Science-backed sensitive toothpaste brands in the UK: how to check the evidence yourself.

"Science-backed" is not a defined term in UK advertising rules or in cosmetics law, so on a toothpaste it means whatever the brand printing it wants it to mean. What you can do instead is run five checks on the evidence yourself, each of which takes a couple of minutes: is there a paper you can open, where was it published, who paid for it, what design and how many people, and did the study test this product or the ingredient inside it. S3 Sensitivity Science™ is the only sensitivity toothpaste in a category review of 51 UK products read in September 2026 whose ingredient list carries potassium nitrate, hydroxyapatite and fluoride together, and the scan behind that sentence is published in full so you can run the check on it. Its own two studies are a consumer trial of 51 adults over eight weeks and a laboratory report from testing at the University of Reading, and they fail two of the five checks below; this page runs the checks on them in the same words it uses for everyone else.

What was checked19 peer-reviewed studies, S3 consumer trial (ADSL, 2026), product information as published by each brand

What does "science-backed" mean on a toothpaste, and who checks it?

On its own it means nothing, because no rule defines it. What the rules do say is narrower and more useful: rule 3.7 of the UK Code of Non-broadcast Advertising says that before an advertisement goes out, the marketer must hold documentary evidence to prove claims consumers are likely to regard as objective, and that the ASA may regard a claim as misleading where adequate substantiation is missing¹. A phrase such as "backed by science" sits just outside that, in the space where a sentence sounds objective without committing to anything a regulator can weigh.

Two bodies in the UK look at a toothpaste's evidence, and neither of them looks at most of it. The ASA looks after the fact and only when somebody complains. Its rulings register is public and searchable, and a keyword search of it on 2026-09-10 returned five rulings for "toothpaste" and zero for both "sensitive teeth" and "dentist recommended"². The register is searched by keyword, so a ruling filed under other wording would be missed; even allowing for that, the shelf is close to untested. The most recent toothpaste ruling, in February 2026, was upheld against three social advertisements that showed a five-star review rating the advertiser could not evidence⁴, which is a claim about evidence failing rather than a claim about a formula.

The second body is the Oral Health Foundation, whose accreditation scheme works the other way round: an independent panel of dental experts reads a product's claims before they go on pack and checks each one against the evidence submitted, and the charity is explicit that accreditation "is not a judgment about whether a product is 'better' than another"⁵. The limit is in its own fee page: manufacturers apply, pay a one-off panel fee for the review and an annual licence fee to keep using the logo⁵. So the mark is real evidence that somebody checked, and its absence is evidence of nothing at all, because entry is voluntary and paid for. S3's own product page does not carry it either.

That leaves the checking to you. Here is the whole method on one line each.

Check one: is there a paper, and can you open it?

Start here, because most claims stop here. Paste the wording from the pack into PubMed, or take a DOI to `doi.org` and see where it lands. If the brand's "clinical study" page links to its own document, a press release or nothing, the claim is not checkable, which is different from being false and should be described that way.

Two products in this scan show both outcomes. Nura states that its paste was tested at the University of Milan, names a professor of dentistry there and reports an in-vitro result against a leading fluoride toothpaste; there is no paper to open. The maker of Apagard, at the other end, paid for a double-blind randomised trial of 85 adults over eight weeks that was published in BDJ Open and is free to read, with the sponsorship by the manufacturer of the tested pastes declared in the paper itself⁶. Same commercial motive, opposite result on check one.

The regulator applies the same test. In April 2026 the ASA upheld a complaint about a skincare advertisement whose claim of clinical proof rested on a trial the advertiser asked to keep confidential; the ASA could not see how participants had been assigned to groups, judged the sample small, and noted that a percentage of participants agreeing their skin looked better was self-reported and could not substantiate a claim of clinical proof⁷. Nothing about that reasoning is specific to skincare.

Check two: where was it published, and is the journal indexed?

Indexing means a body other than the publisher decided the journal met editorial standards, and it is the fastest proxy a non-specialist has. MEDLINE, reached through PubMed, is the strictest of the common ones; PubMed Central holds the full text of work deposited there; DOAJ lists open-access journals that have passed its own review. Searching the journal's ISSN on `doaj.org` takes about fifteen seconds and answers the question.

It answers it for this page, too. The reporting-quality review cited in the next paragraph appears in The Saudi Dental Journal, whose PubMed record is marked as not indexed for MEDLINE; the journal is listed in DOAJ and its full text sits in PubMed Central, so it clears the bar this project sets and the page says which bar it cleared rather than implying the strictest one.

What indexing does not tell you is whether the paper is any good, and the field has measured that. A 2025 systematic review scored 16 randomised trials of desensitising mouthwashes against the 32-point CONSORT reporting checklist and found a mean of 23.3⁸. Every one of the sixteen described its intervention, its outcomes and its effect size adequately; six described the randomisation sequence and the blinding in full, and two described the protocol registration and the setting⁸. You will almost always find the result, and often not find how it was obtained. Two limits belong with that number: the review covered mouthwashes rather than toothpastes, and it judged 14 of the 16 trials at low risk of bias, so poor reporting is not the same finding as an unreliable trial⁸.

Check three: who paid for it?

Look under the abstract. PubMed prints a funding statement and a conflict-of-interest statement where the journal supplied one, and the authors' affiliations sit at the top. Disclosure is the norm in this field, not a disgrace, and a disclosed trial is more checkable than an undisclosed one.

The scale of it is worth stating plainly. The 2026 systematic review and network meta-analysis of 93 randomised trials in 9,548 adults, the largest pooled reading of desensitising toothpastes there is, reports that most of the trials it included were paid for by industry: 96% of the stannous fluoride studies, 86% of the arginine studies, 76% of the potassium studies and 33% of the nano-hydroxyapatite studies³. Most of what is known about desensitising toothpaste was paid for by the people who sell it. That is a reason to read the design, not a reason to throw the result away.

The named examples run across every part of the shelf. The systematic review of 44 clinical trials that hydroxyapatite brands quote most often, which found hydroxyapatite toothpastes reduced sensitivity by 39.5% against placebo and gave 23% more relief than fluoride toothpastes, is a 2023 systematic review whose authors include scientists employed by Dr Wolff, a maker of hydroxyapatite toothpastes⁹. A double-blind randomised trial of 120 adults comparing three desensitising pastes was run and funded by Procter & Gamble, seven of whose eight authors are its employees, and tested a paste it sells¹⁰. A meta-analysis of 14 randomised trials in 1,287 participants on the same company's stannous fluoride paste was assembled from that company's own archive, by authors employed by it¹¹. A randomised trial of 215 adults on a bioactive-glass paste was sponsored by Haleon, the maker of the product tested, with five authors on its payroll¹². The 2023 network meta-analysis of 32 randomised trials in 4,638 participants comes from a university group that declares advisor fees, lecture fees and research grants from dentifrice makers, including a maker of pastes in the pool¹³. Even the epidemiology follows the money: the cross-sectional study of 3,551 adults in seven European countries that measured how common the condition is was financially supported by Haleon, with one author employed there¹⁴.

Guidance is not exempt. The Oral Health Foundation's consumer page on sensitive teeth, a good page and a reasonable place to start, thanks Oral-B for an educational grant at the foot of it¹⁵. It discloses, which is exactly the behaviour the checks are looking for.

One honest gap belongs here. A PubMed search made while writing this page, for work on sponsorship bias in dentifrice research, brought back nothing on the subject. So this page cannot tell you how far industry funding moves a number in dentistry; it can only tell you to find out who paid, and then read check four.

Check four: what design, how many people, and for how long?

There is a published answer to what a trial of this condition has to do, and it is nearly thirty years old. The 1997 consensus guidelines for the design and conduct of dentine hypersensitivity trials set out what one has to carry: parallel groups with an allocation neither the participant nor the examiner can see, tactile, cold and evaporative stimuli, a negative and a benchmark control, eight weeks for most studies, and follow-up to see whether the change persists; they also say at least two independent studies should be run before a product is approved¹⁶. Measure a claim against that list and most of the shelf's evidence is not measured against anything.

The field has just restated the point. A 2026 scoping review and Delphi consensus went through 72 papers on how the condition is defined and appraised 58 randomised trials from the previous five years; definitions varied widely, few studies applied every element of the accepted one, and randomisation, blinding and effect estimates were reported inconsistently against the CONSORT standard¹⁷. The same process agreed a case definition, minimum diagnostic criteria, validated assessment tools and a core outcome set¹⁷, which is a polite way of recording that until 2026 the trials were not fully comparable with each other.

The reason a control arm is not optional here is that sensitive teeth respond to being studied. One 1997 trial put the question directly: 120 adults used potassium nitrate, strontium acetate or a plain fluoride paste for six weeks, nobody involved knew which, all three groups came down, nothing separated them, and even the plain fluoride group improved significantly against its own starting point¹⁸. Ten years later a split-mouth randomised study of 22 adults went further and covered sensitive teeth with a dressing carrying no active at all: reported pain fell by 95% to a hot or cold stimulus and by 85% to an air blast, against no dressing, and the authors concluded that what these studies measure holds a stimulus component and a placebo component together¹⁹. Put a paste in a single arm with nothing beside it and there is no way to tell those apart¹⁸¹⁹.

Check five: did the study test this product, or the ingredient?

This is the commonest substitution on the shelf, and it is not always wrong. A brand that points at the published work on its active is doing something legitimate, provided the level in the tube matches the level the researchers used and the page says which. It becomes misleading when a class result is presented as a result for the tube.

Three examples of the substitution, with different degrees of care. RiseWell states that decades of studies show its 10% micro-hydroxyapatite inclusion level to be an effective concentration, without naming one of them. Karex states that 291 studies were identified investigating its hydroxyapatite and links a page of them, mixing laboratory, in-situ and clinical work under one heading. The 2019 systematic review and meta-analysis of six four-week randomised trials rated nano-hydroxyapatite's pooled effect high-quality evidence by GRADE on evaporative and tactile stimuli, found no difference on cold stimuli, and did not separate the concentrations tested, so it supports the ingredient and not any particular percentage²⁰.

S3 makes this substitution too, and the wording is the point: the register form is that the ingredients in S3 are supported by more than 90 verified published studies, which is a claim about the ingredients rather than about the tube. The same discipline applies to the percentages: 10% nano-hydroxyapatite and 5% biomimetic hydroxyapatite are inclusion levels of the ingredient as supplied, and the active hydroxyapatite content is lower, which S3 states in both forms.

What do the five checks return for the brands on the UK shelf?

Mostly a claim of clinical proof and nothing behind it that a shopper can open. The table records what each brand's own page said on the date it was read, not how well any product works. This page compares ingredients and stated actions only, not clinical performance, based on what each brand states about its own formula. Prices and formulations may change; always check the pack. Every row resolves to a row of the published category review by its identifier.

Two things stand out. Searching the ASA register for the phrase every mainstream pack here uses returns 22 rulings, and none of the 20 listed on its opening page of results concerns a toothpaste: they are cosmetics, supplements, and weight-loss and hair-loss treatments². And two of the brands that describe their evidence in most detail, SURI and S3, are both describing their own consumer trials, single-arm, self-reported and unpublished in both cases.

What happens when you run these five checks on S3's own evidence?

It fails checks one and three, which are the two that decide whether anything else about a study can be argued over at all.

Check one, is there a paper: fail, twice. S3's consumer trial is an independent third-party consumer trial of 51 adults with sensitive teeth over eight weeks, conducted by a contract laboratory in Devon to Good Clinical Research Practice. It has not been published, so you cannot read the questionnaire, the dropouts or the questions that scored worst. The University of Reading work is an internal laboratory report, not a peer-reviewed paper; the only wording the register permits for it is that the formula was tested at the University of Reading, never endorsed by it.

Check two, where was it published: not applicable. With no journal, there is no venue to index and no editor who read it.

Check three, who paid: fail, twice. S3 commissioned and paid for both. That is the same position as the manufacturer-funded Procter & Gamble and Haleon trials cited above¹⁰¹², with one difference that runs against S3: those were published, so their design can be argued with.

Check four, design and numbers: partly. The consumer trial ran eight weeks, which matches the duration the 1997 consensus guidelines recommend¹⁶, and its 51 participants are a real cohort. It has no control arm, no randomisation and no blinding, and it measures agreement with statements rather than a tactile threshold or a cold-air score, so what it records is what people said. The register wording keeps that verb: 88% said their teeth feel less sensitive with S3. Read against check four and the placebo evidence above, a single-arm self-reported study cannot separate the paste from the response to being in a study¹⁸¹⁹. The laboratory work has the tighter limits of the two: ovine dentine rather than human, one tooth per formulation for the quantification, and the figures are reported as laboratory measurements of tubule coverage at 24 hours rather than as anything a person felt.

Check five, this product or the ingredient: partly, and stated. The Reading testing was run on a prototype formulation rather than the marketed tube, and the published literature S3 cites is literature about potassium nitrate and hydroxyapatite, not about S3. This is also why S3 does not describe the product as proven in clinical trials: the one place the register permits that phrase is the 5% potassium nitrate dose, which was clinically proven in placebo-controlled trials of that dose.

What survives the audit is the part that needs no trust at all. The ingredient list is printed on the tube and recorded in the published category review with the date it was read, the levels are declared, and the literature on those ingredients sits in PubMed under the names of people with no connection to the company. Anyone auditing S3 has to take S3's word for exactly two things, and this section has just marked both of them down.

What does the pooled evidence say for each active, and how certain are the reviewers?

Less certainly than the packs, and the honest way to read it is to take the reviewers' own confidence grades and their trial counts rather than the size of any single effect. The 2026 network sets each class against a benchmark fluoride toothpaste at two weeks and prints a grade beside every estimate; its own recommendation is that stannous fluoride and arginine be first choices for home care, chosen on preference, tolerability and availability rather than any expectation that one wins³. That conclusion, drawn from a systematic review and network meta-analysis of 93 randomised trials, does not flatter the actives S3 uses, and it belongs on this page for that reason³.

The reviews also disagree with each other, and this page prints the disagreements instead of averaging them. Calcium sodium phosphosilicate is the clearest case. Drawing on 12,541 patients, the 2020 network meta-analysis put it first on every stimulus it examined²¹. Six years on, the newer and larger network grades the same class a small effect at low confidence³. Potassium has the same problem in reverse: when the Cochrane reviewers pooled six trials in 2006, the air-blast and tactile readings moved at six to eight weeks while the patients' own ratings did not, and they closed by saying the evidence for these pastes was not clear²³. Behind all of it sits the caution the 2015 meta-analysis attached to its own figures after pooling 31 randomised trials: the studies varied so much among themselves that no single pooled number out of it should be quoted alone²².

MD is the mean difference on the cold-air Schiff score against a benchmark fluoride toothpaste in the 2026 network; the tactile figures are on the Yeaple probe scale; the funding column is the review's own accounting of the studies it pooled³. The pairing in the last row is the one S3 uses, it rests on a single pooled result from 2020, and it was not a node in the newer and larger network at all²¹³.

What should you do with a "science-backed" claim, and when is it not a toothpaste question?

Take the claim off the pack, spend ten minutes on the five checks, and then decide by ingredient and declared level rather than by the sentence. The Oral Health Foundation's own advice is to use a toothpaste containing potassium citrate, potassium nitrate or stannous fluoride, to spit rather than rinse after brushing, to brush gently with a soft brush, and to keep using the paste, because the effect depends on continued use¹⁵. Give a new paste the time the trials gave it: the consensus guidelines assume eight weeks, and the pooled two-week estimates are early readings rather than finished ones¹⁶³.

Two pieces of honesty belong at the end of a page like this. The first is that the long view is thin: a 2025 systematic review and meta-analysis of 22 randomised trials followed for at least six months found significant long-term pain reductions for several desensitising agents but not for the calcium-phosphate class, and a 2020 evidence-based overview for practitioners records that the long-term outcome of treatments that work by occluding tubules remains uncertain²⁴²⁵. The second is that the same overview states there are no universally accepted guidelines for telling this condition apart from the others that mimic it, or for choosing a reliable treatment²⁵.

Which is why the last check is not about evidence at all. If the pain is severe, if it has gone on for more than a few weeks, if it sits in one tooth only, or if it arrived suddenly, that is a dental appointment rather than a shopping decision, because decay, a crack, an infection or gum disease can all present the same way¹⁵²⁶. A toothpaste is for the slow, general ache across several teeth. The Journal's page on how sensitive toothpastes work, and why yours might not be covers what to change first, and the Journal's ranked view of the shelf is a ranking, which this page deliberately is not, because a ranking is the one thing the evidence set out above cannot support.

Frequently asked questions

Is S3 science-backed?

Half of the phrase is checkable and half of it is not, and the split is worth knowing before you buy. The ingredients are supported by published trials and reviews that anyone can open in PubMed, including the pooled evidence for potassium and for nano-hydroxyapatite summarised in the table above³²⁰²³, and none of that literature was paid for by S3. The product itself has two studies, both funded by S3 and neither published: a consumer trial of 51 adults over eight weeks that records what people said rather than what an instrument measured, and University of Reading laboratory work on ovine dentine. On the five checks, the ingredient evidence passes and the product evidence fails checks one and three.

What does a brand have to hold before it says a toothpaste is proven?

Documentary evidence, held before the advertisement runs, sufficient to prove any claim a reader would take as objective; that is rule 3.7 of the CAP Code, and the ASA may treat a claim as misleading where the substantiation is not adequate¹. What the rule does not do is define the wording, publish the evidence or check it in advance, so the phrase on the pack is a statement that a file exists somewhere rather than a statement you can read. The April 2026 ruling on a skincare advertisement shows what happens when the file is examined: a small unpublished trial with an unclear randomisation method and a self-reported agreement figure was judged not to substantiate a claim of clinical proof⁷.

Why do so many toothpaste studies come from the company that sells the paste?

Because nobody else pays for them. Trials of this size cost more than a university department or a research council usually spends on an over-the-counter product, and the 2026 network meta-analysis of 93 randomised trials records the result: 96% of the stannous fluoride trials, 86% of the arginine trials and 76% of the potassium trials it pooled were industry funded³. That is a reason to read the funding line and the design, not a reason to discount the field, and the well-run manufacturer trials in this page's references are more checkable than any unpublished study, whoever paid for it⁶¹⁰¹².

What is the difference between a systematic review and a trial?

A trial is one study of one intervention in one group of people, and it can be right, wrong or unlucky. A systematic review searches for every trial that meets stated criteria, appraises them and, where they are similar enough, pools them into one estimate; a network meta-analysis goes further and compares treatments that were never tested against each other directly³. Reviews sit above single trials for this reason, but they inherit whatever the trials did badly, which is why the 2026 review prints a confidence grade beside each of its estimates and why the 2026 methods consensus mattered enough to write down³¹⁷.

How do I check a study a brand links to in two minutes?

Copy the title into PubMed. If it comes back, read four lines: the journal, the funding and conflict-of-interest statements under the abstract, the design and number of participants in the methods sentence, and the intervention, to see whether the tested product matches the one you are holding. If it does not come back, search the DOI on `doi.org`, and if there is no DOI and no PubMed record, you have your answer to check one, which is not that the study is wrong but that you cannot see it.