What this guide is
A creator named Bruno Souza posted an Instagram carousel claiming "Learning Claude right now is like buying Bitcoin in 2017." His hook: 5 prompts that supposedly built him a business making $50K+ per month selling AI services to local companies.
The hook is hype. The structure is solid. We took his 5-prompt skeleton, rewrote each prompt to be specific and constrained, and added the honest stuff he skipped. If you want to start an AI services business for local SMBs, this is the version that works on a Tuesday morning instead of in a slide deck.
What you will get from this guide:
- 5 production-grade prompts you can paste into Claude or ChatGPT today
- The honest math on what an AI services business actually pays at different client counts
- The 4 ways the business model breaks, and how to spot each one early
What you will NOT get:
- Promises about $50K/mo "passive income"
- A magic niche nobody else has thought of
- Anything that sounds like an AI-bro side-hustle pitch
The 5 prompts, rewritten
Prompt 1: Find a niche worth pursuing
The original asked for 10 business types AI could "save 20+ hours per week." That output is too vague to act on. Better:
Help me identify 10 local SMB business types in the US that are good targets for an AI services agency in 2026. For each, output:
- Business type
- Single most common back-office bottleneck (a specific workflow, not "operations")
- Why this niche is underserved by AI today (one sentence)
- Realistic monthly time saved if AI handled that workflow (low-end and high-end estimate, with units)
- One reason this niche might be HARD to win (compliance burden, vendor lock-in, owner skepticism, low budget, regulated industry)
Bias toward businesses with 5 to 50 employees, transactional service work, and owners who are NOT digital natives. Skip industries where AI is already saturated (DTC ecommerce, marketing agencies, software companies).
Format the output as a markdown table.The "one reason it's hard" column is the most important one. Bruno's version implies every niche is winnable. Most aren't.
Prompt 2: Draft a first-touch proposal
The original asked for a "short simple proposal" promising "20+ hours per week saved." Vague promises produce vague responses. Better:
Draft a first-touch proposal email to a [NICHE] business owner introducing AI services. Constraints:
- Under 200 words total
- Lead with one specific operational pain that's true for this niche right now, not a generic "save time" pitch
- Reference 2-3 concrete AI tools by name with what they do and roughly what they cost
- Include one credible outcome with a RANGE, not a specific dollar number ("typically saves X to Y hours per week" not "saves you $X")
- End with a 15-minute call invite with a calendar link placeholder
- Tone: direct peer-to-peer, not vendor-to-prospect
- Banned words: leverage, unlock, transform, harness, journey, ecosystem, seamless, robust, comprehensive, revolutionary, cutting-edge
Output the email body only. No subject line, no signature.The banned words list is the trick. Without it, Claude defaults to LinkedIn-bro voice and the proposal reads like every other vendor pitch the owner has deleted.
Prompt 3: Build a discovery call playbook
The original asked for "5 questions to uncover their biggest time-wasters." That gets you generic questions. Better:
Build me a 30-minute discovery call playbook for a [NICHE] owner. Output four sections:
1. Eight open-ended questions, ordered from least sensitive to most sensitive (general operations → which tools they hate → hours per week per task → loaded hourly cost of the team doing each task).
2. For each question, what I should LISTEN FOR (specific signals that mean "good fit") and what I should LISTEN AGAINST ("not fit" signals). Be concrete: signals like "they already pay $500+/mo for one SaaS tool" or "the owner answers customer emails personally past 7pm".
3. A simple table I can fill DURING the call: task name | current hours per week | loaded hourly cost | AI-handled estimate (low/mid/high) | monthly savings range. Show me a worked example for one row.
4. Two close scripts: one for "this is a good fit, here's the next step" and one for "this is not a fit, graceful exit." Be honest in the not-fit script — no fake-friendly closeout.The "what to listen against" piece is what separates a real consultant from a sales bro. Most prospects aren't a fit. The faster you can hear that on a call, the more time you save.
Prompt 4: Spec the AI tech stack
The original asked for "the best AI agents for copywriting, support, scheduling, and email marketing." That gets you a list of tools without context. Better:
Build me an AI tech stack recommendation for a [NICHE] business. Cover these workflows:
- Customer service first-draft replies
- Appointment scheduling / booking
- Email marketing (campaigns AND transactional)
- Documentation / SOP writing
- Customer follow-up cadence
For each workflow, output:
- Tool name + actual monthly cost in 2026 (real numbers, not "varies" or "depends")
- What a non-tech owner sees in the day-to-day (one sentence)
- What runs in the background (one sentence)
- One specific risk or limitation that an honest pitch would mention (e.g. "the AI replies sound generic without 2 hours of voice training" or "the booking tool double-books if Google Calendar isn't synced")
- Setup time estimate (hours)
Skip enterprise tools that an SMB can't afford or won't trust. Bias toward turnkey SaaS over custom builds for the V1 stack. End with a one-paragraph note on what to add in V2 once V1 is stable.The "one specific risk" column is what makes the proposal land. Owners trust someone who flags problems before signup.
Prompt 5: Design the retainer offer
The original asked for a "$1,500/month AI retainer offer" and how to "stack 10 clients to hit $20K/mo passive income." The "passive" framing is the biggest lie in the deck. Better:
Help me design a monthly retainer offer for AI services to [NICHE] businesses. Constraints:
- Three pricing tiers (low / mid / high) with 2-3 specific deliverables per tier
- Each tier shows what's INCLUDED and what's EXPLICITLY EXCLUDED. Examples of exclusions: "custom code," "after-hours support," "more than 2 revisions per deliverable"
- Each tier shows the realistic monthly hours of MY TIME required to deliver. NO "passive income" framing — be honest about hours
- Include the math for how many clients per tier I'd need to hit $20K/mo, $40K/mo, and $80K/mo gross
- Include a column showing per-client gross margin assuming I subcontract 50% of the work
- End with 5 ways this offer can BLOW UP (scope creep, late payment, churn at month 3, undelivering on time savings, niche compliance failure) and how to mitigate each
Output as a markdown table for the tiers, then a separate list for the failure modes and mitigations.The failure-modes section is the most useful output of all 5 prompts. Bruno's version pretends those failure modes don't exist. They do.
The honest math nobody on Instagram shows you
Here is what an AI services business actually looks like at three client counts, assuming an average $1,500/mo retainer:
5 clients = $7,500/mo gross. Probably 50-60 hours/month of your time. After tools (
$400/mo) + subcontractors ($1,500 if you outsource) + taxes you're at $3,500-4,500/mo net.15 clients = $22,500/mo gross. 90-120 hours/month of your time, MINIMUM. You've now hired one part-time person to handle delivery. Net is roughly $9,000-12,000/mo, which is great, but it is not "passive."
25 clients = $37,500/mo gross. You're now running an agency. Two W2 hires or one full-time + two contractors. You spend most of your week on sales calls and ops, not delivery. Net depends entirely on how well you priced — somewhere between $12,000 and $22,000/mo.
The Instagram screenshot showing $54,000 in one month is real. What's missing: the costs underneath, the months it took to ramp, the churn on the back end, and the fact that gross is not income.
This is a real business. It can pay you very well. It is not passive, and it is not magic.
What to do with these prompts this week
If you are serious about starting an AI services business, here is the order:
- Run prompt 1, pick ONE niche from the output. Not three. One.
- Run prompt 4 for that niche, get the V1 stack, install one tool yourself this week so you actually know how it works
- Run prompt 2, send the proposal to 5 owners in your network (not cold outreach yet — start with people who already trust you)
- Run prompt 3 for any owner who responds. Do the discovery call.
- If a discovery call goes well, run prompt 5 to scope the retainer offer for that specific owner
Don't run prompts 2-5 for niches you haven't validated with prompt 1 first. The biggest mistake new agency operators make is picking a niche based on what they read online instead of what they can actually win in their geography.
If you want help walking through any of this, the scoping call link is free. Thirty minutes, no slide deck, just a conversation about whether the AI services business model is a fit for you specifically.
Let's talk about your AI + SEO stack
If you'd rather skip the how-to and have it shipped for you, that's what I do. Start a conversation and we'll figure out the fastest path to results.
Let's Talk