More insights.
Subscribe to our newsletter.
Deep dives into design thinking, creative process, and the intersection of business and aesthetics.
AI can write the draft, but it's your name that goes on it.
Sooner or later someone in the office asks why you don't just get AI to write the blog. I can see the appeal. A long white paper now takes about as long as making a coffee, and it reads fine. The trouble is that every other firm has the same tool and the same idea, so there's far more AI content about than there used to be, and a lot of it sounds alike.
You can see it on LinkedIn and in the sales emails landing in your inbox: the same smooth paragraphs, the same pet words (delve and tapestry are the ones people make fun of), much of it from a handful of models on their default settings. Paste that onto your website unedited and a reader has no way of telling you from the firm down the road.
Replacing writers is the smaller worry
Most people start by asking whether AI will replace their writers. For a brand, the more pressing risk is that it makes you sound like a commodity. When everyone has the same tools, what's left to set you apart is a point of view, and a model doesn't have one.
I'm not against using it. We treat it the way you'd treat a junior researcher or a sub-editor: it gets you past the blank page and speeds the work up, but it doesn't get a say in what we think.
Where it earns its keep
Research and outlining are the obvious uses. It's good at summarising a dense topic and suggesting a structure you can then argue with. Ask it for ten headlines and one of them might shake loose the one you actually use. It's handy for repurposing too, turning a transcript of your managing director's talk into a first draft of a LinkedIn post, say.
All of that is exploration. It can help you find the right question, but the answer has to come from your own experience, which is what the reader came for.
Where AI content does the damage
Trouble starts when the model holds the pen. AI prose manages to be plausible and empty at the same time. It has none of the rhythm, or the odd specific detail, you get from someone who has done the work, so it reads like a press release nobody wanted to sign.
Worse, models make things up. We've seen drafts that invented statistics, and drafts that cited laws that don't exist. Publish one of those and nobody is going to blame the software. In a regulated sector like finance or healthcare, generic advice written by a model is the last thing you want going out under your name.
Someone has to read every word before it goes live
If AI is anywhere in your process, whoever signs it off needs a few firm rules. Don't trust a statistic from a model: if a draft says "70% of users prefer X", find the primary source or delete the sentence. Rewrite anything leaning on the usual giveaway words like elevate or unlock. Then hold what's left up against your house style. A voice guide matters more once AI is involved. If yours only exists in people's heads, write it down, and keep it short enough that people will use it.
Search rewards experience a model doesn't have
Google's own guidance on creating helpful content leans on E-E-A-T: experience, expertise, authoritativeness and trustworthiness. The first E is the one to notice. A model has never run a project or had to let a client go; all it does is predict the next word.
Publishing more won't get round that. What does well in 2026 is writing that's specific, by someone who has done the work and is willing to hold opinions a model would sand down into something safe. That kind of detail is also what gets quoted when an assistant answers the question.
People at both ends, the model in the middle
Stop trying to automate the whole chain. A person decides the argument and who it's for. The model can rough out a structure and some first-draft paragraphs. Then a person takes it back, checks every claim, rewrites it in your voice, adds the examples only you have and signs it off. The model only ever touches the middle, which is the easy part anyway.
To see how far this applies to you, take the last few things you published with AI's help, imagine a competitor's logo at the top and read them again. Anything that would pass as theirs needs rewriting.
Related Brainstorm services and work
Brand strategy and identity · Content, website and growth services · Discuss a distinctive content system
Updated 24 September 2026. Dave Curtin, founder and Creative Director of Brainstorm, leads website strategy, brand direction and technical delivery for Brainstorm clients.
AI can write the draft, but it's your name that goes on it.
Sooner or later someone in the office asks why you don't just get AI to write the blog. I can see the appeal. A long white paper now takes about as long as making a coffee, and it reads fine. The trouble is that every other firm has the same tool and the same idea, so there's far more AI content about than there used to be, and a lot of it sounds alike.
You can see it on LinkedIn and in the sales emails landing in your inbox: the same smooth paragraphs, the same pet words (delve and tapestry are the ones people make fun of), much of it from a handful of models on their default settings. Paste that onto your website unedited and a reader has no way of telling you from the firm down the road.
Replacing writers is the smaller worry
Most people start by asking whether AI will replace their writers. For a brand, the more pressing risk is that it makes you sound like a commodity. When everyone has the same tools, what's left to set you apart is a point of view, and a model doesn't have one.
I'm not against using it. We treat it the way you'd treat a junior researcher or a sub-editor: it gets you past the blank page and speeds the work up, but it doesn't get a say in what we think.
Where it earns its keep
Research and outlining are the obvious uses. It's good at summarising a dense topic and suggesting a structure you can then argue with. Ask it for ten headlines and one of them might shake loose the one you actually use. It's handy for repurposing too, turning a transcript of your managing director's talk into a first draft of a LinkedIn post, say.
All of that is exploration. It can help you find the right question, but the answer has to come from your own experience, which is what the reader came for.
Where AI content does the damage
Trouble starts when the model holds the pen. AI prose manages to be plausible and empty at the same time. It has none of the rhythm, or the odd specific detail, you get from someone who has done the work, so it reads like a press release nobody wanted to sign.
Worse, models make things up. We've seen drafts that invented statistics, and drafts that cited laws that don't exist. Publish one of those and nobody is going to blame the software. In a regulated sector like finance or healthcare, generic advice written by a model is the last thing you want going out under your name.
Someone has to read every word before it goes live
If AI is anywhere in your process, whoever signs it off needs a few firm rules. Don't trust a statistic from a model: if a draft says "70% of users prefer X", find the primary source or delete the sentence. Rewrite anything leaning on the usual giveaway words like elevate or unlock. Then hold what's left up against your house style. A voice guide matters more once AI is involved. If yours only exists in people's heads, write it down, and keep it short enough that people will use it.
Search rewards experience a model doesn't have
Google's own guidance on creating helpful content leans on E-E-A-T: experience, expertise, authoritativeness and trustworthiness. The first E is the one to notice. A model has never run a project or had to let a client go; all it does is predict the next word.
Publishing more won't get round that. What does well in 2026 is writing that's specific, by someone who has done the work and is willing to hold opinions a model would sand down into something safe. That kind of detail is also what gets quoted when an assistant answers the question.
People at both ends, the model in the middle
Stop trying to automate the whole chain. A person decides the argument and who it's for. The model can rough out a structure and some first-draft paragraphs. Then a person takes it back, checks every claim, rewrites it in your voice, adds the examples only you have and signs it off. The model only ever touches the middle, which is the easy part anyway.
To see how far this applies to you, take the last few things you published with AI's help, imagine a competitor's logo at the top and read them again. Anything that would pass as theirs needs rewriting.
Related Brainstorm services and work
Brand strategy and identity · Content, website and growth services · Discuss a distinctive content system
Updated 24 September 2026. Dave Curtin, founder and Creative Director of Brainstorm, leads website strategy, brand direction and technical delivery for Brainstorm clients.