A managing director sits down with ChatGPT and types “what should we do next as a business?” They expect a response that’s close to a strategic plan. The answer sounds credible, and therein lies the problem.

AI is being treated as a general answer machine, when it behaves more like a junior analyst with access to the internet. There is a way to get the best out of the technology. Asking for answers is not that way.

A recent Harvard Business Review article looked into using AI for strategic advice[1]. It concluded that the leading LLMs all gave the same type of answers. They compared them to a freshly-minted MBA parroting the latest management thinking.

LLMs reflect the data they are trained on. When the internet is full of fashionable business advice, you are going to get that repeated back to you. The researchers ran over 15,000 trials with the most popular LLMs to make sure that prompt quality was not affecting the outcome. They showed that the advice was the same, regardless of the type of company under consideration, and whatever its stage of development.

So what is the latest strategic advice for business leaders? The research asked about seven scenarios, each with a binary outcome. The only one where there was a split opinion was whether it is better to exploit your existing market, or open up a new one. The advice for all the others was the same.

The LLMs said it is better to pursue a differentiated, high-value product rather than to become a leader on cost. So no new Walmart or Wetherspoons. And no reinvesting profits into lower prices, which is Warren Buffett’s preferred consumer strategy.

You should augment human labour rather than automate it away, while prioritising long-term earnings over quarterly returns. Decentralised decision making is preferred to enforcing corporate consistency, and innovation should be incremental rather than radical.

Finally, when faced with the choice of cooperating or competing with rivals, collaboration is preferred. The exception was DeepSeek’s latest model, which said to compete.

None of this is necessarily poor business advice. It may be right for your business. But bear in mind that everyone is receiving it.

We cannot, therefore, just ask AI for a strategy. What should we do instead?

The researchers revealed three interesting findings. The first was that reversing the order in which an option was presented changed the advice. This was not enough to change the overall conclusion, but enough to muddy the water on some questions. It seems LLMs have a sequencing preference, which is worth considering when asking for a choice between two or more items.

The second finding is that context has far less impact than expected. Feeding the LLM with details about the company resulted in the same advice. It seems as if the answer was shaped to fit the context, rather than the context shaping the answer. This is an important finding, because most prompt advice assumes that telling the LLM about your situation is important detail.

The final finding is that when offered the choice of doing both options, the LLMs became confused. I’ve talked before about bosses who believe they are clever by replying “both” to an either/or question. They think they are pushing their team to deliver more. In fact they are doing the same as the LLM. It sees merit in both options and cannot decide.

So we should not ask for advice, even after adding context, and we should treat non-binary outcomes as red flags. The correct approach is criticism.

The right way to use AI for strategy is as a critic rather than an advocate.

Have LLMs generate alternative answers to strategic questions and highlight the blind spots in both. If you decide on a strategy, use an LLM to steelman the opposite conclusion. When there is a binary choice, have the LLM surface examples of both options at work, and relate the results to your business.

One CEO I am working with passes responses between SuperGrok and the latest ChatGPT model. The LLMs expose weak arguments and highlight gaps in each other’s answers. When writing important legal or financial strategy, the CEO will iterate several times before settling on a solution.

Prompting an LLM about what to do will return the latest trendy thinking. This is not designed for your company, and giving the LLM details of your business does not change this. Ignore advice that sees merit in both options, because it will not help you decide.

It’s an old adage that you hire a strategy consultant to tell you what you already know. You need an outsider to say it to convince everyone else. This works because the people that know the business best are those living and breathing it day-by-day.

This truth has not changed, even in the age of AI.

[1] Harvard Business Review, March 2026