Heads up! Starting in March all of my writing is going to be on Substack. You don’t have to do anything to keep getting my essays on AI, tech, and brand marketing… the emails will just be coming from Substack.
You’ll still get regular free newsletters, but I’m also writing these long form essays and tutorials on AI tools and strategy. Last week I wrote about Claude Code and how it’s changing the way all my clients and friends do their work. Anthropic just released an even simpler version dubbed Cowork—and I know this sounds cliched… but if you’re not using one of these tools you’re falling behind.
Sign up as a paying member and get 50% off your first year
Ok, now on to the newsletter!
A couple of weeks ago I read a whitepaper that blew my mind.
Researchers trained AI models on a steady diet of viral posts. Clickbait, outrage bait, low-information content.
Then they tested the models.
The results: worse performance on logic questions, lost the thread in longer prompts, started taking shortcuts while still sounding confident.
Then they tried to "clean them up" with better data. The models improved, but never fully recovered.
Here's what this means for you: AI becomes what it eats.
The Invisible Problem
Most founders don't realize they're slowly poisoning their AI.
They feed it vague prompts, accept mediocre outputs, and never curate the inputs. Then they wonder why everything sounds generic and “sloppy”.
Your prompt history is just another diagnostic tool. It shows where you put your attention and focus. What you actually prioritize versus what you pretend matters in meetings. And whether you can deconstruct hard problems into clear instructions.
When you type "give me marketing ideas" you're confessing you have no framework. No angle. No real strategy… and the AI responds accordingly.
AI is a Parrot
Let me be direct: AI doesn't think. It reflects.
If you feed it templates, it gives you templates. If you feed it emotional intelligence and cultural nuance, you get outputs with soul.
Most people treat AI like a magic box. Type something. Hope for gold. Get mediocrity. Blame the tool.
That’s the wrong target.
The competitive advantage isn't a fancy prompt. It's a clean information diet.
Imagine training a new employee using only viral hot takes. They'd get really good at confidence, punchlines, and attention. And really bad at deep thinking, careful reasoning, and staying consistent.
That's what you're doing to your AI when you don't curate what goes in.
Build a Taste Stack
The businesses, brands, and people who are going to win do things differently. They build what I call a Taste Stack. Here’s what I’ve been helping my clients put together the last few months.
Reference Library: Tagged, ranked, annotated examples of what "good" looks like in your domain.
Reusable Briefs: Context plus constraints. Not "write a post" but "write a 200-word contrarian take on [topic] for B2B founders, ending with a sharp question. Here are 3 examples of hooks and CTAs that I have liked and written in the past.”
Edit Checklists: Your non-negotiables. The quality bar that separates you from the slop.
Model Presets: Your standards, baked in. Voice. Tone. Structure. Decision-making style. All of that needs to go in the project knowledge and/or custom GPT you’re using.
Before you ship anything, run the Parrot Test:
- Signature: Can you spot your fingerprints in 3 seconds?
- Slop: Did you kill all the AI tells? Those em dashes are sneaky!
- Specificity: Real context, not platitudes?
- Scarcity: Uncommon references or insights?
- Stakes: A decision someone could disagree with?
Pass 5 out of 5 or keep editing.
Why This Matters Now
AI made it easier to publish but it didn’t make it easier to stand out. If anything, it made standing out and having a real opinion even harder. Because when anyone can spin up decent content in seconds, last year's "solid" now reads like noise.
The solution isn't simply to produce more content; it's about owning powerful ideas, because while content provides answers, true ideas have the power to fundamentally change beliefs.
You know this… you can feel it when you wade through social media posts. We’ve been flooded with AI-generated everything and people crave authenticity more than ever. The winners won't be the most technical but the ones with the clearest direction and defined taste.
One way I’m helping businesses with this challenge is by building prompt libraries. When this happens, they're encoding how they think and turning gut instinct into repeatable systems. This makes strategy teachable and repeatable instead of trapped in someone's head.
If you only remember one thing from this message, remember this: Your AI's information diet is your competitive moat.
Feed it garbage and you’ll get garbage outputs. Feed it clarity, context, and standards… then you’ll get compounding leverage.
I’m going to talk about this in more detail on my Substack. I’ll share the steps I go through to build a Taste Stack and prompt libraries. Subscribe here and get access to my best thinking and frameworks.
And If you want help building your Taste Stack or auditing what your AI is actually learning from you, book a call. I'll show you exactly where your information diet is breaking down and how to fix it.
Talk soon,
Pete Sena
Please take a second and hit the FWD button ▶️, to share this email with your favorite growth-obsessed peer. I want to inspire 1 million awesome minds like you.
