comes at second three.
• Ignoring what they said. If they mention a board meeting next week, the next question is about the board meeting, not question three on your list.
• Listing features. Three proof points, used only when they answer something the person said.
• Vague endings. "I'll leave it with you" is not a next step. "Shall I pencil in Thursday?" is.

An illustrative florist calling hotels about weekly lobby arrangements found her biggest giveaway was the last one: she had excellent conversations that ended with "let me know". Rewriting the final line of her map to "Could I bring two sample arrangements on Tuesday morning?" turned two of her next five calls into visits. If the next step is a meeting with a presentation, writing a sales deck with AI that wins meetings picks up from there.

Sales call maps: questions from owners

Can I record sales calls to feed the AI?

Often, but recording rules differ between places, and some require everyone on the call to agree. The safe habit is to say at the start that you'd like to record to keep accurate notes, and to stop if the person declines. Keep recordings only as long as you need them, and don't upload them to a consumer AI plan that trains on your data.

Should I tell prospects I used AI to prepare?

There's no need to announce that you prepared with AI, any more than you'd mention using a notebook. What matters is that everything you say is true and that you, not a tool, are the one talking. If you use an AI voice agent to make the calls themselves, that's different: people should be told they're speaking to an automated system.

How long should a call map be?

One side of paper, or one screen you can read at a glance. If it runs longer, you'll start reading from it, which is the thing you're trying to avoid. Keep the opener to one line, the questions to five or six, and each objection answer to a sentence or two; put the detail in a separate sheet you only open when asked.

Further reads

• How to Write Cold Emails With AI That Stay Out of Spam — Warm up cold calls with an email that actually lands.
• How to Write Business Proposals Faster With AI — Turn a good call into a proposal the same day.
• AI Call Summaries: Log Every Phone Enquiry in Your CRM — Log every phone conversation without typing notes.
• How Small Businesses Use AI in Sales: 10 Real Examples — Other ways small firms use AI across the sales process.
• How to Review AI Call Transcripts for Quality and Compliance — Check AI call transcripts before you rely on them.
• Winning Back Lapsed Dental Patients With AI Messages — How to find lapsed dental patients, split them by why they stopped coming, and use AI to write reactivation messages that don't guilt or scare.
• AI Lead Generation vs Hiring a Salesperson: A Cost Comparison — Itemised costs of an AI prospecting stack and a salesperson, a caterer's year-one comparison per meeting and per client, and when each wins.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: vendor help pages for ChatGPT voice conversations and Gemini Live; general guidance on call recording consent. Example figures are illustrative.