Before you give AI your prompts, give it your values

We spend a remarkable amount of time learning how to prompt AI. There are frameworks, prompt libraries, courses and endless advice about how to get better answers from ChatGPT, Claude and other large language models. Much of it is useful, because clearer context and better questions usually do produce better responses.

But when it comes to AI prompts and values, most of us are concentrating on the wrong end of the conversation. As AI plays a bigger role in how we think, make decisions and solve problems, I’m beginning to wonder whether we’ve been too focused on the prompts and not focused enough on what we actually stand for.

Before asking an AI what to do, do we need to be clearer about what matters to us?

Understanding your values

That becomes particularly important when we move beyond relatively mechanical tasks such as drafting an email or summarising a document. Increasingly, people are using AI to think through career choices, business decisions, financial trade-offs and other questions where there may be no objectively correct answer. At that point, a technically impressive response is not necessarily a good response, because the quality of the answer depends partly on the values against which the options are being judged.

I’ve worked with individuals and leadership teams on values for many years, and one thing that continually strikes me is how powerfully values influence our behaviour, while often remaining below conscious awareness. In an earlier article I used the example of Emma, a trainer who was running late for an important workshop when she saw an elderly woman fall and hurt herself. Emma is a qualified first aider. If she stopped to help, she’d be late for the client; if she continued, she would be leaving somebody who may need assistance. Two qualities Emma values, professionalism and helping others, had come into conflict. How she responded depended on the hierarchy she gave those values in that particular situation.

We make these kinds of choices all the time. Security can conflict with freedom, ambition with family, loyalty with honesty, achievement with wellbeing. Becoming more conscious of the values involved doesn’t make the dilemma disappear, but does help us see what we’re actually deciding between.

That could become one of the most useful roles for AI.

The real problem isn’t how well you prompt

Imagine asking an LLM whether you should accept a promotion. It can compare salary, responsibility, career prospects, travelling time and working hours, but it can’t know what a good decision means for you unless you provide some context. If financial security is your overriding priority, the decision may look very different from the way it looks if freedom, family time or meaningful work sit higher in your personal hierarchy.

Rather than asking, “Should I take the promotion?”, a richer conversation might begin with something like: “I value financial security, autonomy and time with my family. This opportunity seems to improve the first but reduce the other two. Help me explore the trade-offs and challenge any assumptions I may be making.”

The difference is subtle but important. You’re not asking the machine to make the decision: you’re asking it to help you think.

A value is more than a word

Another complication is that, even when we believe we’re clear about our values, the words themselves can conceal a surprising amount of ambiguity.

I see this regularly when working with leadership teams. Several people may choose words such as trust, respect, professionalism or honesty, apparently suggesting a high degree of agreement. But it’s only when you ask what those words actually mean to each person that the differences begin to emerge.

Take integrity. For one person it may mean telling the truth regardless of the consequences; for another it means keeping commitments, being transparent about uncertainty, admitting mistakes, or refusing to act in a way that conflicts with deeply held principles. Our cultural background, politics, upbringing and worldview all shape those meanings. So, if we tell an AI that integrity matters to us without explaining what we mean by it, we may be giving much less guidance than we think.

The same applies to words such as success, fairness, loyalty, freedom and responsibility. They feel precise because they’re familiar, but they can carry very different meanings from one person to another.

Translating values into behaviour

That subjectivity around meaning is why I’ve always encouraged organisations to translate values into behaviour. A company value such as “respect” only becomes genuinely useful when people understand what respectful behaviour looks like in practice. The same principle applies when we interact with AI.

If curiosity is one of my values, for example, I might explain that it means remaining open to evidence and alternative explanations, and ask an AI to challenge me when I appear to be settling too quickly on one conclusion. Where integrity matters to me, I could ask it to distinguish clearly between evidence, inference and opinion, rather than making an argument sound stronger simply because I appear to favour it. Personal agency, meanwhile, might lead me to ask for help examining important decisions without taking those decisions away from me.

That’s a very different approach from simply typing a list of five values into a prompt.

It also introduces an interesting discipline. By asking the AI to work with our values, it forces us to articulate those values more clearly ourselves. In that sense, the process can become as useful for the human as it is for the machine.

Why personal values matter more as AI improves

A much larger debate taking place about AI alignment is about how to ensure increasingly capable systems act in ways that remain consistent with human intentions and values.

But another, more personal alignment problem receives far less attention: we want AI to work in ways that are consistent with our values, but how clearly are we articulating those values in the first place?

Daniel Susskind recently made an interesting point when considering how we should prepare young people for an AI-rich future. Rather than trying to predict which particular technical skills will remain immune from automation, he argues for retaining strong human fundamentals and developing the ability to work both with AI and without it.

I think values belong among those fundamentals.

Literacy helps us evaluate what AI writes and critical thinking helps us interrogate what it tells us, but values help us decide whether the options it presents are consistent with the kind of person we want to be and the kind of life we want to create.

Paradoxically, the more capable AI becomes, the more important that may be. An intelligent system may be able to generate dozens of possible routes towards a goal, but it can’t (and shouldn’t) relieve us of the responsibility for deciding whether the goal itself is worth pursuing.

A simple experiment in naming your values

Before your next important conversation with an LLM, try identifying four or five things that genuinely matter to you in that particular context. Then ask yourself what each value means to you, what behaviour would demonstrate it, what would contradict it and how you would want an AI thinking partner to support it.

Then consider where those values might conflict.

That final question is important because values clarity does not eliminate difficult choices: it allows us to recognise them. In that earlier work on values, I’ve described how becoming conscious of what matters to us can help us choose a more constructive response rather than simply reacting to what we feel in the moment, and the same principle applies here.

The purpose of articulating your values to an AI is not to hand your moral compass to a machine. It’s almost the opposite. It’s to make the compass visible enough that the tech can help you notice when you may be drifting away from it.

AI is becoming increasingly capable of helping us decide how to achieve things. We have to be careful about allowing that capability to determine which things are worth achieving.

Before asking AI to help us make better decisions, perhaps we need to become clearer about what better means.

Food for thought

• Before your next big decision, have you actually named the four or five things you want an AI to protect, not just the outcome you want it to help you reach?
• If someone looked at the last month of your conversations with an AI, would your values be visible in what you asked it to do, or only in what you hoped it would tell you?

Remember, when it comes to getting the best results from your AI, stay curious.