AI can do a lot, but it still does not have any taste.
That sounds like a small limitation until you start using it properly. Once the novelty wears off, the question is not whether AI can produce something. It can. The question is whether the thing it produces is any good, whether it should exist, and whether it feels like it came from someone with judgement.
That is where the conversation gets more interesting. Most of the noise around AI is still about capability. Can it write? Can it code? Can it build a campaign plan? Can it produce a strategy? Can it draft a sales page, analyse a report, or summarise a customer interview?
Yes. Often well enough to be useful.
But capability is not taste. Capability gives you output. Taste decides whether that output is worth keeping.
Direction helps, but it is not taste
Most bad AI work starts with bad direction. Ask for “some marketing ideas” and you will get the same recycled mush everyone else gets. Ask for a blog post about AI and business and you will get a blog post that sounds like every other blog post about AI and business.
That is not really the model’s fault. It is doing what it was asked to do.
Give it a clearer goal and the work changes. Tell it who the reader is, what they are sceptical about, what they already believe, what you need them to understand, and what the work is meant to change in their mind. The output gets sharper because the machine has something to aim at.
This is why I do not buy the lazy argument that AI is useless. It is not useless. I use it all the time. The difference between vague AI use and goal-led AI use is enormous, which is why I keep coming back to the importance of context. I wrote about that more directly in Context is King, because the companies getting better results from AI are not always using better tools. They are giving the tools better context.
But even with a clear goal, something is still missing.
A goal gives the work direction. Taste decides whether the direction is worth following.
Taste is knowing what to remove
Taste is not polish.
Polish is easy to mistake for quality because it looks finished. AI is very good at that. It can make weak thinking look structured. It can make a bland idea read smoothly. It can turn uncertainty into a confident paragraph with headings, transitions, and a neat conclusion.
That is exactly why it is dangerous.
Taste is different. Taste is knowing the idea sounds clever, but says nothing. It is knowing the line is technically fine, but trying too hard. It is knowing the campaign is clear, but forgettable. It is knowing the deck is well organised, but has no real point of view.
In marketing, that is not decoration. That is the work.
You see it in the small decisions. What do we say first? What do we leave out? Where are we being too safe? Where are we hiding behind jargon? Where have we made something clearer by removing the very thing that made it interesting?
Good taste often shows up as restraint. It is not always adding the better line. Sometimes it is cutting the line that makes you sound pleased with yourself. Sometimes it is removing the extra proof point because the buyer already has enough to think about. Sometimes it is deciding that the campaign does not need another angle, another audience, another variation, or another round of output.
Taste is knowing when to stop.
Or keep going.
Acceptable work is the real problem
Bad AI work is not the main risk. Bad work is usually obvious. You can spot it because it sounds dead, generic, or wrong.
The real risk is acceptable work.
Acceptable work gets approved. It fills the calendar. It gives the team something to review. It keeps the campaign moving. It makes everyone feel productive because there is visible progress, even if the progress is not moving the business anywhere useful.
This is already a problem in marketing without AI. We have always been good at confusing activity with progress. AI just makes it cheaper and faster to create more activity.
That is why the point in Stop Automating Everything matters here. Automation is useful when it removes friction from the right things. It becomes a problem when it lets you scale work that should have been questioned first.
The same is true of AI content, AI campaign planning, AI reporting, and AI creative. If the thinking is weak, faster execution does not save it. It just gives weak thinking a better outfit.
Intelligence is not judgement
The bit I find hard to believe is the idea that this problem disappears as models get smarter.
I do not think it does.
Load as many IQ points into the AI model as you like, the end result will not automatically be better taste. It may reason better. It may write cleaner copy. It may catch more mistakes. It may produce more useful first drafts. All of that matters.
But taste is not just intelligence.
Taste is judgement shaped by experience, context, standards, preference, memory, and a point of view. It is personal without being random. It is explainable, but not mechanical. Two smart people can look at the same piece of work and disagree, not because one of them has failed to calculate the answer, but because they value different things.
One wants clarity. Another wants edge. One wants safety. Another wants memorability. One wants the version that passes approval. Another wants the version a customer might actually feel.
Those are not just processing decisions. They are taste decisions.
Harvard Business School made a related point in a recent piece on why human judgement still drives innovation. The research it discusses is not really about taste in the artistic sense. It is about judgement, experience, and the ability to know which AI suggestions are worth acting on. That is the bit many people skip. The tool can give you options. It cannot carry the responsibility of choosing well.
Marketing needs more than correct answers
There are plenty of business tasks where acceptable output is enough. Some internal summaries do not need personality. Some admin does not need a point of view. Some process documentation simply needs to be clear and accurate.
Marketing is different.
Marketing is not just information transfer. It is persuasion, timing, framing, tension, confidence, and feel. It is knowing what the customer already suspects. It is knowing what they are tired of hearing. It is knowing when your message is true, but not interesting enough to matter.
This is why taste shows up so clearly in PPC. The platform can optimise delivery. The data can tell you what happened. The reports can show movement. But someone still has to decide what good looks like. I have written about that in What Good PPC Management Actually Looks Like, because good management is not just pulling levers inside an ad account. It is judgement about what the account should be trying to achieve.
The same applies to AI. A model can generate five campaign routes for Meta, LinkedIn, YouTube, or programmatic. It can make them cleaner. It can adapt them to different audiences. It can make the copy sharper if the direction is good enough.
But it cannot know, in the way a human knows, that the third route feels empty, that the second is too safe, or that the first one has the uncomfortable truth in it.
That is taste.
Design has always worked this way
This is not a new problem created by AI. AI has just made it easier to see.
Don Norman’s work on emotional design is useful here because it reminds us that people do not only respond to function. They respond to how things look, feel, behave, and what those things say about them. A product, page, advert, or piece of content can be functional and still fail because it creates the wrong feeling.
That is why taste is commercial, not cosmetic.
The business owner reading your page is not scoring it like an examiner. They are deciding whether they believe you, whether you understand them, whether you feel like their kind of company, and whether your work carries enough confidence to trust.
AI can help produce the page. It can help generate the options. It can even help you challenge the weak points if you ask properly.
But someone still has to feel the difference between clear and bland.
The human role gets more important
I do not see any of this as an argument against AI. That would be pointless. AI is useful, and businesses that ignore it will make life harder for themselves.
The point is narrower and more important.
AI does not remove the need for judgement. It exposes how much judgement was already missing.
That is the same lesson behind What It Actually Means to Manage AI Agents. Agents, automations, and AI workflows do not manage themselves. They need direction, context, boundaries, correction, review, and taste. Without that, you do not get useful scale. You get faster mess.
This is also why AI can make you busier rather than freer. I wrote about that in Why Ai is Making You Busier, and the same logic applies here. If you create more output than you have the judgement to filter, you have not solved the problem. You have moved it.
The advantage will not belong to the people who produce the most. It will belong to the people who can choose best.
Taste is the filter
Intent gives the work direction.
Taste gives it judgement.
Put them together and you get something human. You get work with a reason to exist and a point of view behind it. Take taste away and you are left with output, and output is getting cheaper every day.
That is why I do not believe this is something you simply iterate into the next model. A better model may be more useful. It may be more fluent, more capable, and more impressive. It may help you think more clearly if you use it well.
But taste is not just the next feature.
Taste is the human filter. It is the part that says this is worth saying, this is not, this sounds like us, this does not, this is clear but dead, this is rough but alive.
AI can do a lot.
But someone still needs to know what good looks like.
