Forget prompt engineering. Context engineering is far more important for great content

By Peter Springett · 20 May 2026

Building AI content architectures is Bright Content's main activity. But we also get to train marketing teams on Claude, Gemini, OpenAI and others.

Most employees we speak to already use these tools as simple chatbots. One question, one task. Done. On the surface that might look fine. Employees using AI almost every day? Sounds like success.

The reality is that, yes, AI use in the business is very broad. But it's also quite shallow. And in many cases it's using up time and tokens that eat into return on investment and productivity.

So what's going wrong with AI content and how do we fix it?

Most people are focused on the wrong thing. They've learned to write better instructions with clearer questions and better structure. That's prompt engineering, and it matters. But it only gets you so far if you're starting from scratch every time you open a chat.

Context engineering is different. It's about what you bring to the conversation before you ask a single question. This includes background, project knowledge, the persistent information that means the AI already understands your goals before you type a word.

Most people have figured out the basics of AI prompting. They also know how to attach files to a chat. But because they start every session with a new chat, that information is lost. So they end up entering the same background information, attaching the same files every time they log on to the platform.

Imagine hosting a kick-off meeting for your team and then having to schedule it again every time you move on to the next phase of the project. That's basically how most people use their AI tools today.

Which brings us nicely to context. Imagine giving your chatbot a project memory. Including a set of persistent instructions at the very beginning alongside files and ‘memories’ that get updated with every conversation.

We also call this the knowledge base or even the ‘second brain’ of the project. It’s not a database (too structured) nor a centralised folder (too unstructured and passive). Instead, the platform draws on files that you upload once and once only. You can even point it (Cowork) to an existing folder on your laptop.

But here’s the thing that most people miss. Poor context produces generic content. When AI has no background about your brand, your voice or your audience, it fills the gap with its best guess. And its best guess is an average of everything it’s ever been trained on. Which means average content that anyone could have written.

Context: the critical ingredient for great content

Context is what separates output that sounds like you from output that sounds like a machine imagining what you might say. Prompt engineering can sharpen a question. It can’t fix the fact that the AI knows nothing about who’s asking it. The businesses producing AI content that actually sounds like them aren’t just prompting better. They’re engineering context.

This is where we can help. Context is a skill that you can learn through trial and error. But it takes time and you could end up with employees using incompatible techniques to build persistent project memories.

For the sake of a one-hour session we can teach your team the basics and ensure that everyone’s reading from the same hymn sheet. Which is another way of explaining the importance of context of course.

Put it another way. If you’re not applying context properly, we estimate that you’re wasting up to 90% of your investment in Google Workspace (Gemini), Claude or ChatGPT. No wonder so many businesses (and individuals) say that they’re not getting ROI from their AI investment.

If you’d like to find out how to produce AI content that actually sounds like you, and stop wasting your investment in the process, get in touch today.