The AI Edge

The Briefing

Voice AI for sales calls hit a new gear this week. Platforms like Goodcall (fresh updates as of June 15) and Retell are pitching agents that book, qualify, and even close deals on the phone with no human on the line.

The promise is seductive. The catch is that for anything high-ticket, a robot voice handling the close is still a fast way to torch trust.

Meta rolled out AI business agents that answer questions, collect details, and qualify leads inside your ad funnel. Pair that with the new AI-powered Pixel that auto-adds product and business info, and the grunt work of lead capture keeps shrinking.

If you run paid traffic, less of your week now goes to plumbing and more to the offer itself.

HubSpot expanded Breeze into five specialist agents plus a no-code agent builder and a marketplace where you install agents like apps. Translation for small teams: you can now wire up a CRM that follows up, drafts, and routes leads without hiring for it.

The "agent marketplace" model is how most of us will buy AI features going forward.

At WWDC, Apple opened up Apple Intelligence so you pick which model answers: ChatGPT, Google Gemini, or Anthropic's Claude, each with its own voice. For marketers, that means the model you trust for copy and thinking can ride along on your phone. Pick on purpose. The default is rarely the best one for your job.

The Build

This week the voice-agent hype made me twitchy, so I ran a test on my own sales process.

Here is my bias up front. I have spent 15 years in direct response. The offers I care about most are high-ticket, and high-ticket still closes human to human. Handing a stranger's first real conversation with my brand to a bot felt like a mistake I would pay for later.

So I flipped the idea. Instead of AI taking the call, I had AI prep me for the call.

I keep a markdown file with my whole high-ticket sales framework: the discovery structure, the diagnosis questions, the objection reframes, the case-story slots. I loaded that file into Claude as the rulebook. Then I fed it one prospect's application answers and asked for a one-page call map.

Here is the prompt I used:

You are my sales-call prep assistant.
Use ONLY the framework in the attached file
(high-ticket-sales-scripting). Do not invent tactics.

Here are the prospect's application answers:
[paste raw answers]

Give me a one-page CALL MAP, in bullets, not a script:
1. Their real pain in one sentence (quote their words).
2. The outcome they actually want (not what they typed,
   what it means to them).
3. Three diagnosis questions tailored to THESE answers.
4. The single objection most likely to come up, and the
   exact reframe from the framework.
5. Which case story to tell and why.

If the application answers are too thin to know something,
say "not enough info" instead of guessing.

What came back was tight. A one-line read on the prospect's pain in their own words. Three diagnosis questions that actually fit their situation, not generic filler. A predicted objection ("I need to think about it") with the reframe lifted straight from my own framework, so it sounded like me. And a pick on which client story to use.

I ran it before my calls this week.

What worked: Prep time per call dropped from about 25 minutes to roughly 4. The objection it predicted was right on 2 of the 3 calls, so I walked in already holding the reframe instead of scrambling for it.

Best part: because the framework feeding it was mine, the language came back in my voice, not some generic closer's. I sounded like me, just sharper and faster.

What didn't work: When the application answers were thin, it ignored my "say not enough info" rule and invented pain points that were not there. Confident and wrong is the worst combination on a live call. It also over-scripted on the first pass, handing me word-for-word lines that read like a robot doing a Belfort impression. I had to force it back to bullets so I would still sound human.

The fix was in the prompt, not the model: ask for a map, never a script.

The Play

If you sell anything on a call (coaching, consulting, an agency retainer, a high-ticket program), here is the move while everyone else chases robot closers.

Do not automate the call. Automate the hour before it.

Take whatever sales framework you already trust (yours, a mentor's, a book you swear by) and paste it into your AI as the rulebook. Then feed it your prospect's intake or application answers and ask for a one-page call map in bullets. You keep the human warmth that closes the deal. The AI just makes sure you never walk in cold.

Steal this starter prompt:

Act as my call-prep assistant. Rules come ONLY from the
sales framework I'm pasting below. Don't add your own tactics.

FRAMEWORK:
[paste your sales process / script / notes]

PROSPECT INTAKE:
[paste their form answers]

Output a one-page CALL MAP in bullets (not a script):
- Their pain, in their own words
- What they really want
- 3 tailored questions to ask
- The most likely objection + the reframe from my framework
- The one proof story to tell
If you don't have enough info, say so. Do not guess.

One honest warning from my week: the AI will confidently make things up when the intake is thin.

Read the map, trust your gut over the guesses, and keep the close human.

Try this today. Before your next sales call, paste your sales process plus the prospect's intake answers into the prompt above and print the one-page map.

Walk in with the objection already handled.

Download high-ticket-sales-scripting-1.md