johnmaconline

Talking to AI is a skill.

It’s a skill you can and need to learn. You can get better with practice and iteration. You can get better by gaining a more intuitive feel for how to talk to it.  

LLMs and AI aren’t going away. They’re only getting faster, better, and more deeply embedded in the way we work, create, communicate, and solve problems.

Your competitor is using AI. Your colleagues are using AI. The content creators are using AI. 

You’ll need to use it better or more creatively. 

Because in the future, AI will either be working for you, or you’ll be working for it. 

Look at the newsfeed on LinkedIn. 

It’s 90% useless AI-generated blech. It’s carpeted with beige and unremarkable sameness. The posts start the same way. Look the same. Include the same emojis and bullets. 

But every once in a while, you come across one that hits different. Why? Because it sounds and looks like a real person created it. Therefore, it was either handwritten by a person or created by someone who knows how to utilize the LLM effectively.

In this case, effectively means “sounds like me.”

One of the most powerful things you can do with an LLM is teach it to sound like you. If you’ve written articles, newsletters, blog posts, or even just sent emails, you already have a dataset.

Feed that to the model.

The more you share, the better it gets at sounding like you. Or at least, sounding like a human.

Boss directive:

“I need to reply to a billion emails today. Help me.”

Prompt:

“You are my writing assistant. 
Attached is a bunch of email replies and other snippets that I’ve written in the past. Use these as a reference for tone, style, format, and how I say things. Adhere to this.
Now you’ve read many of my previous emails, posts, and replies. You know my voice — it’s clear, thoughtful, confident, conversational, and a little dry. I avoid buzzwords and prefer simple, precise language. Occasionally, I make a joke.  
Now write email replies in my voice for all of the attached emails. Keep it warm, appreciative, and conversational. Use my phrasing and cadence. Don’t ever make stuff up. Ask me questions if you need context or I must answer a question.”

Next up: Some final thoughts.

We’ve talked about roles, goals, structure, clarity, using more words, asking questions, and slowing it down. These are all important to help you get what you want. 

But also, it’s a conversation. So have the conversation.  

Whichever LLM you’re using, it understands the conversation. It knows the history. It remembers what you told it. Work through the problem iteratively, much like you would with a real person.

Try stuff in layers. Try different stuff. Go this way. Now go a different way. 

If you want a great result, don’t expect to get it on the first try. Instead, build in feedback cycles. Test tone, style, structure. Ask for rewrites. Push it further.

Boss directive:

“We need to figure out the right message for our audience.”

Prompt Sequence:

Round 1 – Exploration:
“You are a senior B2B marketing strategist and copywriter. Start with thinking aloud: ask me 3 clarifying questions about our audience’s biggest pain point, our unique differentiator, and what we believe success looks like (metric, emotion, action). Then generate 5 headline + sub‑headline pairs based on your responses. Each pair should vary in tone (confident, curious, direct, metaphorical, risk‑avoidant).”

Round 2 – Iteration:
“I’d like to combine versions 1 and 3. Now give me 3 more that combine those concepts.”

Round 3 – Refinement:
“Let’s go with numbers 2 and 3. For each, rewrite the headline + sub‑headline to emphasize measurable benefit in the first 3 seconds of reading. Then draft 2 versions of the short body copy (~100 words) for each pair.”

Round 4 – Iteration:
“No, I don’t like any of those. I do like the direction of the first headline, but it needs to feel more urgent. Scrap what you have and give me 3 more headlines + subs that are along the line of the first one.”

Round 5 – Iteration:
“Ok, I like 2. Now let’s go back to the draft versions of the short body copy. Give me 2 ~100 word versions of copy for this.”

Round 6 – Polish & Finalize:
“I like the first one. Now you’re the final editor. Provide the final headline + sub‑headline + final body copy. Then propose 2 alternate body copy tweaks (tone or style change). And provide a short rationale (~50 words) explaining why this version will convert best, based on audience insight and our goal.”

Round 7 – Iteration:
“I like the 2nd one, but can you pull in the style from the first?”

And so on. 

It takes patience and work from your side. Your still the boss. Be the boss. 

Next up: Use my voice.

It’s easy to treat AI like a vending machine: Insert prompt. Get result.

Fair enough. That is probably fine for many of the tasks you give it. But if you want quality and nuanced output for a complex task, you want to think more like a coach, rather than a director shouting “action!”

Great outputs come from great thinking, and great thinking is a process — observe, analyze, decide, create. AI can help at every step if you know how to coach it.

Sometimes the best approach is to use multi-step prompts. Keep the LLM focused on one thing at a time. 

Boss Directive:

“I want to understand why our retargeting is underperforming. Then fix it.”

Prompt Sequence:

1. Start with analysis.
“You are a senior performance marketer with deep experience in PPC and retargeting. I want you to take a slow, methodical approach. Analyze the performance summary below and identify the top three underperforming segments. For each, provide a thoughtful hypothesis as to why retargeting is falling short. Explain your reasoning step by step, and don’t rush to conclusions.”

2. Move to strategy.
“Based on your hypotheses, brainstorm three specific retargeting strategies. For each, explain how it addresses the problem, what assumptions it relies on, and any potential tradeoffs. Prioritize clarity and depth — I’m not looking for buzzwords, I want insight.”

3. Then create.
“For each strategy, draft three versions of retargeting ads (headline + body copy). The tone should be solution-oriented, confident, and empathetic. After each ad, include a short rationale explaining your creative choices. Focus on message-to-audience fit.”

Prompt your AI like you would coach your smartest intern. They can do it. They just need a little coaching. 

Next up: Iteration Beats Perfection

AI can be like your teenager who doesn’t want to mow the lawn. 

It’ll do it, but if you don’t guide it, it’ll rush through and do a crappy job. It’s done, but it might not be good.

If you want it done right, you’ve got to slow it down.

The same is true for an LLM. So tell it to slow down. 

Boss Directive:

“We need a new landing page because our conversion rate is falling. I’d like to see a few designs.”

Prompt:

“You are a senior SaaS marketing copywriter with deep experience crafting high-converting landing page copy, particularly for startups launching new features to existing users and prospects. In this case, we’re introducing a new time-tracking feature for our productivity tool, SlotRight, which is primarily used by freelancers and small agencies. These are busy, independent professionals who value clarity, control, and ease of use. They’re motivated by time freedom and reliability, not hype or jargon.
I want you to go slowly and methodically. First, think through what makes a great landing page design for similar products. Think about the headline and subhead. Then consider the emotional and practical outcomes this product enables. Think about how this fits into the intended SlotRight experience. Think about what kind of language best resonates with this audience.
Then, and only then, develop html prototypes for 5 different versions of the landing page, each with a slightly different angle (e.g. confidence, calm, focus, simplicity, momentum). Choose color themes that fit both the tone of the landing page and our branding. Use natural, benefit-forward language. Don’t be clever at the expense of clarity.
Take your time. Reflect before generating. If anything feels unclear, ask me questions first.”

Maybe seems weird to tell AI to slow down, but like your teenager, it’s just trying to get done. 

NOTE: Speed is also your enemy with data analysis and software development. When using AI to perform these tasks, ALWAYS tell it to be slow and methodical and use language like “accuracy is more important than speed.”

Next up: Prompt the Process, Not Just the Output

AI is a tireless brainstorming partner. That’s one of its best features.

So use it that way.

Don’t ask for “the perfect headline” or “the best CTA.” Ask for 10. Or 25. Tell it you want a whole bunch of ideas to pick from, mash together, or get inspired by.

The point isn’t to choose one and ship it. It’s to explore. When you prompt for quantity, you get diversity. And sometimes, hidden in version #8 is the one you’d never have come up with yourself.

Boss Directive:

“We need some ideas for a new blog article to be targeted at our freelance customers.”

Prompt:

“You are a content strategist and blog writer who specializes in B2B SaaS companies that serve freelancers and solopreneurs. Your job is to generate a list of strong, timely blog post ideas that speak directly to the challenges and interests of this audience.
The goal is to provide helpful, relevant content that positions our company as a trusted resource, drives organic search traffic, and engages readers enough to subscribe or share. We want posts that freelancers will actually click on, read, and feel seen by.
Please generate 40 blog post title ideas that are:

* Focused on productivity, client management, time-saving tools, financial tips, or mindset shifts

* Specific and benefit-driven (not vague thought pieces)

* Aligned with our tone: friendly, practical, and encouraging

* Suitable for freelancers who are juggling multiple clients and trying to grow sustainably

Include a 1–2 sentence explanation under each idea that explains why it would be useful or compelling to this audience. Don’t repeat themes. Don’t make stuff up. Ask me questions if you need specifics or wonder if there are some new directions I didn’t lay out that might be a good path to explore.” 

You’ll get 40 ideas from this. Many will probably suck, just as if you and your team were sitting in the conference room brainstorming. But many will likely be good. 

You can see we’re building these on all of the rules we’ve discussed so far. 

Next up: Slow and Methodical