Question 1

What is the toolkit?

Accepted Answer

A suite of tools for building AI products people trust: anticipate what users will need, measure whether they trust what you shipped, map who's actually using it, and get your bearings on the EU AI Act, all in one place.

Question 2

Do I have to use all four tools?

Accepted Answer

No. Each stands on its own. Reach for whichever question you're asking. They share one worldview (human-centred, evidence-led), so using more than one compounds.

Question 3

Is it free to start?

Accepted Answer

Yes. You can run your first project in each tool for free, with no card required.

Question 4

Do I need to be technical?

Accepted Answer

No. The tools bring the methods and the maths; you bring what you know about your product and your users.

Question 5

Who is it for?

Accepted Answer

Product and design teams building AI-powered products, and anyone responsible for whether those products earn trust and meet the rules.

Question 6

Where is my data stored?

Accepted Answer

Your data is hosted in the EU and handled GDPR-first; payments are secured by Stripe.

Question 7

Is there a free plan?

Accepted Answer

Yes, €0, forever. It gives you everything you need to try each tool on one project, with no card.

Question 8

How much is Pro?

Accepted Answer

€29 / month, or €290 / year (yearly saves ~17%, two months free). A launch offer takes 50% off your first three months, from €14.50 / month.

Question 9

What does Pro unlock?

Accepted Answer

Unlimited projects, trends over time, reliability stats, shareable stakeholder reports, and AI-assisted canvases, everything across the toolkit.

Question 10

Can I cancel anytime?

Accepted Answer

Yes. Try every tool free first, and cancel anytime. Your subscription is managed through Stripe.

Question 11

Is there a team plan?

Accepted Answer

A Team plan (€99 / month, up to 5 members) is coming soon, as an upgrade from Pro.

Question 12

Which plan am I on?

Accepted Answer

You start on Free automatically. Upgrade to Pro from the Plans page whenever you're ready.

Question 13

What is the AI-UX Self-Assessment Questionnaire?

Accepted Answer

The AI-UX Self-Assessment Questionnaire is a diagnostic tool that helps organizations measure the perceived quality of AI-powered user experiences.

It was designed to capture what technical metrics can't: how users feel about interacting with an AI system, whether they trust it, understand it, feel empowered by it, and believe it treats them fairly.

In a world where AI is becoming embedded into everything from healthcare to banking, user trust isn't a luxury. It's a requirement.

Question 14

How does the AI-UX Self-Assessment Questionnaire work?

Accepted Answer

The AI-UX Self-Assessment Questionnaire works by asking users to rate their experience with an AI system across seven key dimensions (such as trust, transparency, and fairness) using simple, statement-based prompts.

Each dimension includes 3–5 Likert-scale items (e.g., *"I trust the recommendations made by the system"*) that capture how users perceive and interact with the AI in real-world contexts.

Once responses are collected, you can process the scores in this calculator and analyze them to identify strengths, surface experience gaps, and guide future improvements.

Question 15

What does the AI-UX Self-Assessment Questionnaire measure?

Accepted Answer

The AI-UX Questionnaire measures how users perceive the quality, reliability, and integrity of their experience with an AI system.

It evaluates seven critical dimensions (Trust, Transparency, Agency & Control, Accountability, Fairness, Privacy & Data Practices, and Perceived Quality & Ease of Use) to capture how well the system aligns with human expectations, supports autonomy, and fosters long-term engagement. Together, these dimensions provide a holistic view of the system's impact on user confidence, satisfaction, and ethical acceptance.

Question 16

When should I run an AI-UX Self-Assessment Questionnaire study?

Accepted Answer

You should run an AI-UX Questionnaire study at key moments when you're looking to evaluate or improve the user experience of an AI-powered system. This includes:

- **Before or after launching a new AI feature**, to gather feedback on user perceptions and readiness.
- **During usability testing**, to supplement qualitative insights with structured perception data.
- **At regular intervals**, to monitor how trust, control, and satisfaction evolve over time.
- **After major updates**, especially if they affect automation, personalization, or decision-making logic.
- **When adoption is low or user feedback is unclear**, to identify hidden friction points or trust barriers.

Question 17

What are the pros and cons of the AI-UX assessment?

Accepted Answer

**Pros**

- **Structured insight across 7 key dimensions:** measures critical aspects like trust, transparency, control, and fairness, a holistic view of user perception.
- **User-centered diagnostic tool:** captures subjective experience, helping teams understand how users feel, not just what they do.
- **Actionable and benchmarkable:** generates scores that can be tracked over time or compared across products and user groups.
- **Flexible format:** available in PDF, Google Forms, and Tally, easy to embed in any workflow or research stack.
- **Non-invasive and easy to deploy:** short and straightforward for users to complete, typically under 5 minutes.

**Cons**

- **Relies on self-reported data:** subjective perceptions may not always align with actual behavior or system performance.
- **Requires contextual interpretation:** results need to be read in relation to system type, user group, and maturity level. There are no universal "good" scores (yet).
- **Not a substitute for usability testing:** best used alongside other UX methods (like interviews or behavior tracking) rather than in isolation.
- **Benchmarking still in progress:** while the tool supports score tracking, industry-wide benchmarking is still being developed.

Question 18

Can other factors affect the AI-UX score?

Accepted Answer

Yes, several external and contextual factors can influence the score beyond the system's actual design or performance:

1. **User expectations and prior experience:** users with past negative experiences with AI (or very high expectations) may rate a system more critically, regardless of actual performance.
2. **System maturity:** early-stage or experimental AI systems may receive lower scores due to bugs, limited capabilities, or a lack of refinement.
3. **Use case and domain sensitivity:** AI in high-stakes domains (e.g., healthcare, finance) tends to be judged more strictly than in low-risk applications (e.g., entertainment, shopping).
4. **Cultural and demographic differences:** perceptions of trust, fairness, and privacy vary across cultural and social contexts.
5. **Communication and onboarding:** if the system is poorly introduced or lacks clear explanations, users may feel confused or skeptical, even if the AI performs well.
6. **Device and interface constraints:** a good experience on one platform (e.g., desktop) may not translate to another (e.g., mobile), skewing scores by access channel.
7. **Recent incidents or media coverage:** public scandals or news about AI misuse can prime users to distrust AI in general, impacting their answers even if your system is compliant and ethical.

Question 19

Can I edit the wording of a standardized UX questionnaire?

Accepted Answer

Best practices generally recommend **not changing the wording of standardized items**, especially in validated instruments like the System Usability Scale (SUS), UEQ, or NASA-TLX. These tools have undergone statistical validation, and changes can affect their reliability, validity, and comparability.

**But it's OK to edit standardized UX questions … to a point.** Some customization is expected, particularly inserting **product- or service-specific references**. For example:

- It's common in SUS to replace generic terms like "the system" with the actual product name (e.g., "this chatbot" or "SmartLearn").
- The same applies here: you'll notice placeholders like **[AI system]** in most of the 23 items. These are meant to be **adapted to your specific usage context** (e.g., "this AI-powered assistant," "the recommendation engine," or "AutoBudget").

This contextual tailoring helps participants relate to the questions without compromising their meaning, **as long as the core intent of the item remains unchanged**.

Question 20

Should I weight the scores in the AI-UX questionnaire?

Accepted Answer

**Most of the time, no.** The questionnaire is designed to give you **simple average (mean) scores** across dimensions and overall. In most cases, unweighted averages are **good enough**, especially if your sample roughly reflects your real user base.

The **fundamental reason to use weights** is when some responses should **influence the overall results more or less than others**, because your sample **doesn't reflect the real composition** of your user base, or because some groups are **deliberately overrepresented** (e.g., novice users in a usability study).

**When should I consider weighting?** Only if your sample **doesn't represent your user base well** *and* you have **reliable reference data** about your actual user population. For example, if your sample skews toward experts but your real users are mostly beginners, you might adjust scores to reflect the true landscape.

| Reason | Example |
| --- | --- |
| **Unintended sample imbalance** | Your survey accidentally includes too many advanced users, but your customer base is mostly new users. |
| **Deliberate oversampling** | You intentionally included more beginners, but now need to adjust scores to reflect the real user mix. |
| **Defined importance differences** | You want to give more weight to feedback from high-impact use cases, regions, or user types. |

**When *not* to use weights:** when your sample proportions closely match your user base; when you care more about group comparisons than overall metrics; when you don't have reliable reference data; when key-group sample sizes are very small (weighting amplifies noise); or when precision matters more than representativeness.

> Weighting improves how *representative* your overall scores are, but it can reduce how *precise* they are.

Unless you're making high-stakes decisions or publishing externally, it's usually better to get the sample right up front than to fix it with weighting after the fact.

Question 21

How do I calculate a weighted mean?

Accepted Answer

**Group-level weighting (simpler).** If you know the group averages, use:

```
Weighted Mean = (Group 1 Mean × Group 1 Weight) + (Group 2 Mean × Group 2 Weight)
```

Make sure your weights add up to 1. For example, novice users (75% of real users): avg 70.8; expert users (25%): avg 76.9:

```
Weighted Mean = (0.75 × 70.8) + (0.25 × 76.9) = 72.3
```

**Case-level weighting (more precise).** If you have individual responses and know each group's proportion in your sample and population:

```
Group Weight = Reference Proportion / Sample Proportion
```

Then multiply each participant's score by their group weight before averaging. You get the same result as the group-level method, with more flexibility for advanced analysis (e.g., regressions or filtering multiple variables).

Question 22

Should I weight percentages in the AI-UX questionnaire?

Accepted Answer

In UX research, percentages usually represent the **proportion of participants** who selected a response (e.g., "Agree" on a Likert item, or "Yes" to a success criterion). If your sample isn't representative (say, too many advanced users and too few novices) you might **weight** these percentages to reflect the full picture.

**Consider weighting percentages when** your sample is unbalanced (by experience, role, geography, etc.) *and* you have reliable reference data on how your real population is distributed. If you don't have a good reference or your samples are small, use unweighted percentages and note the limitation.

**What can be weighted?** Almost any percentage: the % who agreed with an item, the % who rated a dimension positively, or **net scores** (Net Positive, Net Trust) and binary questions. That's because these are all means of coded values (a percentage is the mean of 0s and 1s; a net score is the mean of −1/0/+1), so the same weighting logic applies.

**Do I need to weight all of them?** No. Most teams can work with unweighted percentages during discovery, internal benchmarking, or iterative testing. But if you're comparing across time, reporting to stakeholders, or generalizing to your full user base, group-level weighting is a safe, simple accuracy boost.

Question 23

How do I calculate a weighted percentage?

Accepted Answer

**Group-level weighting (simpler).** With the % of "Agree" responses per group and each group's proportion in the *reference population*:

```
Weighted % = (Group 1 % × Group 1 Weight) + (Group 2 % × Group 2 Weight) + ...
```

Make sure group weights add up to 1. Example: 60% of novices agreed, 80% of experts agreed, and your real base is 70% novices / 30% experts:

```
Weighted % = (0.7 × 60) + (0.3 × 80) = 42 + 24 = 66%
```

**Case-level weighting (more flexible).** Working at the individual level: assign each participant a weight of `Reference Proportion / Sample Proportion`, multiply each score (0, 1, or −1/0/+1) by their weight, then average. This gives precise control, useful when applying weights across multiple variables later (e.g., role and experience).

Question 24

How many users do I need for my study?

Accepted Answer

There's no single "correct" number. It depends on your research goals, the diversity of your user base, and how precise you want your results. As general rules of thumb:

- **20–30 responses**: acceptable for exploratory research, early feedback, or qualitative patterns (common in formative usability studies). Interpret cautiously; results can be noisy or biased.
- **50–100 responses**: sufficient for descriptive statistics, benchmarking across time/systems, or early quantitative trends. (Jeff Sauro at MeasuringU recommends 50–100+ for reliable SUS benchmarks.)
- **200+ responses**: required for regression, subgroup comparisons, psychometric validation (factor analysis, Cronbach's alpha), and reliable benchmarking. Needed if you plan to benchmark across products or publish.

If you want statistical justification, consider a **power analysis** to estimate the required sample size from expected effect size, confidence level, and margin of error.

**Other considerations:** factor analysis needs 5–10 responses per item (with 23 items, that's 115–230+); reliability (Cronbach's alpha) can be done with ~100–150 but stabilizes at 200+; comparative subgroup analysis wants 30–50+ per group.

| Purpose | Minimum N (per best practice) |
| --- | --- |
| Exploratory insights | 20–30 |
| Benchmarking or internal comparison | 50–100 |
| Subgroup comparisons | 30–50 **per group** |
| Regression / reliability analysis | 150–200+ |
| Factor analysis / scale validation | 5–10 **per item** (≈115–230 for 23 items) |

Question 25

What is a power analysis?

Accepted Answer

Power analysis is a statistical method used to determine the appropriate sample size for a study. It helps ensure your study has enough statistical power to detect a meaningful effect (if one exists) and avoids wasting resources on a study that's too small or too large.

- **Statistical power** tells you how likely your study is to detect a real effect *if* one exists (correctly rejecting a false null hypothesis). The null hypothesis is the assumption that nothing is going on: no difference, no effect. A commonly accepted level is 80%.
- **Sample size** is the number of participants; power analysis finds the minimum needed for your desired power.

**Why it matters:** it avoids underpowered studies (which miss real effects and mislead), avoids overpowered studies (unnecessarily expensive), ensures statistically robust results, and supports informed decisions about study design.

Question 26

How do I know my sample reflects my target users?

Accepted Answer

Start by defining the characteristics of your target user base (role, AI experience, industry, region) and compare your sample distribution to those known benchmarks.

If your sample significantly over- or under-represents key groups, consider **stratified sampling** during recruitment (pre-stratification) or **weighting responses** after collection (post-stratification). Just be aware that weighting adds complexity and can increase the variability of your estimates, so use it only when necessary and with caution.

Question 27

How many responses are required per item or dimension for reliability?

Accepted Answer

The questionnaire includes 23 items across seven dimensions. To ensure reliable results:

- Aim for **at least 5–10 responses per item**: a minimum of **115–230 total responses** for full-scale quantitative use.
- For **dimension-level** analysis, at least **30–50 responses per dimension** is recommended.

This ensures composite scores (e.g., for "Control" or "Transparency") are based on enough data to minimize noise and reflect consistent user perceptions.

Question 28

Is there a sample size below which results need caution?

Accepted Answer

Yes. If your total sample is **below 20 users**, results may be too volatile or skewed to interpret confidently, especially for specific dimensions or subgroup comparisons.

Use smaller samples (e.g., 10–15 users) only for qualitative insight or early-stage validation, not for quantitative claims or performance comparisons. Always report the sample size alongside your findings to give context for interpretation.

Question 29

What are the pitfalls of the AI-UX Self-Assessment Questionnaire?

Accepted Answer

**It's not diagnostic.** It measures users' perceived experience across dimensions like reliability, transparency, and adaptability, but it won't tell you exactly what to fix. For actionable insight, pair it with usability tests or user interviews.

**Ceiling effect:** high-performing systems may receive consistently strong scores, making incremental improvements harder to detect, a common limitation of perception-based tools.

**Potential biases:** responses can be influenced by the halo effect or social desirability. To reduce bias, collect responses anonymously and consider randomizing question order.

**Context matters:** scores are shaped by user expectations, familiarity with AI, and context of use. A low score may reflect a mismatch, not necessarily a flawed system.

Question 30

What is anticipatory design?

Accepted Answer

Designing for needs before users can articulate them. Proactive, not reactive. This workspace turns it into a repeatable method (Anticipate → Imagine → Shape), not a blank template you fill in alone.

Question 31

Do I need to know the method already?

Accepted Answer

No, the workspace guides you through each phase, and the method page explains the theory if you want it.

Question 32

Is this just templates?

Accepted Answer

It's a structured method with guidance at every step, not a blank canvas: each phase hands concrete outputs to the next.

Question 33

Who is it for?

Accepted Answer

Product and design teams who want to design proactively, not just react to the backlog.

Question 34

How does it fit with the other tools?

Accepted Answer

It answers one of the toolkit's four questions: 'what will they need next?' The others measure whether people trust it, map who's in front of it, and check it against the rules.

Question 35

What does it map?

Accepted Answer

Your users across five stages of readiness (from not-yet-aware to fully committed) so your AI can meet each person where they actually are.

Question 36

Do I need a behavioural-science background?

Accepted Answer

No, the tool brings the models (Prochaska's stages of change and Fogg's B=MAP); you bring what you know about your users.

Question 37

Where does the data come from?

Accepted Answer

You map from what you observe or already know about your users; the tool structures the thinking.

Question 38

Isn't this just personas?

Accepted Answer

It goes deeper: it's about readiness and the drivers of behaviour, not a fixed profile of who someone is.

Question 39

Does this tell me my product's risk tier?

Accepted Answer

No. Our EU AI Act material is educational: it explains how the Act is structured, the risk tiers, key articles, and the design patterns behind common obligations. It does not classify your product and is not legal advice. To self-check where a system might sit, use an external tool such as the EU AI Act Compliance Checker; for anything binding, consult qualified counsel.

Question 40

What do the risk tiers mean?

Accepted Answer

In short: prohibited (banned under Article 5), high-risk (permitted but subject to the strictest obligations), limited risk (transparency duties, e.g. telling people they're dealing with AI), and minimal risk (no mandatory obligations). The guide explains each with examples.

Question 41

When do the obligations apply?

Accepted Answer

The Act phases in: the prohibited-practice ban from 2 February 2025, general-purpose-AI model rules from 2 August 2025, and most obligations (including Annex III high-risk systems) from 2 August 2026.

Questions & answers