There’s a common assumption that AI is replacing the way businesses create content.
That’s exactly where we started. When preparing to launch our new website early 2026, and realising that we hadn’t written a blog in over two years, we needed to take a different approach and build an AI content creation process.
“We are problem solvers, not content authors. We know what we are talking about but can’t always articulate that neatly.”
Being in business 22 years, we could be considered subject matter experts on IT. But when choosing between writing blogs to support our marketing, or, helping clients with their technology – we would mostly choose the latter as it’s our comfort zone.
The real problem wasn’t AI – it was consistency
When we started rebuilding our website, it became obvious that content wasn’t just “a marketing task” sitting on a backlog somewhere.
It was a structural issue.
- We had real expertise internally
- We had clear opinions on IT, cyber, and AI
- We understood our clients well
…but we weren’t publishing anything meaningful.
In fact, over a two-year period, we had effectively published nothing.
That’s not uncommon. Most small to mid-sized businesses don’t have a content problem – they have a translation problem.
The knowledge exists. The output doesn’t.
This isn’t “AI writing blogs”
There’s a lot of noise around using AI to generate content, and a lot of AI slop out there.
That’s not what this is.
“AI writes blogs for our audience, based on our grounded data, knowing who we are, what we can do, and what our audience actually cares about.”
The key difference is how it’s used.
Instead of prompting AI to “write a blog”, we built a structured process that:
- Understands who we are as a business
- Knows our audience and what resonates with them
- Adapts tone and content based on where someone sits in the buying journey
- Challenges ideas instead of validating them blindly
- Checks against existing content to avoid duplication or mixed messages
It can interview us like a journalist, or take a rough idea and shape it into something publishable.
But critically – it’s not just generating words.
It’s working within a defined system.
The important part: what it’s grounded on
The difference between useful AI and generic output comes down to one thing:
What it’s grounded in.
This system isn’t relying on internet guesswork.
It’s built on:
- Internal audience profiles
- Defined personas
- Our own service capabilities and positioning
- A structured understanding of the market
- Cross-referenced public information to validate ideas
“It knows what ‘good’ looks like because it’s been taught on your own information.”
Without that grounding, AI fills gaps with assumptions. With it, the output becomes relevant, consistent, and usable. This is called retrieval augmented generation, or RAG for short.
Designed deliberately – not just “trying AI”
This didn’t start as a polished system.
It evolved.
Initial experimentation quickly gave way to something more structured – because unstructured AI use breaks down fast in a business environment.
So we built it properly:
- Defined what it should do (interview, draft, challenge)
- Defined what it must not do (publish, make decisions, invent information)
- Aligned it to a clear governance framework
- Built in Copilot Studio using the Anthropic Claude Opus 4.6/4.8 model
- Built it as an Advisor-style agent – assisting, not acting
The principle underneath is simple:
AI supports people. It doesn’t replace responsibility. This is core to our AI governance philosophy and amplifies our team, helping IT experts author complex, technical concepts into regular language.
What we deliberately prevented
Working responsibly with AI isn’t about what it can do.
It’s about what you stop it from doing.
This system:
- Does not publish content
- Does not invent facts or fill gaps with fiction
- Does not default to agreeing with everything
- Does not operate without human review
- Does not use sycophantic language to support bad ideas
“If AI is too agreeable, you end up doing stupid things.”
Every output is reviewed, verified, and owned by a person before it’s used.
The role of conversation (and why it matters)
One of the most valuable parts of this system is the interview approach.
Instead of prompting in isolation, it actively pulls information out through conversation.
That’s where the real expertise shows up.
“Sometimes the best ideas come from conversation – it helps capture our natural tone and authenticity.”
This is what bridges the gap between:
- generic AI output
- and something that actually sounds like your business
It’s not perfect (and that’s important)
This isn’t a “set and forget” tool.
It still needs oversight.
Sometimes it:
- overuses bullet points
- feels too rigid
- pushes structure too hard
We need to rein it in from time to time, and sometimes it gives us garbage.
That’s expected, and that’s why Human-in-the-Loop is so important. Real judgement and ownership still sits with people.
What this actually proves
This article isn’t really about blog writing.
It’s a simple, visible example of something bigger. It’s an example of agent design with a goal in mind and solid knowledge of what data is useful.
Behind this sits a repeatable approach:
- Assess the use case
- Define the outcome
- Structure the behaviour
- Ground it in the right data
- Apply clear governance
Why this matters for your business
Most businesses are currently in one of two positions:
- Reckless and irresponsible implementation of AI systems
- Paralysed by uncertainty into safety and long term risk
They both camps are have a fear of being left behind, and are unclear on how to enable AI with control and responsibility. What’s missing is alignment, structure, and governance. Organisations are chasing tools without a clear plan or outcomes in mind.
In our example here, we’ve applied strong AI governance and agent design, knowing what we’ve wanted to achieve at the start. This shifted us from:
- two years of no meaningful content
to:
- consistent, targeted articles with clear purpose and audience
The takeaway
“It’s not difficult to move from AI experimentation to useful, impactful tools – it just needs structure.”
AI doesn’t need to be chaotic or risky. It just needs clear boundaries, the right inputs, and human accountability.
Our approach is to deliver that structure with our Managed AI Services – alongside the technical capability to make AI genuinely useful.
This blog article was created with assistance from our AI Marketing bot; we’re IT experts, not content authors.
