GLP-1 and Nausea: the Label Says 44%, the Placebo Arm Says 16%, and the Same Dose Reads 18% in the Largest Trial

Ask an AI "how likely is nausea on Ozempic?" and the number that comes back is almost always 44%. Unusually for this series, that number holds up.

The US prescribing information for Wegovy puts nausea at 44% of adults on semaglutide 2.4 mg against 16% on placebo. The label does not name the three trials it pools for that table, but their arm sizes identify them exactly: 2,116 on the drug and 1,261 on placebo is STEP 1 (1,306 and 655) plus STEP 3 (407 and 204) plus STEP 2 (403 and 402), and the label's own hypoglycaemia footnote names its "Study 3" at N=403 and N=402, which is STEP 2's pair. Add up those three trials' own nausea rows as the registry posts them — 576 + 237 + 135 of 2,116, and 114 + 45 + 37 of 1,261 — and you get 44.8% against 15.5%, where the label prints 44% and 16%. Within a point on both arms. We cannot reconcile the last 0.8 of a point from anything either source publishes, and we are not going to invent a reason for it.

That near-agreement is worth a moment, because this series has spent three articles explaining why it usually does not happen. The label's abdominal pain row groups seven separate adverse-event terms, so it reads 20% where the single registry term reads 9.7%. Its fatigue row groups two. Its gastritis row groups four. The nausea row groups nothing: in the label's own footnote list, nausea carries no marker at all. Which means the 44% is a figure you can quote as-is — and it also means the interesting questions about nausea are not about the headline. They are about the 16% underneath it, and about what happens to the same molecule at the same dose in a bigger trial.

This is the eleventh piece in our evidence series, after which effects have enough evidence to calibrate a model, the corpus-derived pooled rates, what patients themselves report, gallstones end to end, what is inside a personalised risk number, hair loss end to end, why trial vomiting rates run from 6.5% to 58%, constipation read against the placebo arm, headache read against the placebo arm and abdominal pain, where the label and the registry disagree by a factor of two. Like those, every figure below is quoted from a primary source with the arm and the number of participants it belongs to, and every one of them was read on 17 September 2026 — from the current US prescribing information, from each trial's own results table on ClinicalTrials.gov, or from our own live production API.

What the label says, and the footnote that is not there

The current US prescribing information for Wegovy (semaglutide), revised June 2026, lists nausea first among the *"most common adverse reactions (incidence ≥5%) in adults or pediatric patients aged 12 years and older"*. Its three adverse-reaction tables each give a drug column and a placebo column:

Five footnotes sit beneath Table 3. Footnote a belongs to the abdominal pain row: *"Includes abdominal pain, abdominal pain upper, abdominal pain lower, gastrointestinal pain, abdominal tenderness, abdominal discomfort and epigastric discomfort."* Footnote b belongs to fatigue: *"Includes fatigue and asthenia."* Footnote d belongs to gastritis: *"Includes chronic gastritis, gastritis, gastritis erosive, and reflux gastritis."* The nausea row carries no marker. (Quoted from the prescribing information via DailyMed, set ID ee06186f, revised 6/2026.)

Section 6.1 describes Table 3's population as *"3 randomized, double-blind, placebo-controlled trials that included 2,116 adult patients with obesity or overweight treated with 2.4 mg WEGOVY injection for up to 68 weeks and a 7-week off-drug follow-up period"* — three trials it never names. The arithmetic above identifies them, and it is worth stating plainly that this is our reconstruction from arm sizes, not something the label says: the three registry records whose 2.4 mg arms sum to exactly 2,116 and whose placebo arms sum to exactly 1,261 are STEP 1, STEP 3 and STEP 2, and footnote c's "Study 3, WEGOVY N=403, Placebo N=402" matches STEP 2's arms exactly. Their individual nausea figures are 44.1% against 17.4% (STEP 1), 58.2% against 22.1% (STEP 3) and 33.5% against 9.2% (STEP 2, on the 2.4 mg arm) — a 25-point spread between the three trials the label presents as one 44%.

Two more sentences from the same document matter more than the percentage, and we quote them rather than paraphrase. On when the reactions happen, section 6.1: *"These reactions were most frequently reported during dosage escalation."* On what to do about it, section 2.2: *"If patients do not tolerate a dose during dosage escalation, consider delaying dosage escalation for 4 weeks."* The 44% is a cumulative figure over the whole trial, not a description of any given week on the drug, and the label itself says the timing is front-loaded.

What the trial registries say, arm by arm

Our database cited 28 distinct studies for a stated nausea rate when this article was written at 06:50 BST on 17 September 2026 — more than any other effect we track. Twelve of those post a placebo or comparator arm of at least 20 people. Every row below was read from that trial's own ClinicalTrials.gov adverse-events module on 17 September 2026, from its non-serious events table, for the single term "Nausea". The counts are the registry's; the percentages and the differences are our division and subtraction.

Four more studies compare a GLP-1 against another active drug rather than against placebo, and they are the ones a reader chasing a "which drug is worse" answer should look at, because both arms were watched the same way. SURPASS-CVOT (NCT04255433) puts tirzepatide at 1,665 of 6,647 (25.0%) against dulaglutide 1.5 mg at 1,484 of 6,647 (22.3%) — 13,294 people, a 2.7-point gap. SURPASS-SWITCH (NCT05564039) reads 35 of 139 (25.2%) against 27 of 143 (18.9%). ACHIEVE-3 (NCT06045221) puts orforglipron 12 mg and 36 mg at 112 of 424 (26.4%) and 116 of 423 (27.4%) against oral semaglutide 7 mg and 14 mg at 56 of 426 (13.1%) and 90 of 425 (21.2%). And REALYSE (NCT05035082), a phase 4 study of oral semaglutide in 500 adults with type 2 diabetes in the United States, reads 38 of 500 — 7.6% — against 10 of 492 (2.0%) on other oral glucose-lowering tablets. That comparator arm is not a GLP-1, so it was withheld from our base on 8 September; the oral semaglutide arm stays, and it is the lowest figure any phase 3 or phase 4 trial in our nausea base reports.

The remaining twelve sources have no placebo or GLP-1-comparator arm we can read a difference from. Seven are orforglipron pharmacology studies — dose-escalation, formulation, renal-impairment and drug-interaction work, dosed for days or weeks in healthy volunteers or small cohorts (NCT06440980, 38 arms from 0.0% to 20.7%; NCT05841238, 18 dosing periods from 0.0% to 41.7%; NCT03929744, 21 arms from 0.0% to 50.0%, with placebo parts of 8, 15 and 2 people; NCT05086445, placebo parts of 4 and 11; NCT06023095, a 24-person Chinese study whose two 10-person cohorts read 40.0% and 80.0% against 0 of 4 on placebo; NCT05936138, a single-dose renal study of 24 people; and NCT06370728, a drug-interaction study whose orforglipron-alone and orforglipron-plus-carbamazepine periods read 1 of 30 and 2 of 28). Two post no placebo arm at all: ACHIEVE-J (NCT06010004), where orforglipron reads 12.9%, 19.3% and 27.6% across its three doses, and a chronic-kidney-disease study of dulaglutide with one seven-person arm (NCT05254418, 1 of 7). Three are papers rather than registry records, and are discussed below.

What the comparison actually shows

In every placebo-controlled arm of more than 300 people, the drug adds between 11.7 and 28.0 points, and never subtracts. That is ten arms across five trials — STEP 1, STEP 2, SELECT, SURMOUNT-1 and ATTAIN-2 — covering semaglutide, tirzepatide and orforglipron. Of the three effects this series has now read arm-by-arm against the placebo column, nausea is the only one with no drug arm anywhere below its own placebo: headache had several, and abdominal pain had four of eight sources with at least one. Whatever else is arguable about GLP-1 side-effect numbers, that the drugs cause nausea is not.

The placebo arms themselves run from 0% to 25%, and that range is the reason the headline figure travels badly. ATTAIN-J's 60 Japanese participants on placebo reported no nausea at all; ACHIEVE-1's 138 reported 2.2%; SELECT's 8,801 reported 3.8%; STEP 2's 402 reported 9.2%; STEP 1's 655 reported 17.4%; the adolescent oral-semaglutide trial's 66 reported 13.6%; and STABLE's 48 people on placebo plus a behavioural weight-management programme reported 25.0%. The same inert injection or tablet produces anything from no recorded nausea at all to a quarter of the arm, depending on who is enrolled and how they are asked. A 44% that is not read beside its own 17.4% is being quoted as if the placebo number were zero.

Dose-response is real but not strictly monotonic at the top. Nausea rises with dose in ATTAIN-2 (20.1% → 31.1% → 36.4%), ATTAIN-J (13.1% → 15.8% → 33.3%), ACHIEVE-J (12.9% → 19.3% → 27.6%) and ACHIEVE-3's oral semaglutide arms (13.1% at 7 mg → 21.2% at 14 mg). It does not in SURMOUNT-1, where 10 mg tirzepatide (34.1%) reports more nausea than 15 mg (31.9%), or in ACHIEVE-1, where 12 mg orforglipron (19.0%) reports more than 36 mg (17.0%). Both non-monotonic pairs are within a few points, on arms of 137 to 636 people, so the honest reading is that the top two doses of these drugs are not reliably distinguishable from each other for nausea — not that the higher dose is gentler.

The same molecule, the same dose, 44% and 18%

The sharpest number in this article is not the 44%. It is that STEP 1 and SELECT tested the same molecule at the same dose by the same route and posted 44.1% and 18.1%. Both are semaglutide 2.4 mg once weekly. Both post to ClinicalTrials.gov with a 5% reporting threshold. SELECT is the larger by a factor of seven (17,604 participants against 1,961) and ran far longer — adverse events collected to week 240, against week 75 in STEP 1 — so the usual explanation, that a longer trial accumulates more events, points the wrong way: the trial with three times the follow-up reported less than half the nausea.

What differs is who was enrolled and what they were enrolled for. STEP 1 recruited adults with obesity or overweight into a weight-reduction trial; SELECT recruited adults with established cardiovascular disease and overweight or obesity into a cardiovascular-outcomes trial, where the primary endpoint is heart attacks and strokes over years. We can state that difference because both registry records state it. We cannot tell you from the registry how systematically each trial solicited gastrointestinal symptoms at each visit, because the records do not publish that in a form we can quote — and that, rather than any property of the drug, is the most likely place a 26-point gap comes from. The practical consequence for anyone quoting a number: the nausea rate of a GLP-1 is not a property of the drug alone. It is a property of the drug, the population and the way the trial asked. REALYSE's 7.6% on 500 people taking oral semaglutide in a phase 4 study in the United States is the same point from the other end of the range.

Our own pooled estimate, and why it is half the label's

Our nausea endpoint published a pooled clinical estimate of 22.5% (95% CI 7–51%), from 133 eligible stated rates across 28 distinct studies, source diversity "very high", when this article was written at 06:50 BST on 17 September 2026. That is almost exactly half the label's 44%, and the reason is arithmetic rather than disagreement.

Our pooled figure collapses each study to one entry — the unweighted mean of every arm the registry posts, carried at the size of its largest arm — and then weights those entries by stated sample size, discounted by how confident our extraction was in reading the figure. The consequence is that the two cardiovascular-outcomes trials dominate: SELECT enters at 18.1% with n=8,803 and SURPASS-CVOT at 23.7% with n=6,647 (the mean of its tirzepatide and dulaglutide arms, both GLP-1 drugs), and between them they outweigh every weight-management trial in the base. STEP 1 enters at 44.1% with n=1,306; SURMOUNT-1 at 30.5% with n=636; REALYSE at 7.6% with n=500; ACHIEVE-3 at 22.0% with n=426; STEP 2 at 32.7% with n=403; ATTAIN-2 at 29.2% with n=331. The pooled number is 22.5% because the largest trials of this class are cardiovascular-outcome trials in older, sicker populations that report nausea in the high teens — not because anybody thinks Wegovy's weight-management figure is wrong.

Two features of the base are worth knowing before quoting the interval. First, seven of the 28 studies are orforglipron phase 1 pharmacology work, and they contribute 89 of the 133 stated rates — most of the rows, almost none of the weight, and a spread from 0.0% to 80.0% that is a large part of why the interval runs from 7% to 51%. Second, one entry is an artefact of the arm-averaging rule worth naming: STABLE enters at 36.8%, the mean of its 73.6% treatment phase (weeks 0–60, n=72) and a 0.0% continuation phase (weeks 60–72, n=10). Neither number is wrong and the mean is what our published method produces, but a 120-person appetite-and-eating-behaviour study is not evidence about the class at the weight its 36.8% suggests.

If you want the Wegovy 2.4 mg number, quote the label: 44% against 16% on placebo, N=2,116 and N=1,261. If you want one trial's registry record, quote STEP 1: 44.1% against 17.4%. If you want the whole class with its heterogeneity visible, quote ours, with its interval and its source count. None of the three is the other two, and for nausea — unusually — the first two agree.

The three papers in the base

Twenty-five of our 28 nausea sources are trial registry records. The other three are papers, and because a paper's numbers are prose rather than a table, we re-read all three in full from NCBI before this article.

PMID 33185364, a placebo-controlled trial of subcutaneous semaglutide in non-alcoholic steatohepatitis, states *"nausea, 42% vs. 11%"* for its 0.4 mg group of 82 against placebo — our row carries 42% at n=82, the drug arm. PMID 37062422, a phase 1 study of TG103, an investigational long-acting GLP-1 fusion protein, in 24 dosed healthy Chinese subjects, states nausea at *"20.8%"* — ours exactly. The third, PMID 41445996, is a narrative review of GLP-1 drugs in plastic surgery whose stored evidence sentence reads *"SUSTAIN-3 showing … nausea (22%)"* — a real, verbatim figure, but one the review is quoting from a trial rather than measuring. Whether a secondary source that names a trial should count as its own entry in a "distinct studies" total is a methodology question we have open at the time of writing; it is disclosed here because it is the kind of thing a reader should be able to see rather than discover.

What the FDA's adverse-event reports show

A different kind of number sits outside the incidence pool on purpose. FDA FAERS publishes the share of *spontaneous adverse-event reports* for each drug that mention a term, and nausea is the most-reported of the terms we track: 11,506 of 82,377 semaglutide reports (14.0%), 7,118 of 43,057 liraglutide reports (16.5%), 12,040 of 94,688 exenatide reports (12.7%), 12,028 of 122,084 tirzepatide reports (9.9%), 9,986 of 101,618 dulaglutide reports (9.8%) and 142 of 3,364 lixisenatide reports (4.2%), each as read from our own endpoint on 17 September 2026. A share of complaints is not a share of patients — it answers "of the people who filed a report about this drug, how many mentioned nausea", not "how many people on the drug got it" — so our eligibility rules report these separately rather than average them in.

Two honest caveats about that list, both of which apply to our own published surface. Our endpoint also carries orforglipron at 50.0% — that is 1 report out of 2 on file, a figure that is arithmetically correct and substantively meaningless; we print it with its count so it can be discarded, and we are naming it here rather than hoping nobody divides. And each row's denominator is the total on file at the moment that row was collected, so the semaglutide rows on our site do not all share one denominator: the nausea row was read against 82,377 reports and the abdominal pain row against 99,460. Every row states its own total in its stored excerpt, but the API currently publishes the share and the count without the denominator beside them. That is a disclosure gap on our side, not a data error, and it is on our list to fix.

What patients themselves report

Of the 26 distinct community reports in our corpus, 10 mention nausea — a 38.5% reporting frequency, the second-highest of any effect we track, behind vomiting at 42.3% (11 of the same 26). That is not an incidence rate and we never treat it as one: it is the share of people who chose to post about their GLP-1 experience who mentioned nausea at all, and at ten reports it is a count, not a rate. The pool is small and cannot currently grow — Reddit has blocked our collector since 28 May 2026 — so that figure is fixed at its date rather than continuously updated. See that article for the full accounting.

What we checked, and did not find

Three consecutive articles in this series have been preceded by reading every source of the effect's base, and each one found a live wrong number: a comparator arm of an anticonvulsant pooled as a GLP-1 rate, a bloating figure filed as abdominal pain, and a four-arm trial's whole population published as one arm's denominator. So it is worth reporting the negative result plainly. All 130 stored registry rates, across all 25 registry records, were re-fetched from the ClinicalTrials.gov API this morning and compared against what we store, arm label by arm label: every numerator, every denominator and every arm attribution matched the registry. The three papers behind the remaining three rates were re-read in full. Nothing was withdrawn, and no published nausea figure moved.

What the same reading did surface is a gap of a different kind, and it is the most useful thing in this article for anyone judging how far to trust a pooled figure from anybody. STEP 3 — one of the three trials behind the label's own 44% — was not in our nausea base at all. It posts its results publicly, its nausea row reads 237 of 407 against 45 of 204, and we quote it above from the registry directly, because our collector had simply never fetched that record: trials enter our corpus either through a rotating keyword search of ClinicalTrials.gov or through a short pinned list of pivotal studies, and STEP 3 was in neither. Checking how general that is, we ran a scripted census against ClinicalTrials.gov and our own production store. Every phase 3 or phase 4 trial of the nine drugs we track that has posted its results: 319 studies. Of those, 143 enrolled 400 people or more, and 137 of the 143 post a non-serious adverse-event row naming at least one of the fourteen effects a registry adverse-event table can express in our taxonomy (emotional blunting has no MedDRA term the registry posts, so no trial record can supply it — "muscle loss" is not one of the fifteen effects this site publishes in the first place: it is part of a wider 18-id collection taxonomy, never shown on any page, so it was never eligible to be counted here either way). That last set is the one that matters, because those are the records that could contribute an incidence rate to this corpus today. We have read 10 of the 137. The other 127 have never had their posted results collected, among them SUSTAIN 6, SOUL, SURMOUNT-2, PIONEER 3, ATTAIN-1 and SURPASS-3. Counted by people rather than by trials: 41,902 enrolled participants in the records we hold, 140,182 in the ones we have not read. SURPASS-2 and SURPASS-4 are in the corpus by another route — their journal abstracts — but not through the arm-by-arm registry table this census counts, and what those abstracts give is thin: all four SURPASS-4 rows are withheld as range midpoints, and of SURPASS-2's five rows only two carry a live rate, both of them stated verbatim for its semaglutide 1 mg comparator arm (diarrhoea 12%, vomiting 8%) while its tirzepatide figures, published as dose ranges, are withheld. STEP 3 was added to the pinned list on 17 September 2026 and will be collected in full on the next pipeline run. The other 126 are not one fix but two. Reading the arm structure of all 127 records: 62 of them have no arm naming a drug outside our nine-drug list (placebo arms aside), so those are gated on collector capacity, not on correctness. The remaining 65 carry at least one arm that does — insulin glargine, insulin aspart, sitagliptin, empagliflozin, and combination products such as IcoSema — and our collector skips placebo arms by name but has no general rule for a non-GLP-1 comparator arm, which is precisely the defect that has had to be corrected by hand three times this month. Bulk-adding those 65 before the rule exists would put insulin's nausea rate into a GLP-1 average.

*Correction, 17 September 2026.* When this article first published earlier the same day, this paragraph put the gap at "107 trials". That figure came from an ad-hoc query run while writing, and it could not afterwards be reproduced; the scripted census above replaces it and puts the gap larger, not smaller. The census is stated here in full so that anyone can re-run it: intervention name matching each of the nine drugs, phase 3 or phase 4, results posted, deduplicated by NCT id, enrolment of 400 or more, and at least one non-serious adverse-event row whose MedDRA term maps to an effect we publish. The same pass corrected a second sentence: this paragraph originally said that *most* of the unread trials randomise against a non-GLP-1 active comparator, which was an impression, not a count. Reading the arm structure of all 127 records puts it at 65, just over half, with 62 clean — so roughly half of the gap needs no new rule at all. Both numbers move over time — as sponsors post new results, and as our collector reads more of them — so this is a capture taken on 17 September 2026, not a constant.

So the honest statement of our coverage is: every figure we publish is traceable to a trial that states it, and the set of trials we have read is a small and non-random subset of the trials that exist — 10 of 137, on the strictest count we know how to make of our own gap. Nobody publishing a pooled GLP-1 side-effect number can say much better than that, and most do not say it at all.

The base was clean partly because the gates that catch this class had already fired on it. Five nausea rows sit withheld in our store today, each for a reason that is printed in the row: the "Other Oral Glucose-lowering Medication" arm of REALYSE and the carbamazepine-only period of the orforglipron drug-interaction study (comparator arms, not GLP-1 rates); SELECT's serious-adverse-event row, 12 of 8,803, which would otherwise have been pooled with its 18.1% all-cause figure; a range midpoint from a SURPASS-4 abstract that states "5-9%" and never states its middle; and a Nature genome-wide association study of 27,885 GLP-1 users that discusses differential nausea risk but states no rate at all. A clean base is not the same as a base nobody checks.

If you are on a GLP-1 now

This is educational content, not medical advice. Nausea is the most common adverse reaction on every label in this class, and the label is specific about when it happens — *"most frequently reported during dosage escalation"* — and about one response to it: *"If patients do not tolerate a dose during dosage escalation, consider delaying dosage escalation for 4 weeks."* That is a decision for your prescriber, not a self-service instruction, but it is worth knowing that a slower escalation is a recognised option in the label itself rather than a concession.

The part worth quoting rather than paraphrasing is what nausea can lead to. Under acute kidney injury, the label reports that *"the majority of the reported events occurred in patients who experienced gastrointestinal adverse reactions leading to dehydration such as nausea, vomiting, or diarrhea."* Nausea that stops you keeping fluids down, for more than a day or two, is not just uncomfortable — it is the documented route to the more serious problem, and it is a reason to contact a clinician rather than to wait it out. Nothing here should be used to start, stop or change a medication or its dose without your prescriber.

Every figure in this article was read on 17 September 2026 from the cited primary source or from our live production API, and is stated as of that date. Query everything behind it — all 15 effects, both tracks, every source with its URL — at magistra.health/en/data-api, with the full eligibility methodology at magistra.health/en/methodology.