Review your study

Note: This feature is only available on AI moderated studies and is currently in open beta. Functionality may change leading up to general availability.

Writing a study that works well for an AI moderator takes a different kind of care than writing one a human moderator will run. Dscout AI can review questions and their settings in your AI moderated study to ensure participants have a good experience during their sessions and you receive quality insights to analyze after.

Note: Dscout AI does not review any screener questions you have in your study or settings on the Overview tab. It specifically looks at your study questions on the AI moderator tab.

To review your study with Dscout AI:

Navigate to the AI moderator tab of your study.

Click the Review your study icon in the top right corner of the builder.

Click Begin review.

Now, Dscout AI assesses the questions and settings across your study. When it's done, its suggestions are grouped into:

Across study: Issues that involve more than one question, like flow or redundancy.
Question-specific: Issues tied to a single question.

Each finding is flagged Critical (red), Warning (yellow/gold), or Suggestion (blue), and you can expand its Details to see Dscout AI's explanation alongside the question as it's currently written:

The resolution options presented depend on whether the fix can be applied in one click. Depending on the suggestion, you may see a combination of the following options:

Accept: Applies the suggested change right away. This can't be undone, so review each suggestion before accepting.
Reject: Dismisses the finding.
Change setting: Jumps you to the setting the finding is about, so you can update it yourself.
Add question: Adds a suggested question directly to your study.
Review question: Opens the question so you can decide how to address the finding.
Delete question: Removes a question the review recommends cutting entirely.

Note: If you've edited a question since the review ran, its card shows the old version next to the new version, and offers Accept anyway in place of a plain Accept.

Re-run a review

If you make changes to your study, you'll see a notification in the review window suggesting you re-run your review.

To re-run a study review:

Open the Review your study panel.

Click Re-run review. A confirmation modal is displayed.

Click Re-run review to confirm.

Re-running replaces the findings from your last review, and per-question flags disappear once the new review starts.

What suggestions will I get?

As with all generative AI tools, it's important to keep in mind what Dscout AI is looking for. This can help clarify any suggestion that may seem at odds with your research goals.

Tip: Dscout AI's findings are a starting point, not a verdict. If you wrote a question a certain way on purpose, you can reject the finding with no consequence.

Here are some examples of what Dscout AI looks for when providing suggestions:

Across study

Redundant questions: Two or more questions collect the same information without a clear reason.
Earlier question primes a later one: A question plants an answer, or frames how a later one gets answered.
Contradictory or illogical order: Questions contradict each other, or the order doesn't fit the objective.
Objective facet not covered: Something your research objective calls for that no question actually addresses.

Question-specific (all questions)

Double negatives: Stacked negatives make it hard to tell what a "yes" or "no" would even mean.
Ambiguous qualifiers: Vague terms like "often," "recently," or "a lot" that mean something different to every participant.
Inclusive phrasing: Language about identity, ability, or circumstance that could alienate part of your sample.
Leading or biased wording: Steers participants toward a particular answer.
Compound or double-barreled: Two separate questions stacked into one, so it's unclear which one the answer is about.
Not aligned to the objective: Doesn't map to anything your study set out to learn.
Placeholder or gibberish: Text that isn't blank, but doesn't ask anything real ("TBD," "test," template text).
Instruction, not a question: Reads as a note to the moderator or AI, not something meant for the participant.
Leading or vague follow-up guidance: The guidance for follow-up probes plants the answer, or is too vague to be useful.

Task questions

Task has no end state: No way for the participant to know when they're done.
Task prescribes the path: Hands over the exact steps instead of a goal, turning discovery into instruction-following.
Several tasks in one: Multiple distinct tasks stacked into a single turn.
No scenario or context: A bare instruction with no reason behind it.
Named stimulus has no link: References a specific site or file, but no task link is attached.
Awkward read aloud: Doesn't hold up when the moderator reads it exactly as written.

Talk out loud questions

Not genuinely open: A question that a single word or a yes/no fully answers.
Awkward read aloud: Doesn't hold up when the moderator reads it exactly as written.

Scale questions

Unlabeled or unbalanced poles: Both ends of the scale are unlabeled, or labeled with poles that aren't true opposites.
Point count is not intentional: More scale points than the topic really needs, or a midpoint that works against your intent.
Scale type doesn't match the construct: For example, forcing a frequency question onto a satisfaction scale.
Wording contradicts the configured range: The question says one range explicitly ("1 to 10") while the scale is set up differently.

Single- and multi-select questions

Options aren't mutually exclusive and exhaustive: Choices overlap, or a clean option is missing.
No usable escape option: No None/Other/N/A for participants who don't fit any listed option.
Wording doesn't match select-one or select-many: A "select all" question reads like pick-one, or the reverse.
Too many options for a spoken turn: More choices than a participant can hold in mind while they're read aloud.
Order bias isn't controlled: Answer order could sway the result, but randomization is off.

To learn more about how Dscout AI can help speed up your research, see What is Dscout AI?.

For information on how Dscout works to keep your data safe when building AI features, see AI privacy and security in Dscout.