AI can craft a pretty believable brand strategy, but will it hold? Here are some things to know before you forward it to the team.

With all the things it’s proving good at, why don’t we outsource brand strategy to AI entirely?

You may read this as common sense or as a brash provocation, depending on your beliefs. Personally, I use AI liberally when needed, so I don’t really fall into the camp of its detractors, even for strategy. And yet, exactly because I have been using it quite a bit, I’m still perplexed when I see a completely AI–made brand strategy.

Yes, it shows.

And I’m not talking about em dashes or “it’s not this… it’s that” pettiness. There is something more profound that you should address if you are serious about your brand strategy (by the way, see our checklist about what makes a solid brand strategy).

But let’s be honest: I understand why companies ask AI to write their brand strategy. If you see it as an “internal document” no one will really see, it means saving time and money you can then use for more visible parts of the work. If you just need it as a piece of paper or a very broad reassurance, it works.

We have seen quite a few companies come to us with an AI–generated “draft” brand strategy, asking for a review or total rehaul. Companies who know brand strategy is a serious matter, who – after trying with AI – sensed something was not quite there.

Humans can usually tell.

Understanding the fuzzy part of brand strategy

Strategy has been defined in a zillion different ways. One that I love for its simplicity is this: “Strategy is an informed opinion on how to win”.

This short sentence packs a few grains of wisdom. “Informed opinion” is its core: strategy is based on information, but information is not enough to lead to a strategy. You need an opinion. Merriam-Webster defines an opinion as “a personal view, judgment, or belief about something that is not based on absolute proof or certainty”. 

And this is where things get interesting. Strategy deals with the unknown and the unknowable. It deals with possible futures and it tries to maximize the chances to win. A strategy – even a brand strategy – is never a fail-proof formula, it’s a direction where it’s more probable that you will win.

Taking a stance on such delicate matters is where you usually need a human to take the leap for you.

AI lives in the past

‎

In our crazy-weather times we are all very familiar with forecast services and apps.

Weather forecast is a good metaphor for strategy: it tries to give you the most probable prediction even though it’s not 100% reliable. You know there is a 90% chance it will rain, so rescheduling your picnic becomes a good decision. Add the fact that most weather apps now integrate AI.

And yet, different apps will give you different forecasts.

But weather forecasting has a big advantage: it can rely on very recent data. Satellite views, sea-level and temperature sensors, weather stations. This insane amount of real-time data helps the models make sense of very complex systems to predict things more accurately.

AI doesn’t really work that way (at least not yet). Large language models are mostly trained on past knowledge – however recent – and they try to infer the future in a probabilistic way. They might mix in web search, which is also a backward-looking process.

Imagine if weather was predicted entirely that way, by looking at historical series. Not only would it be very imprecise, but it would entirely miss out-of-scale phenomena.

When it comes to strategy, that involves people and their behavior at scale, that kind of real–time monitoring becomes harder. Not only because of the technical part (sadly, we are increasingly tracked) but because not everything that matters can be measured.

Which leads me to the distinction between hard facts, non–trivial facts and weak signals.

Hard facts, non-trivial facts and weak signals

Brand strategy has to take into account many kinds of information.

Some can be easily gathered, and that’s usually the hard facts.
Who are the main competitors? What’s their performance? How large is the market? What is the average customer lifetime value in the industry? This information is available, sometimes open and sometimes behind a paywall (which is another issue for AI).

It’s easy to think hard facts are all you need. But hard facts are just a starting point.

Next, you have non–trivial facts. These are facts that are not directly related to your brand but that matter to your brand anyway. Here, you need a leap that comes from an opinion. For example, Netflix has long been saying that their main competitor is sleep. So if you are doing strategy for Netflix you might want to do some research about sleep trends in your market. In this case the data is again available – AI might look it up for you – but the idea to consider it is something that has to be prompted, or it simply won’t happen. In other words, you need a strong opinion to open that direction of research.

Finally, you have weak signals. These are the most creative part of brand strategy work and the one that strategists are sometimes less confident mentioning, as it’s not backed by a Nielsen research. And yet, weak signals make all the difference.

I’m talking about signals that the strategist has perceived that are not statistically relevant yet, but may help intercept a changing tide. Weak signals are the core of cultural strategy. Since culture is a moving thing, data is usually too old to understand it and it’s usually only understood in retrospect. Think about magazine articles about youth cultures: once they are printed, they are usually old.

If people start appreciating colors after a long period of monochrome buying, it might take a while before it shows in the data (for example, a market-wide survey of car purchases). As a brand, though, you may not want to be that late to that party.

On the other hand you might have seen that your circle of friends is now dressing more colorfully. That a director that usually wears black shows up for Variety’s interview in a teal t-shirt. That your neighbour, a serious and somewhat gloomy man, just bought a cherry-red SUV. Yes, I’m making the brand strategist look like a detective, but the parallel holds. 

Weak signals are absorbed in different ways: reading, watching movies, talking to people, watching people in the wild. 

In other words, weak signals can only be acquired by a diverse and somewhat messy human experience, which is exactly what AI can’t have at the moment.

‎

The probabilistic trap: how AI harms differentiation

Another issue with AI–generated brand strategy lies at the core of how AI works. Suppose you have all the possible data. Then AI should be able to give you the right brand strategy, right? Right?

Well, let me tell you: what good is a brand strategy if everybody is using the same?

Given the same inputs – and most of us rely on the same data – AI will give you a different flavor of the same response. It’s bound to do so, because it’s a probabilistic tool. Its reasoning, its choice of words, is driven by what’s more probable given its training set. If three companies want to launch a new burger joint in town, AI will serve them all similar strategies.

Which is a real problem, since brand strategy is all about differentiation. Strategy is a complex, not straightforward game. 

A quality brand strategy takes this into account and tries not to jump to the first and most obvious conclusion. This is where strategy gets creative and tries to add something that is not already in the data, to create its own space.

It makes me think of an old Mickey Mouse story, where Goofy was the only one who could defeat a perfectly rational enemy because he was so illogical and unpredictable (which is also the logic behind drunken kung fu).

How to use AI in brand strategy

Does this mean we should abandon AI when doing brand strategy? Absolutely not. I will be the first one using it. But there are proper ways to do so.

1. Give it the main direction
Remember the Netflix vs sleep example? You decide what you aim your spotlight at. And a proper strategy needs to go beyond the obvious, even in its research phase. AI can help you accelerate this phase, not replace it.

2. Use it as a multiplier, not an originator
Don’t wait for AI to give you the initial spark, see it more as a flammable liquid. Once you have found a spark, AI can help you expand its territory rapidly. AI can give you quantity, not accuracy. Don’t expect it to find the solution, but to give you a broader view of what’s possible.

3. Use it to question your assumptions
Different AIs – Claude, ChatGPT, Gemini – behave differently in terms of sycophancy, i.e. the tendency to always tell you that you are right, no matter what. Rest assured, though, that they will try to come up with an answer even if your question is bonkers. In other words, they are not usually a critical partner to work with. If you want your thesis to be questioned, you usually need to prompt for it. Ask your LLM to pierce holes in your vision, or to offer a counter vision. 

4. Don’t trust its reassurance
AI might start mixing up sensible true pieces of advice with hallucinatory ones, and keep reinforcing the latter with even more hallucinations. If you ask, it might confess – or not. Reality checks are entirely up to you and they must be frequent and thorough: the more you let AI elaborate on a false basis, the more time you are going to waste.

5. Ground the knowledge
When possible, use RAG (Retrieval Augmented Generation) platforms like NotebookLM by Google for your research phase: you can feed them a certain amount of docs and only interrogate those, reducing the risk for hallucinations to a minimum.