There is a subtle but important difference between “help me think through this decision” and “tell me what I should value here,” and it is easy to slide from the first into the second without noticing, especially when a model answers both questions with the same confident, well-organized prose. This article is a workflow for staying on the right side of that line: using AI to clarify and stress-test values you already hold, without asking it to manufacture values for you or to settle a genuine values conflict on your behalf.
Why this distinction matters
Psychologist Shalom Schwartz’s widely used theory of basic human values identifies ten motivationally distinct values — among them self-direction, security, benevolence, achievement, and tradition — that are recognized across cultures, and shows that pursuing one value routinely creates tension with another: security and self-direction, for instance, or achievement and benevolence (Schwartz, “An Overview of the Theory of Basic Values,” Online Readings in Psychology and Culture, 2012). The practical implication is that most hard personal decisions are not “which option is objectively correct” but “which of two things I genuinely value am I willing to weight more heavily this time” — a question that requires knowing your own values well enough to name the tension, not a question any external tool can answer for you.
This is also why asking AI directly “what should I value” or “what should I do” on a genuine values conflict is a structurally different request than asking for decision support. A model has no values of its own to draw on beyond patterns in training data and tuning toward broadly agreeable answers; when asked to resolve a conflict between your own competing values, it will produce something fluent and confident, without any actual standing to make that call for you. The EU AI Act’s prohibition on AI systems that use manipulative or deceptive techniques to materially distort a person’s decision-making, causing them to take an action “they would not have otherwise taken” (EU AI Act, Article 5(1)(a), Regulation (EU) 2024/1689), is aimed at a different and more severe problem than this article addresses, but it points at the same underlying principle: the line between helping someone reason and substituting for their own judgment is one worth guarding deliberately, not something to trust a tool to respect on its own.
What AI is genuinely good at here
Used correctly, AI is a strong tool for three specific tasks in values-based decisions:
- Naming values you have not articulated. Asked the right way, a model can help you turn a vague feeling (“this doesn’t sit right with me”) into a named value or principle you can then examine and weigh deliberately.
- Testing consistency. Given a value you claim to hold and a decision you are leaning toward, a model can point out where the two do not actually line up, or where you are applying the value selectively.
- Surfacing the tradeoff explicitly. Many hard decisions feel harder than they need to because the real tension — security versus autonomy, loyalty versus honesty, ambition versus rest — has never been named plainly. A model is good at helping produce that plain statement, even though only you can decide how to weight it.
What it cannot do: supply the value itself, tell you which of two genuine values should win in your specific situation, or take on any of the responsibility for the choice once it is made.
A worked example
I am trying to decide between staying in a stable job and taking a
riskier opportunity. Here is what I've told myself matters to me:
[list]. Before giving me any recommendation, do two things: first,
name the specific values in tension in this decision (not a general
list, the ones actually present in what I described). Second, ask me
which of those specific values I would protect first if I truly had
to choose only one, and wait for my answer before saying anything else.
Notice what this prompt does not ask for: it does not ask the model which option is better. It asks the model to name the tension and then wait for the human to actually make the value judgment, rather than defaulting to producing a confident recommendation the person can quietly outsource their responsibility to.
The values-conflict worksheet
| Step | What you do | What AI does |
|---|---|---|
| 1. Name your values, unaided | Write 3-5 values relevant to this decision, in your own words, before opening any AI tool | Nothing yet — this step must be yours |
| 2. Test consistency | Describe the decision you are leaning toward | Points out where the leaning decision does or does not match the values you named |
| 3. Name the conflict | — | States plainly which two named values are actually in tension here |
| 4. Weigh, yourself | Decide which value you are choosing to weight more heavily this time, and why | Nothing — this is the one step no tool should do for you |
| 5. Document | Write down the conflict and your reasoning, dated | Can help format the record, not generate its content |
The full values-conflict worksheet gives you this table with space for your own decision, plus a place to log a real human conversation about the choice — not as a formality, but because values conflicts, unlike factual questions, usually benefit from being said out loud to someone who knows you.
Do not use this workflow for crisis decisions, decisions made under coercion, or regulated decisions with legal or professional requirements attached. It is built for ordinary, non-emergency choices between competing goods you hold yourself — not for situations where safety, law, or someone else’s authority over the decision changes the picture.
A common misconception
The misconception is that a “good enough” values framework will eventually produce a clean, tie-breaking answer, and that AI’s fluency at producing confident-sounding conclusions is evidence it has found one. Most real values conflicts do not resolve cleanly — Schwartz’s own research is explicit that competing values create structural tension precisely because pursuing one has real costs to another, not because you have not yet found the right framework. The goal of this workflow is not to eliminate the discomfort of the trade-off. It is to make sure you are the one making it, with the tension named plainly, rather than deferring it to whichever answer sounded most confident.
This connects directly to the broader decision workflow in using AI for better decisions — that article’s devil’s-advocate step and this article’s consistency-testing step are complementary: one stress-tests your reasoning about facts and options, this one stress-tests your reasoning about what you actually value. It also overlaps with what not to delegate to AI: a genuine values conflict fails the “authorship” test in that framework, because the choice needs to be recognizably yours, made for your own reasons, not adopted because a model made a plausible case for it.
Try it today
Pick a decision you are currently sitting with that involves two things you genuinely care about pulling in different directions. Write your relevant values down yourself, unaided, before opening any AI tool. Then use the worked prompt above to name the specific tension and test whether your leaning is actually consistent with the values you wrote down. Make the final weighting call yourself, write it down with today’s date, and if the decision is significant, say your reasoning out loud to one person who knows you before acting on it.
