02 // THE AUTHENTICITY PROBLEM
Here's the problem nobody talks about.
AI can generate 1,000 video scripts in an hour. Great. 999 of them sound like AI. And users can tell. And Facebook can tell. And your CPMs go up because the algorithm knows nobody wants to see another generic "game-changer" ad.
The issue isn't output volume. It's output quality. Specifically, authenticity.
Why AI Outputs Sound Like AI
AI models are trained on internet content. Billions of pages of blogs, articles, marketing copy, product descriptions. The result is a machine that's really, really good at producing text that sounds like... the internet.
That's the problem.
When you ask Claude or ChatGPT to "write a video ad script," it pulls from every ad script pattern it's ever seen. And most ad scripts on the internet are mediocre. So you get a mediocre average of mediocre inputs.
Specifically, AI defaults to:
- Polished over real. Perfect grammar, perfect structure, zero imperfection. Humans don't talk like that. Especially not in a selfie video.
- General over specific. "Many people struggle with weight loss" instead of "I spent six years doing keto, Whole30, all of it." Vague = AI. Specific = human.
- Corporate over conversational. "Game-changing solution" instead of "this actually works and I'm kinda mad nobody told me sooner."
- Feature-dumping over storytelling. Lists of benefits instead of a real experience. Nobody watches a 30-second ad that reads like a product spec sheet.
The fix isn't better AI. It's better prompts. You have to tell the model exactly what authenticity looks like, because it won't figure that out on its own.
The Cliches That Kill Performance
You've seen these a thousand times. Your audience has too. Here's the kill list:
- "Are you struggling with X?"
- "Game-changing solution"
- "Transform your life"
- "Let me tell you about..."
- "No more [problem]. No more [problem]. Just [solution]."
- "Real results"
- "Join thousands of satisfied customers"
- "Start your journey today"
- "What if I told you..."
- Perfect grammar throughout (no contractions, no stutters, no corrections)
Every single one of these is an AI tell. When a user sees them in their feed, their brain pattern-matches to "ad" and they scroll. When Facebook's classifier sees them, it flags the creative as low-quality generated content.
The irony: most affiliate marketers are using AI to produce the exact copy patterns that AI detection systems are trained to catch.
How Platforms (and Users) Spot AI Content
Facebook's side: Facebook has content quality classifiers running on every ad. They're not just checking for policy violations. They're scoring creative quality. AI-generated content that follows generic patterns gets lower distribution and higher CPMs. Facebook wants authentic content in the feed because that's what keeps users on the platform.
Your "game-changing solution" ad isn't getting banned. It's getting deprioritized. Which is worse, because you don't even know it's happening. Your CPMs just quietly climb and you blame the offer.
The user's side: Your target audience has seen hundreds of AI-generated ads this week. They've developed pattern recognition whether they realize it or not. The moment copy feels "off" (too polished, too structured, too perfect), they scroll.
Think about it from their perspective. They're sitting on their couch, scrolling Instagram. They see a video that opens with "Are you struggling with weight loss?" They've heard that exact phrase from 47 different ads this month. Their thumb keeps moving.
Now imagine they see: "Okay so... I need to tell you something about these GLP-1 injections that's probably gonna piss off every diet company in America."
Different reaction. Because it sounds like a person, not an algorithm.
The Compliance Tightrope
Here's where it gets tricky for affiliates.
You can't be too aggressive. Compliance violations get your account shut down, your offer pulled, or worse. Especially in regulated verticals like health, finance, and insurance.
You can't be too vanilla. Generic, safe copy doesn't convert. If your ad sounds like a pharmaceutical disclaimer, nobody's clicking.
AI defaults to vanilla. When you ask it to write health copy, it hedges everything. "May help with weight management." "Consult your healthcare provider." "Results may vary." All technically correct. All completely useless for performance.
The skill is navigating the grey area between compliant and compelling. And that's a prompting problem, not an AI problem. You need to teach Claude where the line is, what's on each side of it, and how to write copy that lives right on the edge.
We'll cover this in detail in Section 08. For now, just know: the compliance challenge is solvable with better prompts.
GLP-1 Example: Bad AI Script vs. Good AI Script
This is the best way to show you what we're talking about. Same offer. Same audience. Same AI model. Completely different prompts. Completely different outputs.
The Bad Version (Generic Prompt, Generic Output)
Prompt used: "Write a 30-second video ad script for a GLP-1 weight loss offer."
Output:
"Are you struggling with weight loss? Tired of diets that don't work?
Let me tell you about GLP-1. It's a game-changing solution that's helping
thousands of people finally reach their goals.
No more yo-yo dieting. No more feeling stuck. Just real results.
Click the link to learn more and start your transformation today!"
What's wrong with this:
| Problem | Why It Fails | |---------|-------------| | "Are you struggling with weight loss?" | Cliche opener. Every AI ad starts this way. Instant scroll. | | "Game-changing solution" | AI tell #1. Nobody talks like this in real life. | | "Helping thousands of people" | Vague social proof. No specificity. Unbelievable. | | "No more X. No more Y. Just Z." | AI pattern. Readers have seen this structure 100 times. | | "Real results" | Means nothing. What results? For who? When? | | "Start your transformation today!" | Pushy CTA. Screams "ad." | | Perfect grammar throughout | No human speaks in complete, polished sentences on camera. | | Zero personal detail | No name, no timeline, no specific struggle. Could be for any product. |
This script would get eaten alive in a competitive auction. High CPM, low CTR, terrible ROAS.
The Good Version (Engineered Prompt, Authentic Output)
Prompt used: A full prompt with role assignment, context loading, output formatting, authenticity markers, and constraints. (We'll break this down completely in Section 03.)
Output:
"Okay so... I need to tell you something about these GLP-1 injections
that's probably gonna piss off every diet company in America.
I spent six years doing keto, Whole30, all of it. Lost the same 20 pounds
four different times. And I'm not gonna lie, I felt like an idiot.
Then my doctor, like my actual doctor, not some Instagram guru, told me
about semaglutide. And I was like... wait, is this the Ozempic thing?
Here's what nobody tells you. The reason every other diet failed wasn't
because I lacked discipline. It's because my hunger hormones were literally
working against me. GLP-1 fixes that. Not by willpower. By biology.
I'm down 30 pounds in 12 weeks and I'm not white-knuckling it. I'm
just... not as hungry. Which sounds simple but it changes everything.
Link in bio if you want the details. This isn't for everyone but if
you're tired of losing the same weight over and over, you should
probably know this exists."
Why the Second Version Works
| Element | Technique | Effect | |---------|-----------|--------| | "Okay so..." | Hesitation opener | Sounds like someone actually talking, not reading a script | | "gonna piss off every diet company" | Controversial hook | Creates an enemy, builds curiosity | | "six years doing keto, Whole30, all of it" | Specific diet names | Audience sees themselves in the story | | "Lost the same 20 pounds four different times" | Specific failure pattern | Relatable. Most dieters have lived this exact cycle. | | "felt like an idiot" | Self-deprecation | Vulnerability builds trust. AI never admits weakness. | | "my actual doctor, not some Instagram guru" | Conversational aside | Establishes authority while being funny | | "wait, is this the Ozempic thing?" | Acknowledge awareness | Doesn't insult the audience's intelligence | | "hunger hormones were literally working against me" | Mechanism reframe | Removes shame, explains why diets fail | | "I'm just... not as hungry" | Understated result | More believable than "transformed my life" | | "This isn't for everyone" | Qualifier | Increases credibility by self-selecting | | "Link in bio" | Soft CTA | Non-pushy, lets the interested self-select |
The Side-by-Side
| Element | Bad AI | Good Prompt | |---------|--------|-------------| | Opening | Cliche question | Hesitant, controversial hook | | Tone | Polished, perfect | Conversational, raw | | Specificity | "Thousands of people" | "Six years, 20 pounds, four times" | | Authority | Implied ("thousands") | Explicit ("my actual doctor") | | Mechanism | None (just "results") | Biology explanation (hunger hormones) | | CTA | "Start your transformation today!" | "Link in bio if you want details" | | Authenticity markers | Zero | Hesitation, self-deprecation, asides, specifics |
Same AI model. Same offer. The only difference is the prompt.
That's what this guide teaches you to build.
The difference between these two scripts isn't "better writing." It's understanding what makes human communication feel human, and knowing how to prompt for it.
The building blocks for that are in the next section.