Read the copy on ten fintech websites this month and you will notice something odd. It is all fine. The sentences are clean, the claims are clear, the tone is calm and capable. And it is almost impossible to tell which company you are reading. Strip the logos and the brand names and most of it would slot interchangeably into any of the others.
This is not a coincidence, and it is not bad writing. It is the predictable result of everyone reaching for the same handful of tools, fed by the same internet, and asking them to write in the same sensible register.
The thing worth understanding is what that does to a brand over time, and why the response that feels obvious, use the tools more, makes the problem worse rather than better.
The floor went up, the ceiling came down
The honest case for AI in a communications function is strong, and I want to make it before I complicate it. The tools have raised the floor. Almost nothing that ships now is badly written. The first draft that used to take a junior a morning arrives in seconds and it is competent. For a four-person fintech marketing team or an NGO comms lead doing the work of three people, that is not a toy. It is genuine access to a standard of output that used to require headcount they could not justify.
So the floor is up, and that is a real gift. The trouble is what happened to the ceiling at the same time.
A model writes towards the middle of everything it has read. That is its job. Ask it for a paragraph about trust, or security, or putting customers first, and it returns the consensus version, the average of a million pages that have said something similar. The output is fluent precisely because it is familiar. It has rounded off the edges, because the edges are where the disagreement and the specificity and the odd phrasing live, and the model has been trained to smooth exactly those things out.
When one company does this, it gets competent copy. When every company in a category does it, drawing on the same two or three models trained on the same corpus, the whole category slides towards a single confident, competent, generic middle. Everyone sounds like a well-briefed stranger describing your business. Nobody sounds like the business itself.
That is the trade hiding inside the productivity gain. The floor under your writing went up. The ceiling on how distinctive it can be came down to meet it.
Distinctiveness was always the asset
The reason this matters commercially is that “sounding generic” can read like an aesthetic complaint when it is actually a growth one.
A brand is, in plain terms, the set of associations that let someone recognise and choose you without thinking hard about it. Recognition is cheaper than persuasion. If a customer already half-knows who you are and what you sound like, every subsequent message does more work for less spend. That recognition is built out of distinctiveness: the particular way you put things, the point of view you hold that your competitors do not, the phrases your customers could finish for you.
Homogenised copy quietly dismantles that. Not in a dramatic way, and not in any single piece. It happens by degrees, one sensible on-brand-enough draft at a time, until the things that used to mark you out have been averaged away and you are paying full price to acquire customers who cannot tell you apart from the three companies they are also considering.
When everyone can produce competent copy, competent stops being a differentiator. It becomes the price of entry. The scarce asset, the thing that is now genuinely hard to buy, is sounding like yourself on purpose.
What is actually load-bearing in a voice
If distinctiveness is the asset, the practical question is which parts of your voice you are willing to defend, because you cannot defend all of it and you should not try. Plenty of your copy is plumbing. The cookie notice, the password reset email, the third paragraph of a feature explainer. Let the tools have it. That is exactly the work the raised floor is good for.
The job is to know which parts are load-bearing and treat those differently. In most organisations they fall into a small number of places.
- The point of view. The actual argument your organisation makes about its market, the thing you believe that your competitors either disagree with or have never bothered to say out loud. A model will not generate this, because it is by definition not the consensus.
- The things only you would say. The specific stories, the real examples from your own work, the analogy your founder always reaches for, the lesson you learned the expensive way. These are unrepeatable because they happened to you and to nobody else.
- The house phrasing. The handful of recurring lines and turns of phrase that have become shorthand for how you see the world. A&C’s whole name is one of these. They are small, and they are doing more identity work than any paragraph.
That is the short list worth protecting. Notice that none of it is about polish, and all of it is about specificity. The tools are excellent at the first and structurally poor at the second, which is the whole reason the division of labour works.
Everything outside that list can be drafted, assisted, accelerated. The skill is not refusing the tools. It is knowing where the line sits, and being deliberate about the few places you choose to stay on the human side of it.
Treating it as a discipline, not a feeling
The reason this rarely gets managed well is that the drift is invisible in the moment. No single AI-assisted draft looks like a problem. Each one is fine. The cost only shows up in aggregate, months later, when someone asks why the brand feels flatter than it used to and nobody can point to the decision that made it so, because there wasn’t one. There was a thousand small defaults.
Which is why distinctiveness has to be a decision someone owns, not a quality you hope survives. In practice that means a few unglamorous things: deciding in advance where AI does not get used, keeping the load-bearing phrasing written down somewhere it can be defended, and having someone senior enough to read a sensible, competent, on-brand draft and say that it is also indistinguishable from everyone else’s, and send it back. That last judgement is the hard one, because the draft is not wrong. It is just not yours.
The honest counterargument is that this is a luxury a stretched team cannot afford, that when you are doing the work of three people the average draft in ten seconds is exactly what you need and worrying about distinctiveness is a problem for companies with more time. I have real sympathy for that, and I would not tell a small team to write everything by hand out of principle. The tools are too useful for that to be good advice.
But the answer to a cost you cannot avoid is not to pretend it isn’t there. It is to know what the tools quietly take in exchange for what they give, and to price it in. Spend the raised floor on the plumbing, deliberately, so that you have the time and the attention left to protect the few things that make you sound like no one else. The teams that come out of the next couple of years sounding like themselves will not be the ones who used AI least. They will be the ones who decided, on purpose, where not to.
This is the kind of judgement we hold for communications leaders at A&C, often as part of a Fractional Communications Director engagement: knowing which parts of a voice are load-bearing and making sure they survive contact with the tools. If your copy has started to sound competent and anonymous and you want it to sound like you again, let’s have a conversation.