The System Usability Scale was built for software that behaves the same way every time. AI doesn’t. AI-UX Score is a respondent-backed assessment of how people experience your AI across seven dimensions of trust and usability, the things SUS was never designed to see.
First project free. No card. PDF and CSV export.
Classic UX research measures how fast someone finishes a task, how many clicks it took, and whether the interface was clear. Those questions still matter, but they assume a system that responds the same way every time. AI is probabilistic, adaptive and occasionally wrong in ways a user cannot predict.
Built for software that behaves the same way every time.
Built for systems that are probabilistic and sometimes wrong.
When your product’s behaviour is probabilistic, you need a metric built for probability: one that measures perceived trust, control and accountability, not just whether the button was easy to find.
Real users rate their experience right after using your AI feature. Their answers roll up into seven dimensions.
Whether people believe the system is reliable and safe to depend on.
Ask: Do users rely on the output when it actually matters?
Whether people can tell what the system is doing, how sure it is, and when they are dealing with AI at all.
Ask: Can users tell how confident the AI is, and when it is the AI talking?
Whether people can steer, correct, edit or override the system.
Ask: Can users override the AI when it gets things wrong?
Whether it is clear who is answerable when the AI is wrong, and whether there is a fallback path or a route to recourse.
Ask: When the output is wrong, does the user have somewhere to go?
Whether the system treats comparable people comparably, and holds output quality steady across different users and inputs.
Ask: Does everyone get output of the same quality?
Whether people feel their inputs are safe, and understand how their data feeds or trains the model.
Ask: Do users feel safe with what they type in?
Whether the output is accurate and fluent, and whether getting there costs more effort than the result is worth.
Ask: Is the result good, and worth the effort it took?
Not a spreadsheet of averages, but a living report your team can read at a glance and take straight into planning.
Acceptable overall, but real friction is showing. Focus on the weakest dimensions.
24 responses
The shape tells you where trust leaks before you read a word: one headline score, and a seven-axis radar anyone can act on.
"People don't trust it" becomes "Accountability 2.9, Privacy 3.5." Every dimension gets its own score and grade; open any row to read the questions behind the number.
Run it again after every fine-tune or redesign. Ten rounds in, the number that used to be a gut feeling is a line you can defend.
Take one clean PDF into a design review or leadership deck, or pull the raw responses into your own analysis. Both come on any plan; the reliability stats that back every number, Cronbach's α and 95% confidence intervals, come with Pro.
A score is the start, not the point. Here's what the assessment turns into: a clear diagnosis, and a backlog you can ship.
Every weak dimension becomes ranked, concrete design work, grounded in what that dimension measures. A plan to start from, not a verdict to sit with.
The action plan gives AI-generated suggestions: a starting point to review with your team, not a substitute for talking to users. A Pro feature.
Generate a lightweight assessment link, or embed the survey right inside your product flow. Users answer in under five minutes.
Responses synthesise automatically into your seven-axis radar, with a sub-score for every dimension and a read on how much confidence your sample supports.
Get a prioritised gap report: exactly where trust is leaking, and which design changes to make next.
for your first 3 months, then €29/mo.
Book an AI-UX audit: find where users lose trust, and get a blueprint to fix it.
They are companions, not rivals. SUS gives one usability number, validated on conventional software where behaviour is fixed. AI-UX Score is built for probabilistic systems: seven dimensions SUS never covers (trust, transparency, agency, accountability, fairness, data safety, quality). Use SUS for the interface, and AI-UX Score for the AI. Many teams run both.
Around 25 gives a reliable read for most teams, and the tool tells you how much confidence your sample supports. Fewer gives a directional signal; 50 to 100+ is better for benchmarking over time.
Yes. The seven dimensions are fixed (that keeps scores comparable), but each item has an [AI system] placeholder you set to your context ("this assistant", "the recommendation engine").
Yes. Your first project is free: one project, up to 25 responses, no card, enough to run a real assessment and get your score and grade across all seven dimensions. Pro lifts the cap to unlimited responses, projects, and rounds.
Your real users, ideally right after they use the AI feature. Gather them from your users, not your team: a group rating its own work is exactly the self-assessment this instrument exists to replace.
Yes. Generate a lightweight assessment link, or embed the survey right inside your product flow. Users answer in under five minutes.
Set up your first AI-UX survey in under two minutes. No card.
Something else? Send a note