Trust in Automation / AI (Jian, Bisantz & Drury)
About this scale
A 12-item measure of human trust in automated and AI systems, including 7 trust items and 5 distrust items. Despite predating modern AI, it remains the most widely used trust-in-automation scale and is routinely adapted to AI agents, recommender systems, and large language models.
Sample item(s)
Reliability
Reported internal consistency: alpha 0.85–0.90. As a rule of thumb, Cronbach’s α ≥ .70 is considered acceptable, ≥ .80 good, and ≥ .90 excellent for established measures.
Notes
Foundational classic; combine with newer AI-specific scales (e.g., Glikson & Woolley) for cognitive vs. emotional trust.
Source & citation
How to administer & score it
All 12 items are typically rated on a Likert-type agreement scale (check the original article for the exact anchors — keeping them unchanged preserves comparability with published findings). Score the scale as the mean (or sum) of its items so results stay comparable across studies that use different response ranges. Include at least one attention check in longer surveys, and randomise item order within the scale block to reduce position effects.
In the original validation, reported internal consistency was alpha 0. Always re-estimate reliability in your own sample: report Cronbach's α (and, if you run CFA/SEM, composite reliability and AVE). An α noticeably below the published range usually signals a translation issue, careless responding, or a mis-scored reverse-coded item — not necessarily a bad scale.
Adapting this scale? A 5-point validity checklist
- Keep the response anchors. Changing a 7-point scale to 5 points changes the psychometrics; only do it with a reason you can defend.
- Translate properly. Use forward and back translation with bilingual reviewers, and pilot the translated version before full data collection.
- Change the referent, not the meaning. Adapting "my organisation" to "my university" is fine; rewording items until they measure something else is not.
- Report every change. Reviewers accept adaptations that are documented and justified; silent modifications are a rejection risk.
- Re-test validity. After any adaptation, re-check reliability and discriminant validity (Fornell–Larcker, HTMT) — our reliability & validity guide walks through the thresholds.
Frequently asked questions
Can I use the Trust in Automation / AI scale in my research for free?
Most academic scales are free for non-commercial research use with proper citation, but permission terms belong to the original authors/publisher — check the source article (cited above) and request permission where indicated.
Do I need the full item wording?
Yes — always work from the original publication's item set rather than paraphrases, so your instrument matches the validated version. This page summarises the scale; the citation above is the authoritative source.
How do I build a questionnaire with this scale?
You can assemble it (with consent text, demographics, and attention checks) in the free SmartForm builder, then verify reliability live as responses arrive, and test your full model in SmartSEM.
Use it in your research
Need to check the reliability or psychometrics of this scale in your own data? Use the free DrAIStudio research tools: