AI Content Creator
A worked example of what good AI adoption looks like
We use AI to help produce our own content. It works because of the preparation behind it, not because of the tools. Here is the whole process, and it is the same one we run with clients, whatever they are trying to use AI for.
Why most AI content fails
The tool is almost never the problem
Point an AI tool at an organisation with no documented voice, no agreed facts and no review step, and it will produce plausible, generic and occasionally wrong material at speed. Everyone can tell. Nobody can quite say why.
Content is simply the most visible version of a problem that shows up everywhere AI gets used: the quality of what comes out depends entirely on what was prepared first.
So we did the preparation on ourselves before we did it for anyone else.
The four stages, applied
What it took to let AI near our own content
01 — Assess
AI Opportunity Assessment
We decided what AI was allowed to touch before we adopted anything.
It helps with research summaries, first drafts, turning one piece into formats for different channels, and checking readability. It does not write our point of view, our client stories, or anything that needs judgement about a real organisation. Naming that line first is what made the rest work.
02 — Prepare
Preparation & Foundations
Most of the work happened before any tool was involved.
Our tone of voice written down. An agreed set of facts about what we do and don't do, so nothing gets invented. Source material organised in one place instead of scattered across drives. A clear boundary around anything client-confidential, which never goes near a model. And time spent making sure everyone using it knew where that boundary sat.
03 — Support
Implementation & Human Support
Off-the-shelf tools, configured against prepared material. No bespoke build.
Nothing was custom-built, because nothing needed to be. For an organisation our size that is usually the honest answer, and it is the answer we give clients when it applies to them. Where a bespoke build genuinely is justified, we define the requirements and bring in trusted specialist partners, then stay involved while the change lands.
04 — Maintain
Ongoing Human Support
Every published piece is edited and approved by a person, and the source material is kept current.
AI drafts. People decide, correct, cut and sign off, and everything appears under a named author who stands behind it. When output is wrong we fix the source rather than the prompt: nine times out of ten it is an outdated fact, a missing definition or a process nobody had written down. That upkeep is the work, and it never stops.
Where we draw the line
What we won’t do, here or anywhere
The same boundaries we help clients set for themselves. They are published in full in our AI Use Policy.
Publish AI-drafted content without a person editing and approving it
Invent statistics, case studies, testimonials or client quotes
Put client-confidential information into a public AI tool
Present something as written by a person when a person didn't shape it
Your use case is different. The pattern isn’t.
Whether you are looking at enquiry handling, reporting, case notes, first-line support or something else entirely, the sequence is the same: assess where AI genuinely helps, prepare the information and processes behind it, support the implementation modestly, and maintain it with a person accountable for what goes out.
Organisations that skip to implementation get exactly the results you would expect. The ones that prepare first tend to find the technology was the easy part.
What would this look like in your organisation?
Start with a free AI Opportunity Assessment. We'll look at where AI could create real value, what needs to be in place first, and tell you honestly if the answer is not yet.
A structured look at where AI could create value. Free, with no obligation.