OutcomeWalk into a conversation knowing what actually matters.
Most preparation is either nonexistent or performative: skimming a profile so you can mention their dog. Neither helps. What helps is arriving with a real hypothesis about what this person is dealing with.
Give the model what's publicly available (their role, their company, their industry, anything they've published) and ask it to reason about pressures, not to recite facts. What is someone in this role likely being measured on this quarter? What would make them look good? What are they probably tired of hearing from people like me?
That last question is the valuable one and nobody asks it. It tells you what not to say.
Two boundaries: everything you feed it should be information the person made public. Don't paste private conversations or client records into a general assistant. And the output is a hypothesis to test in the conversation, not a dossier to perform. If you walk in with a script, they can tell.
Drill15 min
Pick a real upcoming conversation. Run the prep prompt with only public information. Write down the one question it surfaced that you would not have thought to ask. Ask it.
Prompt: Conversation prep
Prep me for a conversation. Reason about pressures, don't recite facts.
WHO: [role, company, industry (public info only)]
CONTEXT: [how we connected, what they know about me]
WHAT I OFFER: [briefly]
Give me:
1. What someone in this role is probably being measured on right now
2. What would make them look good to their boss or their spouse
3. Three things they're likely tired of hearing from people like me
4. The one question I could ask that would show I actually thought about them
5. The strongest honest reason they should NOT work with me
Mark anything you're inferring rather than knowing. Do not invent details about this specific person.
Paste into ChatGPT, Claude or Gemini. Replace every [bracketed] placeholder with your own details first, then review the edited prompt before you send it.
Objection Practice
25 min
OutcomeUse AI as a sparring partner so you're not rehearsing on real people.
This is the most underused application of these tools in sales, and the one with the fastest payoff. You can run a difficult conversation twenty times before you have it once.
Set it up as a role-play with teeth: tell the model exactly who to be, how skeptical, and instruct it to stay in character and not go easy on you. Left to itself it will fold after one good answer, which teaches you nothing. "You are politely skeptical and you do not become convinced unless I genuinely address your concern. If I dodge, notice it."
Then debrief. Out of character: where was I weakest, what did I not answer, where did I talk past you. That debrief is worth more than the roleplay.
What it can't do is give you real emotional stakes. Nobody's rejection actually stings here. Use it to get fluent on the words so that in the real conversation you have attention left over for the person.
Drill25 min
Run the objection you dread most, three times, with a different angle each time. Then ask for the debrief. Most people find the same weak spot all three rounds. That is the one to fix.
Prompt: Objection role-play
Role-play with me. Stay in character until I say "debrief."
YOU ARE: [a specific prospect: role, situation, what they care about]
YOUR STANCE: Politely skeptical. Interested enough to keep talking, not convinced. You do NOT become convinced unless I genuinely address your concern. If I dodge, deflect, or answer a different question than you asked, call it out the way a real person would.
THE OBJECTION YOU RAISE FIRST: [the one you dread]
Open with one line, in character. Then respond to me naturally, one exchange at a time. Don't summarize. Don't coach me mid-scene.
When I say "debrief," step out and tell me:
- Where I was weakest
- What I never actually answered
- The one sentence that would have worked better than what I said
Paste into ChatGPT, Claude or Gemini. Replace every [bracketed] placeholder with your own details first, then review the edited prompt before you send it.
Follow-Up That Doesn't Feel Automated
20 min
OutcomeBuild a follow-up cadence people answer.
Most money is lost in follow-up, and most follow-up fails for one reason: it's about the sender. "Just checking in," "circling back," "wanted to bump this" all announce that you want something and are offering nothing.
The fix is a rule: every follow-up carries something. An article that's actually relevant to what they said. An answer to a question they raised. A short observation about their situation. If you have nothing to carry, you're not ready to follow up.
AI's role here is drafting the specific version at speed. The reference point comes from you, your notes on the conversation, and the model turns it into three sentences that sound like you noticed. That's a two-minute job that most people skip and then blame the prospect.
A cadence that works without being oppressive: day 2, day 7, day 21, then quarterly. And a clean exit at the end: "If the timing isn't right, tell me and I'll stop." People respect it, and it clears your list of ghosts.
Drill20 min
Take your three oldest un-followed-up conversations. Write one carrying-something follow-up for each. Send them today. This drill has the shortest path to results of anything in this program.
Prompt: Follow-up message
Write a follow-up message.
WHAT THEY SAID: [your notes: their situation, concerns, exact words if you have them]
LAST CONTACT: [when, and what happened]
WHAT I'M CARRYING: [the article, answer, or observation. If this is blank, tell me I'm not ready to send anything]
RULES:
- Under 80 words
- Lead with the thing I'm carrying, not with me
- Reference something specific they said, in their words
- No "just checking in," "circling back," "bumping this," "hope you're well," or "touching base"
- One clear, small next step, or none at all
- Voice: [paste voice profile]
Give me two versions: one warmer, one more direct.
Paste into ChatGPT, Claude or Gemini. Replace every [bracketed] placeholder with your own details first, then review the edited prompt before you send it.
Personalization at Scale, Without the Ick
25 min
OutcomeSend fifty messages that each pass the 'would I be embarrassed' test.
You can now personalize at volume. Whether you should depends on one test: if this person saw the template and the other forty-nine messages side by side, would they feel respected or played?
Mail-merged first names fail that test. Everyone knows what a merge field looks like. What passes is personalization on something that required a human decision: which of your three offers is relevant to them, a specific thing they posted, the actual problem their industry has this quarter.
Practically that means segmenting into small groups with genuinely different messages rather than one message with slots. Five groups of ten, each getting something true about that group, beats fifty near-identical notes.
And the volume ceiling is real: if you can't handle the replies personally, you're not scaling outreach, you're generating a backlog and a reputation. Send what you can answer.
Drill25 min
Take a list you were going to blast. Sort it into three groups by what those people actually have in common. Write three messages. Notice that this is both more effective and less work than agonizing over one universal message.
Prompt: Segment instead of blast
I have a list of people to reach out to. Help me segment instead of blast.
THE LIST: [describe who's on it]
WHAT I OFFER: [briefly]
1. Propose 3-5 segments based on what these people actually have in common: their situation, not their demographics.
2. For each segment, name the specific problem that segment has and what would make my message worth answering.
3. Write one message per segment, under 90 words, in this voice: [paste voice profile]
TEST EACH ONE: if a recipient saw all the versions side by side, would they feel respected or played? Tell me which of your drafts would fail that test and fix it.
No merge-field personalization. If the only custom thing is their name, you've written a blast.
Paste into ChatGPT, Claude or Gemini. Replace every [bracketed] placeholder with your own details first, then review the edited prompt before you send it.