The race to the bottom in online GLP-1 prescribing, and the better way to grow

A race nobody announced

Online prescribing of GLP-1 medicines has become one of the most commercially contested spaces in European healthcare, and the contest that matters is not about price or brand. It is about the width of the front door. Loosen the eligibility interpretation. Thin the verification. Accept the applicant the provider next door turned away, and do it faster than they do.

The commercial logic is easy to follow. Acquisition in this category is expensive, and every applicant rejected at intake is marketing spend written off. If a rival declines a patient and you accept them, their acquisition cost becomes your revenue.

What strikes me, watching this from inside, is where the effort is going. Senior product, engineering and clinical policy time is the scarcest resource any of these companies has, and a great deal of it now sits at the front door. Rewriting questionnaires to reject fewer applicants. Re-reading the licensed indication for the generous interpretation. Redesigning funnels so that last week's decline converts this week.

Each of those decisions is rational on its own. Together they are corrosive. When every operator optimises for the patients everyone else refused, the marginal patient in the system is, by construction, the one the most cautious provider judged unsuitable, and the sector's effective eligibility threshold quietly becomes whatever its least selective member will accept. Nobody voted for that outcome. It emerged from a hundred reasonable-sounding funnel optimisations.

Eligibility criteria are not a marketing setting. They are a clinical screening instrument, and when an industry starts competing on how little screening it can get away with, the people who absorb the risk are the patients taking the medicine, the clinicians signing the prescriptions, and eventually the companies themselves.

What the gate is actually for

Any clinician will recognise an intake questionnaire for what it is. It is a screening test, and like every screening test it trades sensitivity against specificity. A well-designed intake identifies the patients for whom a GLP-1 is appropriate and safe while excluding those for whom it is not: people below the licensed weight thresholds, people with active eating disorders, people who are pregnant or planning pregnancy, and people whose comorbidities or medications make remote initiation the wrong route.

Widen the gate without changing anything else and the arithmetic is unforgiving. Specificity falls, and the false positives are not an abstraction. They are a person with a history of anorexia obtaining an appetite suppressant, or an early pregnancy exposed to a medicine whose European label contraindicates it.

On pregnancy, the exposure data are so far moderately reassuring. A 2026 systematic review in Diabetes, Obesity and Metabolism found preconceptional or early-pregnancy exposure not consistently associated with increased maternal, fetal or neonatal risk. Not consistently associated is a phrase that describes a data gap rather than a green light, and the label position stands. On eating disorders the evidence is journalistic rather than epidemiological, but the signal is loud. Broadcasters have repeatedly documented people with healthy body mass indexes, and people with eating disorder histories, obtaining injections through online services with minimal challenge.

Who actually carries the risk

It is worth being precise here, because the risk is not distributed evenly.

The patient holds the clinical risk. They take the medicine.

The prescriber holds the professional risk, and this deserves more attention than it gets. The doctor or pharmacist independent prescriber who signs each prescription does so against their personal registration, and no employment contract transfers that accountability to the company whose funnel delivered the patient. A prescriber asked to work at the pace and margin that a loosened gate implies is being asked to underwrite, with their licence, commercial decisions they did not make. In my view this is the quiet fault line in the model. Services that widen intake without strengthening the information reaching their prescribers are spending their clinicians' professional capital as if it were the company's to spend.

The company holds the residual risk: enforcement, reputational damage, and the strategic exposure that comes from a whole category being regulated around its worst practices. All three rise together when the gate widens alone, and only one of the three parties was in the room when the decision was made.

What I am not arguing

Widening eligibility is not inherently wrong. When a provider extends its criteria and, in the same release, adds the mitigations that make the extension safe, that is exactly how clinical services should evolve. Stronger verification. Deeper conditional questioning for the newly admitted group. Tighter follow-up. Defined escalation routes. Some of the best work in this industry looks precisely like that.

My objection is to widening as a standalone act, where the criteria move and nothing else does. The first pattern is clinical development. The second is risk transfer dressed as growth. From outside, the two can look identical on a website. From inside, everyone knows which one they are running.

It is also worth saying that the licensed indication is already generous. The European labels for semaglutide and tirzepatide permit treatment from a body mass index of 30, or 27 with a weight-related comorbidity. That is the lawful outer boundary, and it is far wider than what public systems will fund. NICE TA1026 recommends tirzepatide only from a body mass index of 35 with a comorbidity, and NHS England's phased rollout began at 40 with four qualifying conditions. Private providers operating at the licensed floor are already serving a population the public system cannot reach for years. The race is not really about the thresholds printed on the website. It is about how honestly those thresholds are enforced, and that is where the sector has been caught out.

The regulators have stopped waiting

Anyone who believes eligibility drift is a theoretical concern has not been reading the enforcement record of the last twenty months.

In February 2025 the General Pharmaceutical Council rewrote its guidance on pharmacy services provided at a distance and added weight-management medicines to the list of higher-risk categories. The change was surgical. Prescribers can no longer rely solely on an online questionnaire, and must independently verify weight, height or body mass index through timely two-way communication, access to clinical records, or contact with the patient's own general practitioner. That is a regulator saying, in terms, that it no longer trusts the industry's self-reported intake data. The Council has since told ITV News that it had taken enforcement action against around eighteen pharmacies whose online checks fell short, and its assessment of photograph-based verification was blunt. One photograph, by itself, is not sufficient.

The press has repeatedly demonstrated why. In June 2025, BBC journalists obtained weight-loss injections by entering false information and uploading photographs, with no consultation. In January 2026, ITV News found people with healthy body mass indexes using artificial intelligence photo-editing to fake their weight. In February 2026, BBC Northern Ireland ordered injections from two major online providers using inaccurate details and an out-of-date photograph, and received the medication the next day. Both firms committed to tightening their safeguards.

I include that record not as flagellation but as evidence of how systemic the problem is. Verification weakness is not a rogue-operator issue at the industry's fringe. It runs through its centre, and the honest response is to fix it and say so.

Advertising has followed the same arc. Prescription-only medicines cannot be promoted to the public, under CAP Code rule 12.12 and the Human Medicines Regulations 2012 in the United Kingdom, and under Article 88 of Directive 2001/83/EC across the European Union. Yet the Advertising Standards Authority's April 2026 enforcement report analysed over 95,000 paid online advertisements and assessed roughly 900, from 38 advertisers, as likely to be breaking the rules. In February 2026 the Authority issued its first rulings against social-media affiliate posts promoting weight-loss injections through referral links and discount codes, naming several of the sector's larger providers. The message was unambiguous. Affiliate marketing is not a loophole.

The European picture is quieter but pointed, and it is worth being precise about what the law does and does not say. No European Union or European Economic Area member state has a rule specifically banning legitimate asynchronous GLP-1 prescribing by licensed clinicians, and claims to the contrary overstate the position. What exists is a lattice of general constraints: Germany's remote-treatment provision in section 7(4) of the Model Professional Code, permitting remote-only care where it is medically justifiable; France's video-mandated and volume-capped teleconsultation framework; and the Irish Medical Council's telemedicine guidance. Alongside those sits escalating enforcement against illegal supply. Ireland's Health Products Regulatory Authority reported 431 GLP-1-related website takedowns or amendments by the end of October 2024, up from 172 the year before. France's medical council has issued disciplinary warnings over sham teleconsultation prescriptions. The European Medicines Agency and the Heads of Medicines Agencies issued a joint public-health warning in September 2025 on falsified GLP-1 products sold through fraudulent websites.

Regulators working across two jurisdictions' worth of frameworks are converging on the same conclusion. The sector's weakest operators are defining how the whole sector gets regulated. Every provider that competes by thinning its checks is writing the case for the next restriction, and the 2025 distance-selling guidance did not appear from nowhere. It was drafted in direct response to demonstrated failures. A race to the bottom does not end with one company winning. It ends with a regulator redesigning the track.

The steelman: the commercial pressure is real

It would be dishonest to pretend the commercial argument for widening eligibility is foolish, and I have no interest in pretending. Acquisition in this category is genuinely expensive, conversion funnels are genuinely leaky, and a stricter intake genuinely rejects revenue. The pressure to grow the eligible population is legitimate, and any medical director who treats commercial colleagues as adversaries for feeling it has misunderstood the job.

The question is not whether to grow the population a service can safely treat. It is which mechanism you use. That is where the loosen-the-gate strategy fails on its own terms, because the economics of this category are not acquisition economics. They are retention economics.

The number the industry is not solving for

The persistence data should alarm every operator in this market.

In the largest European cohort published to date, 157,822 Danish adults initiating semaglutide for obesity reported in JAMA Network Open in August 2026, 49 per cent had discontinued within twelve months. A United Kingdom primary-care cohort in type 2 diabetes found one-year discontinuation of 41.1 per cent (95% confidence interval 40.1 to 42.2), with the highest discontinuation, 49.2 per cent, in patients with a body mass index below 30. That is precisely the population that loosened gates admit. United States managed-care data presented at AMCP 2026 tracked 53,183 initiators and watched persistence fall from 65 per cent at 120 days to 34 per cent at one year.

Discontinuation is not a neutral event. An 18,187-patient analysis presented at the American Diabetes Association's 2026 sessions found that patients who discontinued were 41 per cent less likely to achieve a 5 per cent weight reduction (relative risk 0.59; 95% confidence interval 0.57 to 0.62) and 47 per cent less likely to reach glycaemic target (relative risk 0.53; 95% confidence interval 0.51 to 0.55).

My reading of these data is that the industry has been solving the wrong equation. The binding constraint on growth in this category is not the size of the eligible population. A 2026 Lancet Diabetes and Endocrinology analysis of 99 countries, applying the label criteria to measured national survey data, put eligibility at 42.8 per cent of adults across Europe and North America (95% confidence interval 41.4 to 44.2), and that estimate counted only hypertension and diabetes as qualifying comorbidities, so the true label-eligible figure is likely to be higher. The overwhelming majority of those adults are not in treatment with anyone.

The binding constraint is that half of the patients who do start are gone within a year. An industry facing a retention problem has chosen to compete on acquisition, which tells you more about how these businesses measure themselves than about where their value is created.

Read those numbers as a commercial operator and the strategy critiques itself. A provider that spends heavily to acquire a marginally eligible patient, initiates them through a minimal service, and loses half of them inside a year has not lowered its acquisition cost. It has doubled it, added regulatory exposure, and manufactured a cohort of former patients whose experience of GLP-1 therapy was a few months of nausea and an unanswered message thread. On the United Kingdom evidence, the marginal patients admitted through a widened gate are the likeliest to leave. Loosening eligibility to cut acquisition cost is a strategy of filling a leaking bucket by widening the tap.

Grow capability, not criteria

Here is the argument I want the industry to sit with. The same data that condemns the loose-gate strategy points directly at the alternative, because the strongest predictors of persistence are not patient characteristics. They are service characteristics.

In the AMCP cohort, patients with five or more prescriber contacts were reported to be substantially more likely to remain on therapy at 120 days than those with fewer. That is an association rather than proof of causation, but it is consistent across the literature. The structured-follow-up tradition in weight-maintenance trials shows regain reduced by roughly a third compared with minimal contact. Industry-affiliated real-world data point the same way, and I grade it as such, because these cohorts come from providers with an obvious interest in the result. A United Kingdom digital service's published cohort in Diabetes, Obesity and Metabolism found digitally engaged patients lost 21.5 per cent of body weight at eleven months against 17.0 per cent in less engaged patients, with hazard ratios for every weight-loss milestone favouring engagement. Whatever discount you apply for affiliation, the direction of the whole evidence base is one way. The service around the injection determines whether the injection works, and whether the patient stays.

That is the commercially and medically superior route to a wider population, and here is the opinion I hold most strongly in this debate. The industry has run out of excuses for not building it.

Five years ago, adaptive clinical questioning, outcome analytics and personalised follow-up at scale were genuinely hard: bespoke engineering, expensive data science, uncertain regulatory reception. None of that is true in 2026. Conditional logic is a solved product problem. Every established provider already holds the longitudinal patient data, and analysing it requires intent rather than invention. Artificial-intelligence-assisted pathway design and content generation have collapsed the cost of producing personalised education and structured follow-up to a fraction of what a single television campaign costs. The tools that make deeper care cheap arrived at exactly the moment the sector chose to compete on shallower intake instead, and I find that choice increasingly difficult to defend.

Adaptive intake

Asynchronous medicine has quietly acquired a tool that face-to-face medicine never had: the conditional consultation. A well-built intake no longer asks every patient the same forty questions. It asks fewer questions of the straightforward patient and progressively deeper questions of the complex one, branching on an eating disorder signal, a medication interaction or an ambiguous comorbidity, so that scrutiny concentrates exactly where risk concentrates. That is how you widen eligibility and raise specificity at the same time. Not by asking less, but by asking better.

I will be honest about the evidence gap, because intellectual honesty is the point of this article. There is no published sensitivity and specificity literature on adaptive intake questionnaires in this setting, and there should be. Any provider running one at scale is sitting on the dataset that could establish it.

Outcome-linked data

Every substantial online prescriber already holds longitudinal data linking intake characteristics to titration tolerance, side-effect patterns, engagement and twelve-month persistence. Analysing it is not a moonshot. It is routine data science pointed at a clinical question. A provider that can demonstrate, from its own outcomes, which patient profiles do well under which pathway has an evidence-based case for extending its service to populations a static protocol would exclude, and a defensible answer when a regulator asks why.

The regulatory route for the more ambitious end of this work already exists. The MHRA's AI Airlock sandbox has run since 2024, and its second phase explicitly included weight-management technology. The sanctioned path for artificial-intelligence-supported risk stratification in prescribing pathways is open. The gap is not regulatory permission. It is willingness to do the slower work.

Follow-up as product

The persistence data reframes patient education, side-effect coaching, titration support and proactive engagement from cost centres into the core commercial asset. In a category where half of patients leave within a year, the provider that keeps patients well, informed and on treatment does not need to loosen its gate to grow. Its growth compounds through retention, reorder and reputation while its competitors churn through ever-wider funnels.

There is a quieter benefit too. A service with genuine clinical depth is one that patients who are entirely eligible, but unimpressed by a commoditised and identical-for-everyone prescription, will actually choose. The population of appropriate patients who currently decline online care because it looks thin is a growth market nobody is racing for.

Five questions any operator can ask itself

Strategy documents are cheap, so here is an audit any online prescriber, including mine, can apply to itself this week.

1. If a journalist submitted false height, weight and photographs tomorrow, would your systems catch it before a medicine shipped?
2. Does your intake deepen when a risk signal appears, or does every patient answer an identical form?
3. Can you link intake characteristics to twelve-month outcomes in your own data, and have you looked?
4. What proportion of your patients remain on treatment at one year, and can you say, patient by patient, why the others left?
5. When you last widened an eligibility criterion, what safeguard did you strengthen on the same day?

A provider that cannot answer these is not running a clinical service with a commercial engine. It is running a fulfilment operation with a questionnaire in front of it.

Where this ends

The GLP-1 era is the first time online prescribing has operated at genuine population scale, and the sector is writing its own regulatory future in real time. One path is the current race: wider gates, thinner checks, acquisition-cost arithmetic, and a steady drumbeat of investigations and enforcement until regulators impose the discipline the industry declined to impose on itself. The other grows the treatable population the way medicine has always legitimately grown it, by getting better at looking after people and proving it.

I believe in this industry's mission more than most of its critics and, on the evidence above, more soberly than some of its operators. Asynchronous prescribing done well is not a lesser form of medicine. It reaches people whom traditional services have failed for decades, at a price and convenience clinic-based care cannot match, and the population who could benefit is measured in the millions. That is exactly why the race matters. A sector this young, handling medicines this consequential, gets one reputation, and it is currently letting its least careful members write it.

The second path is slower to build and harder to copy, which is precisely why it is worth building. The companies still standing in five years will be the ones that understood that in this category the gate was never the product. The care behind it is.