Promotional claims substantiation, explained.

Promotional claims substantiation explained: proving each line of an ad before anyone signs it

One psoriasis banner, from agency draft to first use

1. The briefThe brand team asks the agency for a half-page web banner on Veltrimab's two-year data, for dermatologists.
2. The draftOctober 7, 2026, 8:41 am: three frames and a safety scroll arrive from the agency, due at Thursday's committee.
3. Every claim tied to its pageSeven claims, headlines included, checked against the library and the label. "Simple to stay on" is caught here, or at committee.
4. The committeeMedical, legal and regulatory reviewers read the piece with its marked-up references and sign, or send it back.
5. First use and Form FDA 2253The banner runs, and the company files it with FDA at that moment on Form FDA 2253.

Step 3 is the subject of this guide. The medicine, people and figures in this example are fictional.

Why a headline counts as a claim, and three more substantiation surprises

What claim, claims library, annotated reference and MLR review mean

Fourteen terms are defined in the one-page summary.

Nine kinds of promotional claim, from a Week 16 percentage to "simple to stay on"

In shortA claim is whatever a reasonable reader takes away, written or implied. FDA reads the whole piece and asks what it leaves in the mind. So a substantiation check has to cover headlines, subject lines, chart titles, images and video voiceover as well as body copy, and it has to split a long sentence into each promise it makes.

Clauses are from 21 CFR 202.1(e).1 That section is written about advertisements. FDA also treats detailing pieces, brochures, mailing pieces and similar matter as labeling, and its letters find misbranding in both.8,3

Why reading a compound sentence as one claim hides two

"Self-injected at home with a prefilled pen after training, and can be kept at room temperature for up to 30 days" looks like one line. It makes three promises: who gives the dose, what device they use, and how it may be stored. Each sits on a different page of the label. A reviewer who ticks the sentence because the first half checks out has approved the other two by accident. Splitting claims is slow by hand, and it is the first thing a careful substantiation pass does.

21 CFR 202.1, Form FDA 2253 and FDA's consistent-with-labeling test: the rules behind every claim

In shortThe core US rule, 21 CFR 202.1(e), is old and still in force. It requires truth, evidence and fair balance, meaning risks presented comparably to benefits. FDA's guidances fill in how to present information that goes beyond the label. What changed is enforcement: FDA's list of untitled letters went from 5 dated in 2024 to 63 in 2025 and 24 so far in 2026.4 In the EU every part of an ad must comply with the product's official summary, and the UK's industry code adds a ten-day clock for producing the evidence.

How the US, EU and UK differ on claims that go beyond the label

Rule text differs by market, and a claim approved in one country may fail in another. A library needs a market on each entry.

What can stand behind a claim: the label, the study, data on file, and how a claims library keeps track

In shortEvidence comes in a rough order of strength for promotion. The label is the safest, then the registration studies, then other published work, then the company's own unpublished data on file. A claims library records which wording rests on which page, and what must travel with it. Its value is highest when a label changes and you need to know every piece that used an old line.

One approved claim with its evidence, qualifier, channels and status

An open-label extension keeps patients from the main trial on treatment, with everyone knowing what they take. As observed means counting only patients still being measured.

FDA asks for the same precision when it gets annotated copies: "specific page and lines" of the label, or "specific page and column/paragraph from other references".21 DIA's best-practice article says references "must be annotated and accessible", and that a core claims document has to be kept up to date and shared with everyone who writes or reviews.27

That last step is where libraries pay for themselves, and where spreadsheets fail. In our fictional case, label version 4 added liver tests before treatment on September 18, 2026. One claim retired that day, and three live or in-flight pieces still used it, one of them a live web page. Without a record of where a claim is used, that search is a phone call to every brand manager.

A correct number can still mislead: five checks a medical reviewer runs on one fictional claim

In shortMatching digits is step one. A reviewer also asks which patients the number describes, what happened to those who left, how missing data were handled, whether the difference is stated in absolute terms, and whether the endpoint was planned. The five checks below use the fictional psoriasis banner.

Check 1: Does the rounding keep the meaning?

The label's Table 5 gives PASI 90 at Week 16 as 84.1% on Veltrimab (810 patients) and 4.9% on placebo (202 patients). PASI 100, completely clear on the score, is 56.2%. Three ways a draft might word those figures:

Check 2: Who is in the denominator?

The two-year claim says 87% of Week 16 PASI 90 responders maintained PASI 90 at Week 104. It reads as a strong result, and it is about a group already selected for doing well. Work it per 100 patients who started treatment:

FDA made a related argument in its April 23, 2026 letter on a consumer web page. Survival figures at three and a half years included a six-month open-label extension, and the agency wrote that events in that period might reflect "differences in patient populations that remained in the study and continued into the open-label extension".26 A qualifier saying "open-label extension" helps. It may not be enough on its own, and that is a judgment for the committee.

Check 3: How were missing patients counted?

The same fictional extension reports PASI 100 at Week 104 two ways. As observed, counting only patients still being measured, the figure is 52.3%. Under non-responder imputation, which counts anyone missing as not responding, it is 44.8%. A headline number picks one, and the footnote has to say which.

Check 4: Is the benefit stated in absolute terms?

84.1% against 4.9% is a difference of 79.2 percentage points, or about 17 times the placebo rate. Both are arithmetically right; only the first tells a reader how many patients gain.

FDA's January 30, 2026 letter on a TV ad objected to a claim that the risk of worsening was "61% lower". The figure came from a hazard ratio, a rate over time, and the letter sets it beside a risk difference of 25.7 points.8 For consumer pieces, FDA's quantitative guidance covers how such figures should be shown.12

Check 5: Was the endpoint planned?

The fictional email says "itch relief as early as Week 1". The study did look at itch at Week 1, in an exploratory analysis (one that looks for leads the study was not built to confirm) with a nominal p-value (a p-value not adjusted for the many endpoints tested). The rule lists "statistical analyses and techniques on a retrospective basis" that are not soundly supported.1

The CFL guidance allows that information about onset before a study's main time point can be consistent with labeling. Its example on the parts of a composite endpoint (one result built from several outcomes) warns that analyses not planned in advance or adjusted "could represent chance findings", and it says disclosing a study's limits does not rescue a study unable to support the message.10 Many companies go further with a house rule: no claim from an exploratory endpoint at all.

What FDA's 2026 untitled letters say about overstated benefits, broad indications and disclaimers

In shortUntitled letters are FDA's public notices that a promotional piece is false or misleading. They surged after September 2025. We read and coded the 24 dated in 2026 so far. By our count, half object to an overstated or unsupported benefit, two-thirds raise risk information, and three-quarters say a disclaimer did not fix the problem. At least four concern pieces for professionals.

Five things FDA quoted back in 2026, and what was wrong with each

Each letter also sets out other concerns. The source list names each product; we leave the companies out of this table since the point is the pattern.

Convenience wording has drawn objections for years. A 2020 letter on an emergency syringe objected to "Easy to use. Easy to know you did it right", because the product took several steps.32 A February 2026 law firm note on that year's wave says comparative positioning must rest on appropriate data, and that theorized convenience or patient preference will not carry it.33

Retired claims, dropped qualifiers and old safety text: how errors get in, and what happens next

In shortMost substantiation errors are ordinary: wording drifts, an old template is reused, a qualifier does not fit a small format, a reference is the wrong version. Inside the company they cost a committee slot and an agency round. Outside it, a letter brings a demand to stop, a written response within 15 working days, and sometimes a corrective campaign.

The causes are our reading from practice guidance and FDA's letters.27,4 No published study counts these errors by cause or by who catches them. Audit a sample of your own to find your mix.

From a lost committee slot to a corrective campaign: the consequences of a weak claim

The largest numbers on record come from government settlements. A 2019 JAMA study counted 103 settlements between drug companies and federal or state governments from 1997, over $11 billion in all, for off-label or deceptive marketing.35 Those cases involve conduct far beyond a single misworded banner, and the figure is dated, so read it as a measure of how seriously the law takes promotion. There is no public figure for what a single untitled letter costs a company in withdrawn materials, corrective work and staff time.

How many pieces reach FDA, how long MLR review takes, and how often a piece comes back

In shortVolume is high and rising: FDA received over 150,000 pieces in 2024. Published time benchmarks range from about 15 to 24 days, and they measure different things. First-pass approval, a piece cleared in its first round of review (its first circulation), runs from 56% to 79% by reviewer role. All these benchmarks come from software vendors; treat them as ranges and measure yourself.

Why a small first-pass failure rate becomes a campaign problem

Veeva's 2021 report found 74% of digital content approved in one review cycle.40 Take that at face value and assume pieces fail independently. A campaign of ten pieces then has about a 95% chance that at least one comes back: 1 − 0.74¹⁰. A June 2026 article from Klick, an agency, published the same arithmetic.42

Pieces also multiply. Veeva's report said content volume "almost tripled" over the three years it studied, with digital content running 3.5 times print.40 Each new format repeats the same core claims. FDA's own count, about two pieces per submission, points the same way: much of the work is many forms of one message.2

What AI can check in a promotional piece, what it cannot, and what regulators have said so far

In shortReading every frame, matching wording, finding the page and spotting a retired claim are good work for software. Judging a piece's overall impression is a person's job. Large language models are also known to cite sources that do not fully support what they say, so each citation has to be checkable. No regulator has written specific rules for AI in promotional review, and FDA says it uses AI to watch drug ads itself.

Where FDA, EMA and the EU AI Act stand on AI in promotional review

How promotional review teams substantiate claims today: who does what, and where the days go

In shortAn agency writes the piece and marks up references, a coordinator checks it against the claims library and books a committee, and three reviewers read and comment. The waiting and the rounds take most of the calendar. No study publishes hours per piece, so the split of hours below is a planning estimate to replace with your own.

Agency producer to regulatory reviewer: who touches a promotional claim

Seven steps between a brand brief and Form FDA 2253

Where the calendar days go in MLR review, by Veeva's 2021 count

Where the working hours go on one banner, by our estimate

In-house coordinators, an agency's medical team, or promotional review software

A three-frame psoriasis banner: seven claims found, one caught, one line changed, Thursday's committee kept

In shortThis is the example from the top of the page, inside the product. At 8:41 am on Wednesday, October 7, 2026, Lena Ortiz, a promotional review coordinator, opens a new banner for Veltrimab. Six agents (pieces of AI software, each with one job) read it and pin each claim to its approved wording and page. One headline has no approved claim behind it. She swaps in approved wording and books the committee, all before the 5:00 pm cut-off.

1

The draft fills the screen, as the reviewer will see it

Lena's home screen opens on the newest piece itself: three frames, 300×600 pixels, and the safety scroll under them. Along the top run the other drafts before MLR. Harborline Health, the agency, sent this one, and it is due to reach 1.1 million impressions on medical sites.

2

Pressing "Pin the claims" runs six agents in view

She presses the button, and the steps run across the top. The piece reader has taken the three frames apart. The claim extractor is underlining claims, headlines included. Next come the library matcher, the reference anchorer, the new-claim catch and the label and safety check.

3

Six pins go green, and "Simple to stay on" goes red

The extractor found seven claims. Six match approved claims word for word, pinned to five reference pages. Pin 3, the frame-2 headline, matches nothing among the 142 approved claims. It tells a dermatologist that patients keep taking the medicine, and no reference in the library measures that.

Readiness, a score out of 100 for how close the piece is to going to committee, reads 88. Sending takes 90, and an open new claim blocks sending whatever the score. The red card offers the nearest approved wording, CL-0130.

4

Pin 1 opens the row behind the 84%

Lena opens pin 1. "84% of patients achieved PASI 90 at Week 16 (vs 4.9% placebo)" is approved claim CL-0102, word for word, and it points to Prescribing Information version 4, page 18, the PASI 90 row. The card quotes the row: 84.1% and 4.9%. One more click opens the page.

5

Asking medical to substantiate would miss Thursday

Lena has two routes. She can ask Dr. Maya Chen, the medical reviewer, to substantiate the headline as proposed claim CL-0152, due October 9. The card then says what that means: Dr. Chen looks for a reference that measures how long patients stay on treatment, and the banner goes to the next committee.

The product shows this route without judging it. A persistence claim might be worth a week's wait, on some brands.

6

The approved dosing claim goes in, and readiness reaches 100

She chooses approved wording: "Two starter doses, then one maintenance dose on a set schedule", CL-0130, section 2.1, page 2. The band turns blue and the card shows the passage. Readiness goes to 100, with undo a click away.

7

Send to MLR books slot 8 of 10, Thursday at 10:00 am

Send to MLR shows the committee, the readiness, the reviewers and what the pack will hold: 7 claims, 2 references, 5 pages. The cut-off is still hours away. Dr. Maya Chen reads for medical, Omar Haddad for regulatory, Dana Okafor for legal.

8

A reference pack numbered to match the piece

The pack builder numbers each claim, marks the five passages with the same numbers, and assembles a 24-page pack. The medical reviewer no longer has to find the sentence. The PASI 90 row is highlighted with pins 1 and 2 in the margin.

An overstated subject line, a cross-trial comparison, a retired claim and a wrong age in the same week's drafts

In shortThe banner was the clean case. Other drafts carry the rest of what MLR sends back: an email with seven things to decide, a detail aid using a claim the label has just retired, and a social post that widens the approved age. Each flag comes with the reason, the page and, where one exists, the approved wording.

1

An email on two-year data, bound for 18,400 dermatologists, scores 58 on readiness

The VELA-2 email has 12 claims. Its subject line, "Clear skin that lasts", reads as complete, permanent clearance. The study shows 56% at PASI 100 at Week 16, and the two-year data come from an open-label extension. The 87% line has lost its qualifier.

2

The rest of the same email: an exploratory result and a comparison

"Itch relief as early as Week 1" rests on an exploratory analysis with a nominal p-value, and the nearest approved claim is itch at Week 16. "Higher PASI 90 rates than other biologic treatments" has no head-to-head study. VELA-2 compared Veltrimab with placebo only, and its own discussion warns against comparing across trials. Further down, a line about tuberculosis conflicts with the label, and the email still carries an older safety text and label link.

These rules are this sample team's settings. A house rule against any exploratory claim is stricter than FDA's guidance, which leaves room for some of this information, with sound evidence and context.10

3

A claim the label retired 19 days ago, still in a detail aid

Prescribing Information version 4, dated September 18, 2026, added liver tests before treatment in section 5.4. So CL-0127, "No routine laboratory monitoring required", is retired, and detail aid PC-2141 still says it. A thread runs from the claim to the passage. The approved replacement, CL-0128, is offered, and the screen says plainly that a conflict with the label cannot be kept as written.

4

A social post that widens the label's age by six years

The Dravenzo post says "For anyone 6 and older living with eczema". Dravenzo is approved from 12 in this sample. The headline "Itch-free nights start here" also overstates a result. Pin by pin, the checker points to the indication on page 1 and the itch row on page 14, and Lena's note to North Pier Creative, the agency, can name each fix and its approved wording.

Eight more screens, from the whole queue to the claims library, the dashboard and the rules a team sets

In shortOne banner is the story. These screens show the rest: every piece in one list, the claims library with a label update on top, why pieces come back, the rules your team sets once, and the record of what each agent did.

From the piece reader to the pack builder: six agents, a coordinator who decides, and a committee that signs

In shortEach agent has one job and limits it does not cross. None can edit the agency's file, send a piece or file with FDA. The coordinator decides every flag, a medical reviewer approves each new claim, the committee reviews and signs, and regulatory files Form FDA 2253 after approval.

Piece reader to pack builder: what each agent does and never does

The coordinator decides each flag, and three reviewers sign

Each agent step, flag decision with its name and time, rule change and piece sent is on the audit trail, as controls that support your compliance with your own quality system and with 21 CFR Part 11 where it applies. Botminds holds a SOC 2 Type II audit and manages information security to ISO 27001.

A draft from any agency in; every claim pinned, new claims caught and a reference pack out

In shortThe solution reads the drafts and references your team already has, applies your library and your rules, and leaves a marked-up piece and a pack for your committee, with a record behind every decision. It sits alongside the systems you keep.

Content systems, reference libraries and committee calendars around a claims check

Built for a portfolio: many brands, agencies and channels under one library and one set of rules, with the same checks applied every time.

What a hand-checked claim costs you in review rounds, lost committee slots and FDA letters

In shortStart from your own costs. Nobody publishes a cost per piece, reviewer hours per piece or a claim error rate, so the numbers that matter are yours. Published figures frame them: how long a job takes, how often a piece comes back, and how FDA's enforcement has moved.

Six costs of manual claims work, and how to find each number in a week

The reviewer's morale, the letter nobody budgeted, and other costs of manual claims checks

What faster claims checks return when you count annotation hours, avoided rounds and committee slots kept

In shortPut in your own numbers. The calculator values hours saved on finding and marking up claims, and second rounds avoided, then sets them against what a tool and your testing cost a year. It leaves out what matters most, fewer public objections and pieces out on time, because those have no honest dollar figure.

Annotation time, review rounds and retired-claim searches: where claims work saves

Work out the return on MLR automation from your pieces, hours and return rate

A team reviewing 60 pieces a year for one brand may gain more from a well-kept claims library in a shared document, an agency briefed on it, and a one-page house rule list, than from new software. Volume, agencies and label changes are what change the answer.

How ready your promotional review team is: claims library, references, agencies, house rules and sign-off

In shortPick the answer closest to your team today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

A low score means building your claims library and house rules first

What to ask an MLR review tool about implied claims, reference pages and retired wording

In shortAsk every vendor the same questions, us included. The answers show how a result is built: whether implied claims get found, whether each pin opens a real passage, how version changes work, who decides, and how you would test it on your own pieces.

We found no published independent comparison of promotional review tools, as of November 2026. A planted-error test on your own pieces is the fairest measure, and it measures your current process too.

Test a claims tool on your own past pieces, with planted headlines and a retired claim

In shortMeasure first, then test on your own past pieces with errors you planted, then decide with four numbers. Keep your reviewers signing throughout. None of these steps commits you to a vendor, and the first two improve your process with any tool.

1. Audit 20 approved pieces against their references, claim by claim, and log each problem. That is your error baseline.
2. Clean up one brand's claims library: approved wording, page, qualifier, channels, status. Time your next ten pieces by step.
3. Gather 40 past pieces across formats. Copy some and plant known errors: a new headline, a dropped qualifier, a retired claim, an old safety block, a cross-trial comparison.
4. Run in parallel on live pieces for eight weeks, with your coordinators deciding and reviewers signing everything as they do today.
5. Decide with four numbers: catch rate by error type, false alarms on clean copies, hours per piece, and first-pass rate at committee.

Scoring a claims pilot: catch rate, false alarms, hours and first-pass rate

Passage accuracy is the check behind the catch rate; the other four are the numbers you decide on.

Claims substantiation questions answered: MLR review, fair balance, Form FDA 2253 and AI

In shortQuestions reviewers, agencies and buyers search for, each answered on its own.