Skip to content

Privacy and preparation

Advertising claim substantiation data: what a claim approval case needs

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

A marketing claim approval training case needs the claim as drafted and as run, each meaning a reader would take from it, the evidence held before launch mapped to each meaning, the reviewer's verdict, every revision with its reason, the sign-off chain, the rights on each asset and the post-launch outcome. Finished copy without that file shows style, not the substantiation decision.

Provenance trail

Advertising claim

This page covers advertising claim substantiation data: what a claim approval case needs.Illustrative

Key takeaways

  • The unit of a claim review case is one claim and one meaning paired with its evidence, not the whole ad.
  • Record the evidence held at review time; support found later is a separate, dated item.
  • Narrowing edits made because evidence fell short are the most valuable revisions to capture.
  • Conditional approvals carry rules the final copy does not show, so record the condition.
  • Agency access to studies and creative rarely equals the right to license them.
  • Hold back live disputes, privileged advice and medical claims unless a specialist clears them.

What evidence belongs in a marketing claim approval training case?

A useful case holds the claim exactly as it was proposed and as it finally ran, every meaning a reviewer read into it, the evidence on file before launch, a line-by-line link from each claim to the evidence that supports it, the revisions that changed the wording and why, who approved what, the rights attached to each asset, and what happened after launch. Strip any of those and the record stops showing a decision and starts showing only copy.

The core unit is not the ad. It is the pair of a single claim and the evidence judged to support it, plus the verdict a qualified reviewer reached. One print ad can carry five claims; one study can support two of them and say nothing about the third. A training case that keeps those pairs separate lets a model learn the judgment that matters: whether this sentence is backed by that proof, at the level the sentence promises.

Why the substantiation file teaches more than the finished copy

Approved copy shows what a brand was allowed to say; the substantiation file shows why. The Federal Trade Commission's policy statement on advertising substantiation treats an objective claim as a representation that the advertiser had a reasonable basis for it, and expects that basis to exist before the ad runs. The FTC's guide for small businesses repeats the point in plain terms: proof comes first, and the kind of proof depends on the claim.

That timing rule is why the file is the training signal. A record that pairs the draft claim with the evidence held on the day of review captures the actual decision. A record built later, with evidence gathered after a complaint, teaches something different, because the original reviewer never saw it.

The same policy statement says that when an ad names a level of support, such as tests prove or studies show, the advertiser needs at least that level. When no level is stated, the reasonable-basis analysis weighs factors including the type of claim, the product, the consequences of a false claim and what experts in the field consider reasonable. Those factors are exactly the context a dataset has to carry, because the same sentence can pass for a dish towel and fail for a supplement.

How to break one ad into claims a reviewer can check

Split each ad into atomic claims before you map any evidence. A claim is one statement a reasonable reader would take away, whether it is printed or implied by an image, a comparison or a testimonial. Reviewers in regulated categories already work this way; the dataset only has to record it.

The FTC's Green Guides, at 16 CFR 260.2, tell marketers to identify every express and implied claim an ad reasonably conveys and to make sure each reasonable interpretation is truthful and supported. Its endorsement guides, at 16 CFR 255.2, add that an ad built on a customer's experience with a key attribute will likely be read as saying that result is typical, and that consumer endorsements are not themselves competent and reliable scientific evidence. Those two rules alone produce several claim types with different evidence needs.

How to break one ad into claims a reviewer can check
Claim typeWhat a reader takes awayEvidence a reviewer expectsGap that often flips the verdict
Express performanceThe product does the stated thingTest or study that measured that outcome under comparable conditionsTest measured a related outcome, not the one claimed
Level of supportTests, studies or experts back the claimAt least the stated level, such as a study when the ad says studies showOnly internal data or a single small pilot
Implied from image or contextA demonstration shows real-world resultsProof the demonstration is representative and not staged beyond disclosureImage retouched or conditions not disclosed
Testimonial typicalityThis customer's result is what most people getData on generally expected performance, or a clear disclosure of itGeneric results-not-typical line with no supporting data
ComparativeBetter than a named or implied rivalHead-to-head test on the attribute compared, current productsCompetitor product reformulated since the test
EnvironmentalRecyclable, compostable or lower impactCompetent and reliable scientific evidence for each interpretationClaim true for one component, read as the whole package
Health or safetyThe product affects health or is safeCompetent and reliable scientific evidence evaluated by qualified peopleCustomer letters or a mechanism study offered instead of outcome data

The claim to substantiation mapping dataset brief

Use the brief below as the field list for every case. Each row describes one claim, so a single campaign becomes several rows that share a campaign identifier. The allowed values keep labels consistent across reviewers and across agencies, which is what makes the cases comparable once they are combined.

At a glance

The claim to substantiation mapping dataset brief

  • Reason codes for revisions: evidence did not reach the stated level; implied meaning not supported; typicality data missing; comparison outdated; client preference with no evidence change; brand voice only.
  • Evidence types: controlled test, consumer perception study, published paper, third-party lab report, internal performance data, expert opinion, customer testimonial, certification document.
  • Minimum completeness: a case needs draft and final text, at least one conveyed meaning, a verdict per meaning and the approval chain; anything less is kept as context, not as a labeled case.
The claim to substantiation mapping dataset brief
FieldWhat it recordsAllowed values or formatRule
claim_id and campaign_idOne claim and the campaign it belongs toStable internal identifiers, no client namesNever reuse an identifier after a claim is split or merged
claim_text_draft and claim_text_finalWording at first review and as approvedVerbatim text, with channel and run dateKeep both even when identical; identical is a result
conveyed_meaningsEvery reasonable interpretation a reviewer listedOne line per meaning, tagged express or impliedAt least one meaning per claim; implied meanings need a reason
claim_typeCategory that sets the evidence barExpress, level of support, implied, testimonial, comparative, environmental, health or safetyPick the strictest type that applies
evidence_refsEach item of support linked to the meaning it supportsDocument identifier, type, date, author role, rights statusEvidence dated after first run is flagged, not hidden
fit_verdictReviewer judgment per meaningSupported, supported if narrowed, unsupported, needs specialistOne verdict per meaning, never one per ad
revision_logEach wording change and its causeTimestamp, from text, to text, reason code, requester roleReason codes from a fixed list, free text only as a note
approval_chainWho signed off at each stageRole, stage, decision, date; no personal namesRecord rejections and conditional approvals as well
asset_rightsRights for every image, voice, study and testimonialOwner, license scope, expiry, release on file yes or noUnknown rights block delivery of that asset
post_launch_outcomeWhat happened after the ad ranRan unchanged, modified, withdrawn, challenged, unknownUnknown is allowed; a guessed outcome is not

Which revisions and sign-offs change what a case can teach

The revision that matters most is the one where wording changed because the evidence did not support the first version. Those narrowing edits, such as moving from clinically proven to tested in a lab study, are the clearest examples of claim-evidence judgment a model can see. Edits for tone or length belong in the log but carry little signal, so the reason code has to separate them.

Sign-offs change what a case can teach when they are conditional. A legal reviewer who approves a claim only with a disclosure, or only for one market, has created a rule the copy alone does not show. Capture the condition as part of the approval record, not as a comment someone has to find later.

Outside review sharpens the label. In the self-regulatory process run by BBB National Programs, the National Advertising Division places the burden of proof on the advertiser and its FAQ notes that summaries or excerpts of test results are unlikely to give enough information to judge reliability. A case where a reviewer asked for the full study rather than a summary teaches the same discipline, so record which version of the evidence the reviewer actually saw.

The FTC's health products guidance gives a useful exception to watch for. A claim that implies scientists agree is false unless the advertiser has evidence of that consensus, even if other research supports the product's effect. Cases like this are valuable precisely because the evidence was strong but did not match the level the words promised.

Who controls the creative, the studies and the client files?

Access to an approval archive is not the same as the right to license it. An agency may hold drafts, studies and approvals in its project system and still not own most of them. Four questions are separate: who can open the files, whether any personal data in them may be used for this purpose, who owns the copyright or the study, and what the client and vendor contracts allow. De-identifying a file answers part of the privacy question and none of the others.

Copyright in agency creative depends on how it was made. The U.S. Copyright Office's Circular 30 explains that a work is made for hire either when an employee creates it within the scope of the job or when a commissioned work falls into specific categories and both parties sign a written agreement saying so; if any requirement is missing, it is not a work made for hire. Many agency contracts then assign rights to the client, which can leave the agency with permission to keep files but not to license them.

Substantiation evidence is often the least licensable part of the file. Client studies, third-party lab reports and consumer surveys usually arrive under confidentiality terms, and testimonials carry the endorser's own release. Record each asset's rights status in the brief and treat unknown as a block, not a default.

  • Agency-created drafts and internal review notes: often agency-held, check client contract assignment clauses.
  • Client-provided studies and data: usually the client's, under confidentiality; needs client permission or exclusion.
  • Stock imagery, music and talent: licensed for the campaign, rarely for dataset use; exclude or reference by identifier only.
  • Testimonials and consumer survey responses: personal data plus a release written for advertising use; review consent scope before inclusion.

Worked example: a laundry stain claim from first draft to approved copy

This example is illustrative and does not describe a real campaign. A household products client asks for a print and social ad saying a stain remover removes 99 kinds of stains in one wash, dermatologist tested, with a customer quote saying it saved a favorite shirt.

  • Claim split: four claims (stain removal range, one-wash performance, dermatologist tested, customer result) and seven conveyed meanings, including the implied meaning that the quoted result is typical.
  • Evidence mapped: one lab test covering 62 stain types under standard wash conditions, one dermatologist patch test summary, and the customer's signed release.
  • First verdicts: stain range unsupported as written; one-wash supported if narrowed to tested stains; dermatologist tested supported, with a note that only a summary was on file; typicality unsupported.
  • Revisions: two wording changes over 2 rounds and 6 working days. The count became removes over 60 common stains and the quote gained a disclosure of generally expected results drawn from the same lab test; the full patch test report was also requested and filed.
  • Approval chain: account lead, then agency legal (conditional on the disclosure), then client regulatory, then client brand.
  • Outcome: ran with the modified wording; no challenge recorded within the 90 days the agency tracked.

One exception from the same example

The client later supplied a newer lab test supporting the original stain count. It is tempting to relabel the first claim as supported. Do not. The reviewer judged the claim on the evidence held at review time, and the policy statement says later evidence of truthfulness does not absolve an advertiser that lacked a reasonable basis when the ad ran. The right record keeps the original verdict, adds the later study as post-review evidence with its date, and lets the buyer decide whether to use that pair as a second case.

The other limit in this example is rights. The lab tests belong to the client and arrived under a confidentiality clause. Unless the client agrees, the case can describe the evidence by type, date and what it measured, but cannot include the reports themselves. A case with described evidence is weaker but still teaches the narrowing decision.

Cases to hold back, narrow or send to a specialist

Some files look rich but should not go into a dataset as they stand. Hold back or narrow them when the decision depended on facts the record cannot carry, or when sharing would breach a duty the agency still owes.

At a glance

Cases to hold back, narrow or send to a specialist

  • Live disputes: claims under an active challenge, regulator inquiry or litigation hold stay out until counsel clears them.
  • Privileged advice: legal memos written to advise on a claim may be privileged; keep only the outcome and reason code unless counsel approves more.
  • Drug, device and medical claims: these sit under additional regulators and specialist review; record them only with a qualified regulatory reviewer.
  • Consumer research with identifiable respondents: remove or aggregate respondent data before any review of licensing scope.
  • Cases with no recorded verdict: an approved final ad with no review notes shows an outcome without the reasoning, so it belongs in context, not in the labeled set.
  • Single-client archives: if one client supplies most cases, confidential patterns about that client can show through even after names are removed; narrow the scope or seek that client's agreement.

Licensing an approval archive rather than a portfolio

Agencies and in-house teams that keep review files hold a rare kind of record: expert decisions about whether evidence supports a sentence, with the reasons written down. That is different from a portfolio of finished ads, and it is the part buyers training review assistants tend to lack.

Agencies and brand teams can license review archives through SourceX, which handles the rights review, contracting, preparation and delivery a claim file needs before it reaches an AI developer. It starts with a short assessment of the company, no files uploaded, and no case leaves the agency until it has approved the scope and terms and signed. A filled-in mapping brief for a small sample is a practical way to see what that scope could include.

Frequently asked questions

What is advertising claim substantiation data?

It is the set of records that show which evidence supported each claim in an ad at the time it was approved. A useful dataset pairs each claim and its conveyed meanings with the evidence, the reviewer's verdict, the revisions and the sign-offs, rather than storing finished ads alone.

Can approved ads alone train a claim review model?

Not well. Finished copy shows what was allowed but hides what was rejected or narrowed and why. The decision a review model needs to learn sits in the pairing of claim and evidence and in the edits made when the two did not match.

Do customer testimonials count as substantiation?

Under the FTC's endorsement guides, consumer endorsements are not competent and reliable scientific evidence on their own, and an ad built on one customer's result will likely be read as claiming that result is typical. A case should record whether typicality data or a disclosure of generally expected results was on file.

Should evidence collected after the ad ran be included?

Include it, but date it and label it as post-review evidence. The FTC expects a reasonable basis before dissemination, so the original verdict must stay attached to the evidence the reviewer actually had. A later study can form a separate case if the buyer wants one.

Who owns an agency's claim review files?

It depends on employment status, work-made-for-hire agreements and the client contract. Agency drafts and notes may be agency-held, while client studies, stock assets and talent releases usually are not. Counsel should confirm the position for the specific contracts before anything is licensed.

Sources

  • Objective ad claims imply a reasonable basis that must exist before dissemination; stated levels of support set a floor; listed factors set the reasonable-basis bar; later evidence of truthfulness does not absolve a firm that lacked a reasonable basis. Source
  • Advertisers need objective proof before an ad runs, and the type of proof depends on the claim. Source
  • Consumer endorsements on a key attribute are likely read as typical results, and endorsements are not competent and reliable scientific evidence. Source
  • Marketers must identify all express and implied environmental claims and support each reasonable interpretation, often with competent and reliable scientific evidence. Source
  • Health claims need competent and reliable scientific evidence, and an implied consensus claim is false without evidence of consensus. Source
  • In an NAD challenge the burden of proof is on the advertiser, and summaries or excerpts of test results are unlikely to be enough to judge reliability. Source
  • A work is made for hire through employment scope or, for commissioned works, specific categories plus a signed written agreement. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies