Comparing different approaches to claims' diagnostics
Claims diagnostics starts with four different questions: what the claim means to a consumer, whether the advertised benefit is true, whether the claim appeals, and whether showing the message changes a specified response. One study rarely answers all four.
Which question does the claim need to answer?
| Buyer question | Evidence to request | What the result does not establish |
|---|---|---|
| What does a consumer understand the claim to mean? | Consumer interpretation research using the exact wording and context | Whether the implied product benefit is true |
| Does the product deliver the advertised benefit? | Substantiation matched to the benefit, population and conditions of use | Whether the wording is the most persuasive message |
| Which candidate claim appeals to the target audience? | A defined rating, MaxDiff, choice or monadic task with its response source stated | Realized purchasing or clinical efficacy |
| Does showing claim A rather than claim B change an outcome? | Randomized assignment, a defined outcome and a suitable comparison | Outcomes beyond the tested task, population and setting |
MaxDiff and TURF are analysis or selection methods, not labels that settle causal validity. A monadic test can use randomized assignment. A survey experiment can estimate effects on stated responses. A simulated choice experiment estimates effects within its modeled population. Each result still needs a clear endpoint and external validation for the decision.
What does health-claim substantiation require?
The FTC's Health Products Compliance Guidance concerns evidence for the advertised product benefit, including implied claims. For health benefits, the guidance generally expects randomized controlled human clinical testing, with the evidence matched to the particular claim.
A test showing that a message increases purchase intent does not show that the product improves health. A synthetic choice test, a focus group or a more persuasive claim cannot replace evidence of the underlying benefit. Have the appropriate legal and scientific reviewers assess substantiation before release.
How should a team assess synthetic claims testing?
Ask the vendor to name the response source, task, target population, model version, comparison data and validation date. Request results for the outcome the decision needs, including misses and uncertainty. A correlation with human survey ratings does not establish clinical benefit, purchase lift or accuracy in a new category.
Subconscious's July 2026 causal-fidelity working paper, which is not peer reviewed, reports mean Spearman rank correlation on estimated choice parameters of 0.55 across roughly 300 replicated studies and 0.73 across 43 passing design filters. Those aggregate parameter-rank results are not a claims-specific accuracy percentage or health-benefit substantiation. They do not establish agreement on treatment-effect size.
What should the research brief contain?
Write the exact proposed wording and every implied benefit a consumer may take from it. Separate the evidence for truth from the evidence for interpretation or appeal. Then define the alternatives, respondent source, assignment method, outcome and decision threshold for the message test.
For a simulated comparison, specify the human or live check required before acting. Report the result alongside the study's population and outcome limits. If a message wins on stated choice but the product benefit remains unsubstantiated, the message is not ready to publish.
Book a decision review to scope a product, pricing or messaging comparison and the evidence it needs. Clinical efficacy and legal substantiation require the appropriate specialist evidence.