Back to blog
Most leadership teams are already using AI, and each person is using it in their own way. Marketing has one tool, sales has another, and someone in the product team has a private setup nobody else has seen. Each produces output that reads well. Put the output side by side and it no longer sounds like one company. That is how a brand gets diluted: as ten good pieces in ten slightly different voices, with no single bad one to point at. How to use AI without diluting your brand is therefore a decision and not a tool choice. It is a decision about what the company sounds like, and what it will not hand to a machine.
High intelligence, no feel for the brand
The simplest way to think about AI is as a team of high achievers with a very high IQ and no EQ. They are fast, well read and tireless. They also have none of what the Germans call Fingerspitzengefühl, the feel in the fingertips that tells an experienced person what is right for this customer, in this tone, at this moment. Without cultural guidance, that team will dilute everything you stand for.
The mechanism is ordinary. AI fills every gap with the average of everything it has read, and for any given category the average is the voice every competitor already shares. Left unguided, your output drifts towards the middle. Nothing in it is wrong, and that is what makes it hard to spot. It could belong to anyone, and a brand that could belong to anyone has stopped doing the work of a brand.
AI does not dilute a brand by making mistakes. It dilutes it by being fluent in everyone's voice.
A brief that starts in the wrong place
In a pilot case we ran, a communication training agency asked a reasonable question: how can we use AI to increase our prospects and leads, and grow sales? Our debrief was that sales are not won at the top of the funnel with AI. They are won by being the proven, obvious choice in the customer's eyes. The first job was to protect that, and only then to build the engine that delivers and scales it.
The agency was already winning customers against competitors, and client retention sat at 80 to 90 per cent. Inside the company, though, people held conflicting views of what its identity actually was. Putting AI on top of conflicting views would have scaled the conflict. A request to use AI to grow sales can hide an earlier question: what do customers choose you for, and would everyone in the company answer it the same way?
The stakes follow from that. A company that wins by being the obvious choice has more to lose from sameness than one that wins on price. Its retention, its margin and its ability to be shortlisted all rest on customers seeing a difference. AI multiplies output without any particular regard for that difference, so the faster it runs, the more it needs a clear picture of what it must not smooth away.
How to tell it is already happening
Dilution is slow, so it helps to test for it. Take five recent pieces of customer-facing text that AI helped to produce: a proposal, a web page, an email sequence, a case study, a post. Remove the logo and the company name, and ask three people who know your market two questions. Could a competitor have written this? And does it sound as if the same company wrote all five? If the answer to the first is yes, the output has drifted to the category average. If the answer to the second is no, the company has several voices and nobody has decided which one is right.
Other signs are quieter. People disagree about which tool or version is the official one. Customer-facing work has no name against it. A new colleague asks what the company sounds like and gets five different answers. None of these starts as an AI problem. AI makes them visible, and then makes them bigger.
Find out what is already running
Before writing anything, find out what is already in use. Ask everyone who uses AI for customer-facing work to show which tool they use, what they ask it to do and what material they give it. Put the answers on one list. Look in particular for personal accounts, tools nobody approved, prompts that copy a competitor's tone because it was the nearest example, and customer information pasted into services the company has not assessed.
The list does two things. It shows how many versions of the company's voice already exist. And it shows where the data is going, which turns a brand question into a governance question, because anything you write down about tone will be ignored if the tools behind it are private. It is the first piece of evidence for the leadership discussion, and one that is hard to argue with. Keep the list current. A quarterly refresh is enough to show whether the boundaries are being used or ignored.
What to write down, and who answers for it
Four things need to be on paper before AI does any customer-facing work. The first is one identity. Leadership signs off a single value proposition: the offer, the benefits and three reasons to believe. If that cannot be agreed, nothing downstream will hold. The second is a source of truth that every tool reads from, combining existing content with a written account of how the company works, its method, its tone and where its information lives. Source of Truth: Where to Play, How to Win for AI walks through how to write one.
The third is a line between toil and craft. Toil is routine work nobody builds a reputation on: admin, scheduling, first drafts of structure. Craft is the work that gives the brand its footprint: the advice, the conversation, the final judgement. Automate the first, protect the second, and keep the do-not-automate list where everyone can see it.
The fourth is a named owner for every output that reaches a customer. In our own working model, AI drafts the structure and a senior person owns the craft and the final words. For a product company the equivalent is a person who checks each piece against the written boundary before it leaves the building. A practical check has three questions. Does it say something only we would say? Does it match the written tone? Is every claim one we can stand behind? If any answer is no, it goes back, and The AI-Augmented Organisation shows how to build a team around that check.
None of this replaces customers. AI cannot tell you whether your identity is the right one. In the case, testing positioning with real customers alongside the first pilot was marked urgent, and the customers' own words became some of the best raw material for the value proposition.
Where to start
Once the boundaries exist, the best first jobs are the repeatable, checkable ones. List the easy wins and put them in order.
Before either, write down the three things a customer would say they choose you for, and the three things you would never let a machine write. If the leadership team cannot agree on either list, that disagreement is the first AI project, and it needs no tool.
Key takeaways
AI dilutes a brand by drifting towards the average of everything it has read. The output is correct and could belong to any competitor, which makes the drift hard to spot.
A request to use AI to grow sales can hide an earlier question: what customers choose you for, and whether everyone inside the company would answer it the same way.
Test for dilution with five recent pieces of AI-assisted text. Remove the name and ask whether a competitor could have written them and whether one company wrote all five.
Put four things on paper first: one signed-off value proposition, a source of truth every tool reads from, a line between toil and craft, and a named owner for every customer-facing output.
AI cannot confirm that your identity is the right one. Test it with real customers alongside the first pilot.
FAQ
How do you use AI without diluting your brand? Agree one written value proposition, build a source of truth that AI works from, decide in advance which work is routine and which is craft, and give every customer-facing output a named owner who checks it before release.
Why does AI dilute a brand? Because it fills every gap with the average of what it has read, and the average voice in a category is the one every competitor already shares. It does not make visible errors. It produces fluent output that sounds like everyone else.
How can you tell whether AI is already diluting your brand? Take five recent AI-assisted pieces, remove the name and ask whether a competitor could have written them and whether they sound like one company. Disagreement about which tool is official is another sign.
What should never be automated? The work that gives the brand its footprint: the advice, the customer conversation and the final judgement. Routine work, such as administration, scheduling and first-draft structure, is the natural candidate for automation.
Where should a product company start with AI? With the identity question, not the tool. Agree what customers choose you for, write down what a machine must not write, and then take on the repeatable, checkable tasks first.
