Growth · The Spark
Using AI to write without destroying your brand
Most teams are either sceptical holdouts or flooding channels with obviously machine written copy. A four stage framework for the leverage without the damage.
The argument about AI and content has moved on from whether to use it. The live question now is how to use it without wrecking the thing you spent years building.
Most teams sit at one of two extremes. Either they are holding out and producing everything by hand, or they went all in and are now filling their channels with copy that anyone can tell was generated. Neither works. Here is what does.
Three things the models still cannot do
The tools have genuinely improved. They handle structure, nuance and voice matching far better than they did. There are still three gaps that matter, and they are the three that decide whether content is worth reading.
They cannot originate from lived experience. A model can synthesise everything already written on a subject. It cannot sit in your customer meetings, feel a launch go wrong, or have the thought in the shower that reframes the whole problem.
They will not take an editorial risk. Models drift toward the consensus, because the consensus is what they were trained on. The content that earns attention takes a position and says something the reader has not already heard nine times.
They cannot build the relationship. Readers can tell the difference between something written by a person who cares about the subject and something produced to fill a slot in a calendar. They are rarely wrong.
None of that makes the tools useless. It means press button, get article produces mediocre work at volume, which is worse than producing less work of higher quality.
The four stages that work
Stage one, the idea is yours. Every piece should start from something a human noticed: a pattern across customer conversations, a question that keeps surfacing on sales calls, a lesson from a project that went sideways. Use the model here for adjacent angles and headline variants, not for the insight itself.
Stage two, research and shape. This is where the leverage is real. Gathering supporting material, proposing structures, finding the holes in your argument, surfacing relevant examples. The principle that changes the output more than any other: feed it your own material. Your previous work, your case studies, your tone of voice. A generic model produces generic copy. A model working inside your context produces something that sounds like you.
Stage three, draft as a conversation. Give it a real brief with your argument and your key points. Let it draft the middle sections while you write the opening and the close, because that is where your voice matters most. Ask for three versions of the paragraph you cannot get right. What does not work: asking for a post about a topic with no further context, accepting a first draft, and letting it write your calls to action.
Stage four, elevate. Every piece needs a human pass, and not only for accuracy. Ask four questions. Does this sound like us. Have I added at least one thing only we could have said. Would I put my name on it. Does the opening earn the second paragraph, or does it read like an encyclopaedia entry.
Where the leverage actually sits
In between sit the pieces where it helps with structure and research but the substance has to be yours. Blog posts, case studies where the detail comes from an actual interview, landing page variants where the underlying proposition still needs a person.
The threshold nobody wants to set
Here is the uncomfortable part. Most generated content is not good enough to publish. Not because the tools are bad, but because teams are not giving them enough context, not editing hard enough, and not adding the parts that make something worth a reader's time.
So set a threshold and hold it. Would anyone save this. Would they send it to a colleague. Would they come back because of it. If the answer is no, do not publish it, however fast it was to produce. Publishing mediocre work at volume damages a brand faster than publishing nothing.
The teams that get this right do not produce more than everyone else. They produce better, faster, and spend what they save on the parts no model can do for them.
Frequently asked
Questions buyers ask about this
Can AI write content that sounds like our brand?
Only if you give it your context. A model working from a blank prompt produces generic copy every time. Feed it your existing best work, your tone of voice guidance, your case studies and your customer research, and the output moves much closer to something you would publish. The context is doing most of the work, not the prompt.
What should we never use AI to write?
Thought leadership, brand positioning, crisis communications and personal founder content. The first two require original thinking by definition, the third has stakes too high for anything but genuine human judgement, and the fourth stops working the moment it stops sounding like the person whose name is on it.
How much editing does AI assisted content need?
More than most teams do. Assume a full human pass on every piece, checking not only accuracy but whether it sounds like you, whether it contains at least one insight only you could have contributed, and whether the opening genuinely earns attention rather than reading like a reference entry.
Does publishing AI assisted content hurt search performance?
The format matters far less than the quality. Thin, derivative content performs badly whoever or whatever produced it, and search and AI systems are both getting better at spotting it. Well structured content carrying original data and genuine expertise performs well regardless of what helped write it.
Where does AI add the most value in a content operation?
Email sequences, repurposing long form work into social, structured product copy and summarising research. These are tasks with clear inputs, a defined format and a low requirement for original insight, which is exactly the shape of work the tools handle well.
Working on a real engine? Start with a conversation.
Tell us where you are. We will tell you what we see and where we would start.