Choosing AI Feature Disclosure Wording Before It Ships
On this page
A product or trust-and-safety leader signing off on a new AI feature must choose the exact sentence beside the activation control: which verb voice describes the data use, whether a retention window is stated, and how visible the opt-out is. Check whether existing evidence supports that wording in the intended screen and audience. When interpretation remains uncertain, compare eligible alternatives before release.
The decision comes down to a few specific wording choices
The same underlying data-use fact can be written several ways, and each version lands differently with a reader deciding whether to trust the feature.
| Wording choice | Interpretation hypothesis to check | Possible concern |
|---|---|---|
| "We use your data to improve the product." | May imply ongoing use | Could be broader than the actual practice; disclose the factual scope |
| "We may use your data to improve the product." | May suggest conditional use | Check whether readers understand the actual conditions |
| "Your data is used to improve the product." | May obscure the actor | Name the responsible party where the disclosure requires it |
| No retention window stated | Reader may not understand how long data is retained. | Test comprehension without omitting required information. |
| Opt-out named and reachable from the same screen | May help readers find the available choice | Verify comprehension and access; do not infer adoption effects without evidence |
Legal review assesses the required facts and wording. A comprehension or behavioral study asks a separate question about interpretation and choices. Compare complete, truthful alternatives and specify whether the endpoint is generated or observed.
Getting the wording wrong is costly in both directions
The FTC’s digital advertising disclosure guidance discusses clear placement, proximity, prominence, and understandable language. The April 2023 joint agency statement affirms enforcement of existing laws for automated systems. Have legal reviewers determine the information the feature must disclose before comparing compliant wording.
Treat interpretation and adoption effects as hypotheses until tested. Complete wording can still be misunderstood, while missing required facts must be corrected regardless of a favorable reaction. Review the actual information obligations before comparing eligible alternatives.
What does disclosure copy do?
Disclosure wording explains a data practice and may affect how a reader understands the available choice. Test comprehension of complete, truthful alternatives. Shi and colleagues’ dark-pattern study examines manipulative versus neutral LLM dialogue in a 34-participant scenario study; it does not establish the effect of these data-use disclosures.
How do you test wording before it ships?
Subconscious can compare modeled choices between assigned, complete disclosure variants in a supported audience. The useful output is a contrast between those exact alternatives, with uncertainty and a recorded task and audience. Agree the configuration before treating a disclosure or opt-in task as supported. A separate comprehension assessment needs agreed questions, scoring and evidence that readers understand the actual practice. Actual opt-in behavior needs relevant human or live evidence; a modeled choice does not establish the behavioral effect.
Confirm audience coverage for the proposed disclosure test. Preserve the question through a matched human comprehension and choice study when the stakes warrant it. Research and how we work describe the distinction between modeled comparisons and human evidence.
What does this kind of test not replace?
A wording comparison does not determine whether a retention window matches actual practice, processors are named correctly, or jurisdiction-specific obligations are met. Qualified reviewers must assess those facts and requirements. Preserve required information regardless of which wording receives a stronger generated reaction.
Where to start
Bring the actual feature, data practices, complete wording alternatives, intended screen and endpoint to a Subconscious decision review. Agree which comparison the study can support, which comprehension or behavioral evidence remains necessary, and who owns the release decision. The brief should preserve the approved wording, required facts, audience, task, uncertainty and acceptance rule. Review applied case examples for context rather than treating them as disclosure-specific validation.