Your AI gives you generic answers because your business is undocumented
If your AI keeps producing bland marketing output, the problem usually isn't the model and isn't your prompting. It's that what the AI would need to be useful (your positioning, your point of view, what your customers actually struggle with) has never been written down. The model can't use what doesn't exist. AI didn't create that gap. It just made it impossible to ignore.
Key takeaways
A capable model with no context gives you average output, because average is all it can infer from a blank brief.
The bottleneck in most businesses isn't AI capability. It's undocumented knowledge that lives in people's heads.
"Context rot" is real. Dumping everything into one long prompt makes output worse, not better. Structure beats volume.
Startups and scaleups are the most exposed, because they've moved fast and written little down.
The fix isn't a better tool. It's the documentation work you probably should have done anyway.
Why does my AI give such generic marketing advice?
Because the brief was generic, even if you didn't mean it to be. You open a chat window, you ask it to write a one-pager about your product, input on your planning doc, or research events that match your ICP, and you've effectively asked a stranger to speak for your business with nothing to go on. It doesn't know how you're positioned, your points of view, or what your buyers pain points are. So it gives you the average of everything it has ever read.
What usually happens next is people decide the AI isn't much good, and go hunting for a better model or a cleverer prompt. I'd look somewhere else first. The model is fine. The brief is empty.
Isn't this a prompting problem?
Better prompting helps, but there's a ceiling on what phrasing alone can do, and most people hit it sooner than they expect. You can ask a perfectly good question and still get a hollow answer back, because the knowledge needed to answer it well was never in the room.
Early on, everyone obsessed over prompt engineering; the cleverest way to word the instruction. The attention now is on what Anthropic and others call context engineering: getting the right information in front of the model, structured well, before you ask it anything at all. And the reason for the shift is the part worth sitting with. Even a weaker model does well with the right context, and the best model on the market can't rescue a poor one. So if you're about to pay for a premium tier to fix output you don't like, the model probably isn't your problem.
So I should just paste everything I have into the chat?
In short, no. Once you accept that context matters, the instinct is to shovel everything into the chat, but that’s often too much and can muddy the waters.
Models have a finite attention budget, and it frays as you fill it (hence why you’ll get Claude ‘compacting our conversation so we can keep chatting’ on longer chats). There's a name for the effect now, context rot: as the number of tokens climbs, the model's ability to accurately pull the right detail back out goes down. So you bury the one thing you needed under a pile of things you didn't, and the model has to dig through all of it. The skill isn't assembling the biggest pile. It's finding the smallest set of genuinely useful information that gets the job done. Think ‘more structured’ rather than just ‘more’.
Why are startups and scaleups hit hardest?
Because growing companies have a tendency to move fast and write things down late. The positioning lives in the founder's head, sales' head, marketing’s head (and often different variants). The real reason customers buy sits in a sales lead's gut, never typed out anywhere. The brand voice is "you'll know it when you see it." None of that was a problem when ten people shared one room and picked it up by osmosis.
Then you hand the work to an AI, and the AI can't sit in the room. It only knows what's been made explicit, and almost nothing has. So the output sounds like a competitor, because everything that makes you different was never captured anywhere a model could reach it.
How do I fix it?
You do the valuable work of writing things down. Your positioning. Your point of view. Who your buyer really is and what they're actually wrestling with. The language that lands, and the language that makes them switch off. Not as one giant document you ram into every prompt, but as structured, separable knowledge the AI can draw the right piece from, exactly when it needs it.
In all honesty, you're not doing this for the AI. A business that can't articulate its own positioning has that problem whether or not a chat window is open. The AI just dragged the gap into the light and gave you a reason to close it. The companies getting real work out of these tools - using AI as a thought partner and an augmenter - aren't the ones with the cleverest prompts. They're the ones that did the thinking, wrote it down, and shared it with their teams and their tools.
That's more than one post can hold, so I've split the rest in two. Next week: what this actually costs you in time, tools and structure, and how to tell if it's working. The week after: the practical build, which documents to create, and the order that makes them compound.
FAQ
Is the latest, most expensive AI model worth it for better marketing output?
Usually not, if "generic" is your complaint. The newest models help on reasoning-heavy work, but bland output is almost always a context problem, not a capability one. A mid-tier model with good context beats a top-tier model with none.
What is context engineering?
Curating the right information for an AI to use before you prompt it (your positioning, customer insight, brand voice and so on) structured so the model can pull on the relevant part. It's the step on from prompt engineering, which only ever dealt with how you word the instruction.
What is context rot?
The observed effect where an AI's accuracy drops as you cram more into its context window. It's why pasting everything into one prompt backfires. The model has a limited attention budget, and noise crowds out the signal.
Does this apply to small businesses, or just big ones?
Most of all to fast-growing small and mid-sized companies, because they tend to have the least written down. The less documented your knowledge, the more generic your AI output.
Where do I start if almost nothing is documented?
Positioning and customer pain, in that order. They move AI marketing output more than anything else, and they're usually the two least documented things you own. The rest of this series walks through the full sequence.
