Why your AI content sounds like everyone else's
TL;DR: Most companies investing in AI readiness are cleaning up the measurable layer — analytics, CRM data, structured records. The interpretive layer — customer language, positioning reasoning, the specific way the business understands its market — isn't getting the same attention. It rarely has a system, a hire, or a process attached to it. And when AI gets briefed from the measurable layer alone, what comes back is organised and generic. The fix isn't better prompts. It's whether the interpretive layer was ever built.
Earlier this year, sitting in a focus group I was running as part of a research project, someone mentioned that companies they know are hiring data analysts specifically to get their data in order before working with AI.
It’s a sensible move. Clean data going in, better outputs coming out. The logic is straightforward.
But it’s also a particular version of readiness — and it raises a quieter question about which layer of the business is actually getting prepared.
Most teams, when they talk about getting AI-ready, mean the measurable layer. Analytics, CRM records, NPS scores, heatmaps, product telemetry. The stuff that lives in systems, can be queried, and presents well in a slide. There’s another layer that doesn’t get the same treatment. And that’s the one that determines whether the AI output is any good.
Why doesn’t the interpretive layer have a system?
The measurable layer is abundant in most B2B companies. Analytics, CRM records, NPS scores, heatmaps, product telemetry. Structured, queryable, presentable in a slide. When someone challenges a decision in a meeting, you can point to a number. The number is defensible. That matters.
The interpretive layer is different. Customer language — the specific words buyers use to describe the problem before they've found a solution. Positioning — the particular way this business understands its market differently from everyone else in the category. The reasoning behind product decisions that customers ask about and that nobody inside can quite articulate to a new hire.
This layer doesn't produce a dashboard. It requires interpretation, and interpretation can be contested. So it doesn't get a system. It doesn't get a hire. It gets assumed to exist because the founder knows it, or because the old positioning document is somewhere on the shared drive.
The data analyst hire is cleaning up the measurable layer. The interpretive layer stays as it was.
What does it look like when the interpretive layer gets skipped?
You see it in how decisions get made. Campaigns built on what looks plausible rather than what customers have said. Product choices justified after the fact. Messaging that reads clean but doesn't quite connect with the people it was written for.
You see it in the positioning document nobody references, because it was written from the inside out and doesn't quite match how buyers describe the problem. In the personas that got generated quickly and then quietly abandoned, because nothing in them matched how customers actually talk.
And you see it in what happens when AI enters the picture. A founder I worked with recently kept sharing AI-generated content that could have described any company in her space. Structurally clean, professionally written, completely generic. When I asked what she'd given the model to work from, it was the website and a few competitor pages. The AI had done exactly what it was asked to do. The problem was upstream.
Why doesn’t a better prompt fix it?
The most common response to disappointing AI output is a better prompt. More context, more structure, a longer system prompt with more specific instructions. It helps a little. The output improves. It’s still generic.
What most teams are diagnosing as a prompt problem is a foundation problem. The prompt is drawing from a starting point that was never properly built — no documented customer language, no clear positioning, no shared story the business actually agrees on. A better prompt from an empty foundation doesn’t produce specific output. It produces more polished generic output. The wrongness doesn’t get fixed. It gets polished.
Feed AI clean, structured data and it organises it well. It finds patterns, produces summaries, gives you averages faster than any analyst could. Feed it the interpretive layer — actual customer language, documented positioning rationale, the specific way this business sees its market — and the output is something the team can work from. Personas that describe real buyers. Messaging that sounds like the business rather than a polished version of the category average.
The companies starting to figure this out aren’t optimising prompts. They’re going back and building the foundation the model needs — talking to customers in a structured way, documenting what they actually said, building a record of the language they use that isn’t filtered through an internal assumption about what they probably meant.
Was the data ever the problem?
Clean data going into AI is a reasonable investment. But measurement has always been a substitute for understanding. AI hasn’t changed that. It’s made the substitution faster, more visible, and easier to scale.
The data analyst hire addresses one half of the readiness question. The other half — whether the interpretive layer was ever built — tends not to come up in the planning meeting. It doesn’t have a job title attached to it. It doesn’t have a system. It gets assumed.
Most teams have more data than they can use. The part that would make it useful still isn’t there.
If you want to understand what building that foundation looks like in practice, my free Research Toolkit is a good starting point.
If you’re ready to find out what’s actually missing, book a 20-minute intro call.
Frequently asked questions
Why does AI-generated content sound generic even when it's well-structured?Because the inputs it was given were generic. AI reflects back what it's fed. If the foundation is the company website, a few competitor pages, and internal assumptions, the output describes the category average rather than the specific business. The problem isn't the model — it's that the interpretive layer, the customer language, positioning reasoning, and market understanding that makes a business distinct, was never documented in a form the model could work from.
What's the difference between measurable data and the interpretive layer? Measurable data is what most companies already have: analytics, CRM records, NPS scores, product telemetry. It tells you what is happening. The interpretive layer is different — it's the specific words customers use to describe their problem before they've found a solution, the reasoning behind product decisions, the particular way this business sees its market differently from others in the category. This layer doesn't produce a dashboard, which is why it tends not to get a system or a dedicated resource attached to it.
Why do companies keep skipping the interpretive layer? Because it's harder to defend in a meeting than a conversion rate. Measurable data is structured, queryable, and presentable. The interpretive layer requires interpretation, and interpretation can be contested. Teams default to what's least likely to be challenged. Research also introduces the possibility of learning something unwelcome — that a strategy, a product decision, or a positioning assumption isn't holding up.
What does it actually mean to build better AI inputs? It means building the interpretive layer that AI needs to produce output worth using. That involves talking to customers in a structured way and documenting what they actually said in their own language. It means recording the positioning reasoning the founder carries in her head and hasn't written down. It means building a resource the whole team can draw from, rather than briefing AI from a website and hoping the output sounds specific.
How does the data analyst hire relate to AI readiness? Hiring a data analyst to clean and structure data before working with AI is a sensible move. But it addresses the measurable layer only. Most of what makes AI output distinctive and useful comes from the interpretive layer — and that layer doesn't get cleaner through data hygiene work. Companies treating data readiness as AI readiness are investing in one half of the problem and leaving the other half as it was.
What's the connection between this and founder-dependent sales? The same dynamic runs through both. In founder-dependent sales, the story lives in one person's head and falls apart when anyone else tells it — because it was never extracted and documented. The interpretive layer problem is structurally identical: the knowledge exists, usually in the founder or a few senior people, but it hasn't been built into a system the business can work from. AI makes both problems more visible by exposing exactly what's missing when the foundation isn't there.