​Read online​

Welcome to Unpacking Meaning. If you received this from a friend and enjoy it, ​subscribe here.

Why your AI outputs sound generic (and how to fix it)

There’s a concept from Tor Nørretranders’ book The User Illusion that explains why a lot of marketers get crap results with AI.

It’s called exformation.

It’s the information you deliberately exclude when communicating, so the other person can actually process what you include.

In short, good communication is knowing what to cut.

Yesterday I had forty-five minutes, on a webinar with Team GPT, to show how I go from research to strategy and copy for B2B SaaS clients. Cramming everything into that time window forced me to see something I’d been doing intuitively: you can get great results from AI only if you know what to cut at every stage.

Most people treat AI like a dumping ground

Here’s how most marketing teams use LLMs:

They open a chat, upload everything from research files, to strategy docs, to interview transcripts, to competitor analyses, and persona cards. Oh and finally that Slack thread from last week’s positioning workshop — because, why not…

Then they ask: “Write me a homepage hero.”

What comes back? Vague. Or bloated. Or weirdly generic despite all that context sitting there.

Here’s why that happens.

You’ve handed the AI the job you should’ve done yourself — deciding what actually matters. The model is trying to reconcile contradictions, average insights, and guess at what’s most important. It’s too much work all at once.

You’d never throw all of that stuff to a junior or an assitant and tell them, “go ahead and do it”.

Would you?

How I actually move context through my process

Here’s what works for me (and what my forty-five-minute constraint on the webinar made obvious).

I move context through stages and across different chats or projects. At each stage, I deliberately compress it, and decide what doesn’t need to get passed on to the next.

Stage 1: Raw research This is the widest point - Customer interviews - Sales transcripts - Support tickets - Competitor copy and analyses - Review mining - Product knowledge

And more.

Everything goes in (as long as we cover internal, external and market research).

But—and this is the key—I don’t reuse this raw material to write copy. Not even to create the strategy, yet. I use it once to build a Research & Insights Report which is the condensed, synthesized version of all the research we’ve done. This is the first step of exclusion.

Stage 2: Positioning and messaging When I move to positioning and messaging, I don’t feed the AI those fifty pages of transcripts. I give it the Research & Insights Report (and the templates it needs to create my strategy documents). That’s it.

From that, I build the Positioning Canvas. Then the Messaging Framework and the Value Proposition Canvas. Each document gets more specific and focused in its use case.

Stage 3: Copywriting When I’m writing homepage copy or a sales deck, I don’t go back to the research report, but use the Positioning Canvas, Messaging Framework, and Value Proposition Canvas as knowledge base data.

Those documents carry forward only what passed two rounds of compression.

The funnel narrows at every stage, the context gets tighter and the signal gets stronger.

Why this is harder than it sounds

Here’s what sucks for most people: excluding information feels risky, and knowing what to exclude is hard work.

What if I need that detail later? What if the AI misses something important? What if I’m cutting the insight that would’ve made the difference?

So people include everything.

Just in case.

But that’s FOMO.

Practicing exformation means making calls. You decide what matters for this specific task and trust that the rest doesn’t need to follow. You have to believe the compression itself is what creates value and clarity.

Raw transcripts are essential for extracting insights and VOC, but they’re too much when you’re drafting a headline.

Positioning matters for messaging. But when you’re writing a product page, you need the distilled promise, not the full strategic rationale behind it.

The skill is knowing what to leave behind.

What actually changes when you do this

When you compress deliberately at each stage:

The AI gets clearer instructions because you’ve done the editorial work. You’ve already decided what matters for this stage, and the model just needs to execute.

Your outputs get better because the signal-to-noise ratio is way higher.

Your process gets faster because you’re not re-litigating the research every time you write a headline. You compressed once, and now you move forward.

Which, let’s be honest, is the work we used to do before AI was a thing. We’re just picking convenience over quality now.

And it’s a shame, because doing this work is how our thinking improves. The act of compressing is the strategy work. Deciding what to exclude forces us to clarify what we actually believe.

The work you can’t automate is deciding what the AI needs to do this specific job. Not everything you know. Just what it needs right now.

DISCOVERY

My takeaways from the London Turing x SYSTM Growth Summit

If you want a deeper look at what I took away from the Turing Fest × SYSTM Growth Summit — especially the parts that matter for messaging in 2025 — I broke down the biggest insights in a LinkedIn post. It covers differentiation, alignment, JTBD, AI’s impact on research, and why most teams still treat messaging like a one-off artifact instead of a system.

You can read the full breakdown here.

Going live today 🔥

I recently joined Emilio Garcia on Demand Gen Studio to talk about how AI is transforming customer research and messaging strategy.
​
We dug into:

💡 Synthetic research (and where it actually works)

🧠 Using AI to uncover buyer psychology

🎯 Structuring research so your copy writes itself
​
​The episode drops today at 3PM GMT.​

RESONANCE

Yamamoto Tsunetomo, Hagakure

Have a great weekend!

Cheers,

Chris

Chris Silvestri Founder & conversion alchemist

🙌🏻 Let’s be friends (unless you’re a stalker)​

When you're ready, here's a few ways I can help

Not sure where to start? Take our free message-market fit scorecard.​