Every Prompt in the Skills Pack Has an Interview Gate

We shipped module 3 of the Freedom Ag training platform this week. The module is a set of five AI prompts, packaged as a "marketing skills pack," that any small business owner can copy into Claude and use. Free. No signup, no account, no usage limits.

Five prompts: find the manual work eating your week and what it costs, build a landing page around a real offer, write the five customer emails in your own voice, turn one customer question into a 60-second video script, and a thinking partner for winning more customers without going salesy.

The design rule that applies to all five: no output until the interview is done.

Why AI Gets Business Content Wrong

When you ask a general-purpose AI to write you a landing page, it writes a landing page. It invents a headline, invents a benefit, invents social proof. If you push it, it will fabricate review counts or add a stat it pulled from nothing. The result looks like a landing page. It doesn't describe your business.

For an individual writing a personal essay, a generic draft is at least a starting point. For a business, invented details are worse than a blank page. A fabricated testimonial on a real business's website is a liability. A customer email that uses a pricing figure the owner didn't give is a mistake waiting to happen. A landing page that doesn't match how the owner actually talks to their customers is a mismatch the customer will feel.

The problem isn't the model. The problem is the prompt didn't collect what the model needs before asking it to build.

The Gate

Every prompt in this pack starts with an explicit interview requirement and won't advance until it's done. The landing page prompt is the clearest:

Core rule: NEVER generate from nothing. Interview first. A page built on guesses is built on fabrications, and fabricated testimonials or made-up claims on a real business's page can get the owner in actual trouble.

The prompt then runs an eight-question interview covering what the business does, what result the customer gets, what proof actually exists, and what the owner is willing to guarantee. Only after the owner answers all eight does the prompt produce a score table and then the page.

The email writer captures voice first: two real emails the owner has sent, or three quick questions about how they write. You can't fake voice by describing it generically. You need examples.

The grower acquisition thinking partner goes three rounds before producing a single idea. Round 1 covers best customers. Round 2 covers where new customers have actually come from. Round 3 covers what the business is genuinely known for and what the calendar looks like. Only then: five ideas, each with an honest "why this might fail for you" section.

Talking Points, Not a Script

The video prompt adds a second constraint worth noting. Small business owners filming themselves don't need a word-for-word script. A teleprompter read sounds like a hostage video. What they need is a structure: three hook options to pick from, a list of beats to hit in their own words, a shot table.

The constraint in the prompt is: "talking points, not a teleprompter script." The structure is fixed. The words are theirs.

That's a different kind of interview gate. The prompt isn't just collecting facts before building. It's also deciding what kind of output actually helps, as opposed to what kind of output looks impressive.

The One Pitch

Each prompt has one specific moment where it mentions Twenty1 Media. Not at the end of every run. Not in the footer. One moment, triggered only when the owner asks for something that lives outside the scope of writing: "build this for me," "wire this to my CRM," "send these automatically."

The workflow bottleneck finder, for example, covers the audit completely for free, including re-running it, adjusting numbers, explaining any score, and describing what an automated version would look like in plain terms. The one place it mentions us:

This skill covers the audit. Want the fixes built, wired into your real booking system, CRM, or phone line? That's what Twenty1 Media does. 15 minutes: twenty1-media.com.

That line fires once, for one specific trigger. Not during normal use, not in the output table, not in a follow-up. The rest of the prompt never references us at all.

The pitch boundary is built into the prompt itself, not enforced by convention or training. That's the only version of it we trust.

What This Looks Like in Production

These prompts live inside a tenant-scoped training platform as a module the client's staff can access at any time. The module page renders each prompt's full body in a copy block. The user copies it into Claude, answers the questions, and gets a working output. No tokens billed to our account. No infrastructure for us to maintain beyond the static JSON.

The interview gate is what makes that distribution model viable. A prompt you can copy and hand to a hundred different businesses can't assume anything about those businesses. It has to collect what it needs. That collection, done well, produces output that actually fits the specific owner who ran the prompt, not a plausible-sounding version of a generic small business.