Value Metric
Preference Score
Preference Score is the share of people who prefer one option — a design, message, or concept — over alternatives.
What is Preference Score?
Preference Score measures how strongly people favor one option over others when shown a choice — two landing pages, several logos, competing value propositions. It comes from preference testing and quantifies subjective appeal, which behavioral metrics alone cannot capture.
Preference is most useful when paired with the "why": knowing that 65% prefer option B matters far more when you also understand the reasons, so you can generalize the insight beyond the specific options tested.
How to calculate it
Preference Score = Respondents Choosing an Option ÷ Total Respondents × 100
- Respondents Choosing an Option
- People who selected the option
- Total Respondents
- All respondents who made a choice
Worked example
If 130 of 200 people prefer version B, its preference score is 130 ÷ 200 × 100 = 65%.
What good looks like
- Check significanceBeat a coin flip meaningfully
With two options, treat results near 50/50 as no clear preference. Ensure the margin is beyond sampling noise before acting, and always capture reasons.
Why it matters
Preference testing lets you gather directional feedback on subjective choices before committing engineering effort — cheaply and early. It is especially valuable for messaging and visual design, where opinion drives response. Its limit is that stated preference is not behavior; a preferred design should still be validated with a real A/B test when stakes are high.
How to improve Preference Score
Capture the reasons, not just the winner
Ask why people prefer an option so you can extract the underlying principle and apply it broadly.
Confirm preference with behavior
When it matters, validate the preferred option in a live A/B test to make sure stated preference converts.
Frequently asked questions
Does a preference test predict real behavior?
Not perfectly. Preference tests capture stated opinion, which correlates with but does not equal behavior. They are excellent for early, cheap direction on subjective choices, but high-stakes decisions should be confirmed with a live A/B test that measures what users actually do.
How many responses does a preference test need?
Enough that the winning margin is clearly beyond sampling noise. A near 50/50 split among two options usually means no real preference. For tighter margins or more options, you need larger samples to distinguish a genuine preference from chance.