FAQ

Setup

I cannot add JavaScript to my Qualtrics questions. Your brand administrator has disabled the JavaScript permission. Send them What we collect and the note for administrators. If external scripts are the concern, the inline header build loads nothing from outside Qualtrics.

Does it work with the New Survey Taking Experience? Yes. The collector works with both the legacy experience and the new one, and it does not need jQuery.

Can I use it together with Prolific's authenticity checks, or other quality scripts? Yes. Lydite writes only its own __js_lyd_* fields and does not change the page's behavior. Run the manual checklist once with both installed before a big study.

How many questions can I monitor? Up to 20 per survey in Qualtrics (__js_lyd_f01 to __js_lyd_f20). Monitor the open-ended questions where AI use matters. Multiple-choice questions give little signal in v0.1.

Does it slow down the survey? No noticeable effect. The collector is under 15 KB gzip and only counts events.

Do I need a secret? No. Without one, a random seed is generated per session, and the traps work the same way; the payload records which word was used. A fixed secret, together with a respondentKey such as a panel ID, makes a respondent see the same word when they come back in a new session. The secret is visible in the page source, so never use a real password.

Respondents

Will respondents notice anything? No. Nothing is shown, and nothing is blocked. The hidden instruction is invisible, skipped by screen readers and does not change the layout. It only appears if someone copies the question text, which is exactly the case it is designed to catch.

What if a respondent has JavaScript disabled? The survey works normally, but there are no integrity data. Those respondents get insufficient_data, and you decide how to handle them. They are not labeled as human or as cheating.

What about respondents who use dictation, screen readers or an input method? Input methods are detected, and the typing rules are relaxed for them. Dictation and some assistive tools can produce no-keystroke flags of medium strength, which lead to review on their own, not exclusion. Check the subgroup table in the report.

Analysis

Can I reject a submission because Lydite flagged it? No. Use flags only to exclude data, and pay people who made a genuine effort. See Responsible use.

What threshold should I use? Use the defaults unless you have a reason to change them, and pre-register them. They are not yet calibrated on benchmark data, and the report states this.

Does the analyzer upload my data? No. The offline page is forbidden by its own Content-Security-Policy from making any request. The R, Python and command-line analyzers read and write local files only.

Why do R, Python and the web page give the same results? They implement the same specification and are tested against the same shared test vectors. CI checks that their output files are byte-identical.

Does Lydite detect AI-written text? Not from the text itself. It looks at how the text got into the answer box, plus the decoy word. Text detectors are unreliable, and Lydite does not use one.