Stop Writing Vague Copy: A Good Marketing Claim Is Precise and Specific

Buried in a twenty-year-old pharmacoepidemiology paper is one sentence that belongs taped to the monitor of every copywriter, brand lead and reviewer in pharma:

A good marketing claim is precise and specific; a particularly good example is a direct quote from the original article.

That is Kari Lankinen and colleagues, writing in Pharmacoepidemiology and Drug Safety in 2004. They did something almost nobody had bothered to do before: they pulled every reference behind a large sample of prescription drug advertisements and checked, one by one, whether the study actually said what the ad claimed it said.

Spoiler: mostly, it didn't.

And here is the part that should change how you write. The problem was not bad research. It was bad claim writing.

What they actually did

The team screened 1,036 drug advertisements from four major Finnish medical journals published in 2002. That narrowed to 245 distinct advertisements carrying 883 marketing claims between them, for 148 drugs from 34 companies.

Every claim was sorted into one of four buckets — unambiguous clinical outcome, vague clinical outcome, emotive or immeasurable, and non-clinical. Then every Medline reference was traced, read by at least two independent reviewers, and graded on a scale running from unreferenced through irrelevant up to strong research-based evidence. Where the two readers disagreed, they went back to the paper and argued it out until they agreed.

In other words, they ran a systematic MLR review across an entire country's medical advertising for a year. Here is what came back.

Finding one: almost nothing was specific

Only 9% of the 883 claims stated an unambiguous clinical outcome. Sixty-eight percent were vague or emotive.

The authors are blunt about why that is fatal, and this is the bit worth internalising: if a claim is too vague or too general, it is impossible to support credibly with any reference. Their own examples:

• "Effective." For what? Compared with what?
• "Rapid." How rapid? Compared with what?
• "Significantly better tolerated." This one is the trap. It borrows the vocabulary of statistics, so it reads as precise — but it tells you nothing. Better tolerated than what, on which measure, by how much, in whom?

The reason is mechanical, not stylistic. Trials produce specific outcomes in specific populations. Nothing in the literature is shaped like "effective," so a vague claim can never land cleanly on the evidence. Write one, and you have guaranteed that whoever hunts for its support will have to stretch something to make it fit.

The data says exactly that happened. Of the claims classified as vague, 50% were referenced to sources that were either untraceable or irrelevant, and another 16% to reviews or other non-scientific publications. Vague and emotive claims were significantly more likely than precise ones to be propped up by a reference that could not carry them.

Finding two: citing a paper does not prove anything

Only 38% of the claims carried a reference at all. When the researchers chased down the 381 references that were there:

Evidence level of the reference	Share
Not retraceable (non-Medline, incomplete or incorrect)	32%
Irrelevant — retrieved, but does not support the claim	21%
Non-scientific support	17%
Limited research-based evidence	19%
Moderate research-based evidence	9%
Strong research-based evidence	2%

Read the top two rows together, because that is the finding: for more than half of the referenced claims, the citation either could not be found or did not support the words on the page. At the other end, 2% reached the strongest tier — striking, given that nearly every heavily advertised product was an established drug with trials and meta-analyses behind it.

Then there is the twist. Among the precise, unambiguous claims — the good ones, the ones the authors held up as the standard — not one was supported by strong research-based evidence, and 8% of their references did not support them either.

So being specific does not make you right. It makes you checkable. That is a different property, and a more useful one, because checkable claims are the ones that survive review. It is also the whole argument for anchoring rather than merely citing: a reference proves someone looked something up, nothing more.

The big three mistakes, still alive in your review queue

The paper's worked examples are not exotic. If you have sat through MLR, you have seen all three this year.

• The flat-out falsehood. An advertisement claimed more endurance. The cited study was run in 10 healthy male volunteers, 7 of whom had moderate to severe adverse reactions — most often recurrent fatigue. The source did not just fail to support the claim; it pointed the other way.
• Population jumping. A claim about improving cardiovascular prognosis in hypertensive patients cited two papers reporting the same Swedish randomised trial in acute myocardial infarction. The trial made no mention of hypertension anywhere. Strong evidence, wrong patients.
• The unfair mashup. A claim that one drug at 4 weeks cured as many reflux patients as a comparator at 8 weeks lifted two arms out of studies that had measured both timepoints for both drugs — a comparison the studies never made, and one that ran against their own numbers.

None of these slipped through a gap in the rules. They published in peer-reviewed journals in a market with binding legislation, a national industry code, and the IFPMA and WHO promotional criteria all in force. Rules on paper caught none of them. Only someone reading the source would have.

The numbers nobody printed

One more finding, and it has aged into a live issue. Across 883 claims, statistical parameters appeared 34 times. Absolute risk reduction showed up in exactly one advertisement. Confidence intervals and number needed to treat appeared in none.

That is the difference between a claim a physician can weigh and one they can only absorb. Relative framing without the absolute numbers is still one of the most dependable ways to make a true statement misleading — and still one of the specific things a medical reviewer is there to catch.

Does a 2004 study still matter?

Fair question, and worth being straight about the limits: this is a 2002 sample of print journal advertising in one country, from before most of the review infrastructure we now work inside. A company running a modern MLR process with a claims library behind it should not be producing a 21% irrelevant-reference rate.

But the failure taxonomy has not moved an inch, and the paper's own comparison table shows the behaviour did not either: 80% of ads unreferenced across 18 countries in 1993; 92% of US ads judged non-compliant with FDA criteria in 1992, only 4% acceptable without revision; 44% of references failing to support their statement in a 2003 Lancet analysis of Spanish journals. Different decades, different regulators, same result.

And one echo is uncomfortably exact. A reference that looks credible, is formatted perfectly, and does not support the claim attached to it is precisely the failure mode of AI-generated citations. In 2004 that took a human rushing a deadline. Today it takes seconds, and the output reads better.

How to write a claim that survives review

If a good claim is precise, and the best claim is a direct quote from the source, then substantiation is not something you achieve at review. You decide it when you choose the words.

Before a claim leaves your draft, check that you can name all six:

1. The outcome. Which measured endpoint, in the trial's own words.
2. The population. Which patients, and whether they match the ones you are promoting to.
3. The comparator. Versus what — placebo, active control, baseline, nothing.
4. The magnitude. The actual number, absolute as well as relative.
5. The timeframe. Over which period, measured at which visit.
6. The source sentence. The one line in the one paper that says it.

Miss any of the six and you have not written a claim. You have written vague copy.

Get the sixth and you are most of the way to anchoring it — highlighting the exact supporting sentence so your reviewer verifies in seconds instead of rebuilding your reasoning from scratch. That claim-and-evidence pair is also the raw material of a core-claims library, which is how a good claim gets reused instead of re-argued every quarter.

Lankinen's team closed on a point that lands harder now than it did in 2004. Companies employ medical departments precisely to support marketing, they noted, and their findings suggested that expertise simply was not in the room when the advertisements were planned. Most of the unsubstantiated claims they found would have been avoided by asking a medical expert first.

That is garbage in, compliance out, made two decades early. Reviewers are not there to repair a claim that was written to be unsupportable. They can only send it back — late, expensively, and more than once.

Substantiable claims FAQ

What makes a marketing claim substantiable? Precision. A claim that names a specific outcome, in a specific population, against a specific comparator, with a magnitude and a timeframe, can be matched to a specific trial result. Vague claims like "effective" or "rapid" cannot be, because no trial produces a result in that shape — which is why Lankinen et al found half of all vague claims propped up by irrelevant or untraceable references.

Is a claim with a reference automatically substantiated? No. In the 2004 audit, 21% of retrieved references did not support the claim they were attached to, and another 32% could not be traced at all. A reference proves someone looked something up. Only the content of the source, checked against your exact wording, proves the claim.

Why is "significantly better tolerated" a problem if it sounds statistical? Because it borrows the vocabulary of measurement without supplying any. No comparator, no tolerability measure, no magnitude, no population. It reads as precise and reviews as vague, which is the worst combination — it invites approval while remaining impossible to anchor.

Does quoting the source article directly solve substantiation? Most of it. A direct quote cannot drift from the evidence, because it is the evidence, and it makes anchoring trivial. What it does not solve is relevance: the quoted result still has to concern your population, your comparator, and your approved label.

Precision is cheaper than rejection

Every failed claim in that study was, at some point, a sentence somebody typed. The audit is really a record of what happens when writing the claim and checking the evidence are two different jobs, done at two different times, by two different people.

PharmaText.ai closes that gap. Our Precision Traceability engine ties each claim to exact coordinates in your source PDFs as it is drafted, so those six questions get answered while the sentence is still yours to change — not three review rounds later. If your team is stuck in the vague-claim cycle, get in touch by email.

Related: see what actually needs a reference for the substantiation rules, linking and anchoring for the anchoring standard, and the MLR review process for what imprecise claims cost you in rounds.

Source: Lankinen KS, Levola T, Marttinen K, Puumalainen I, Helin-Salmivaara A. "Industry guidelines, laws and regulations ignored: quality of drug advertising in medical journals." Pharmacoepidemiology and Drug Safety 2004;13(11):789-795. doi:10.1002/pds.1017. Comparative figures cited within it from Herxheimer et al (1993), Wilkes et al (1992), and Villanueva et al, Lancet 2003;361:27-32.