Every one of these drugs slows the stomach, and a tablet has to leave the stomach before it can be absorbed. From that pairing the internet has built a general warning that swallowed medicines stop working properly. The manufacturers ran the trials that test it, the regulators published the results, and what those results show is narrower and stranger than the warning: the rate of absorption moves reliably, the total amount absorbed mostly does not, and where the total did move significantly it went up. The same slowing is what makes a stomach still full before anesthesia a live concern, so the mechanism is not in doubt — only its consequences for a pill.
What the labels actually name
The tirzepatide label warns that gastric emptying is delayed, asks for caution with concomitant oral medications, and says to monitor patients on medicines that depend on threshold concentrations or have a narrow therapeutic index, giving warfarin as its example. Its one concrete instruction concerns oral hormonal contraceptives.[1]
The semaglutide label makes the same general point, then supplies the single figure it has: levothyroxine exposure increased 33% in a trial with the semaglutide tablet, with advice to consider increased clinical or laboratory monitoring where the therapeutic index is narrow.[2]It also states that in clinical pharmacology trials with the 1 mg weekly injection, semaglutide did not affect the absorption of orally administered medications. Two swallowed drugs, one named for each brand, is the whole of the specific interaction advice these labels carry. Everything else circulating about the class is inference from the mechanism.
There is a third labeled interaction, and it is not about absorption at all. Both labels instruct prescribers to consider reducing the dose of a concomitant insulin or insulin secretagogue such as a sulfonylurea when starting, because the GLP-1 lowers blood glucose on its own and the two effects stack into hypoglycemia.[1][2] That is a pharmacodynamic interaction — two drugs doing the same thing — rather than a pharmacokinetic one, and it is the only interaction in this class that comes with an instruction to change a dose. It applies to people treating diabetes rather than weight, a distinction drawn in the article on the two indications.
What the interaction trials measured
Healthy subjects took metformin, warfarin, atorvastatin or digoxin before and during subcutaneous semaglutide at steady state. Lack of interaction was pre-specified as a 90% confidence interval for the exposure ratio falling inside 0.80 to 1.25. Every one of the four landed inside it, peak concentrations were not meaningfully changed, and warfarin produced no clinically relevant shift in international normalized ratio.[3]
The tablet was tested the same way against lisinopril, warfarin, digoxin and metformin across two crossover trials in 84 subjects. Three showed no apparent effect. Metformin exposure rose 32% (90% CI, 1.23 to 1.43) with peak concentration unaffected — judged not clinically relevant on the grounds that metformin has a wide therapeutic index.[4]A combined oral contraceptive given to 43 postmenopausal women with type 2 diabetes met the bioequivalence criterion against weekly semaglutide for ethinyl estradiol, at a ratio of 1.11 (1.06 to 1.15), and came in 20% higher for levonorgestrel, at 1.20 (1.15 to 1.26).[5]
The reliable effect is on the peak, and it fades
A 2024 systematic review pulled together 22 reports and six prescribing sheets covering injectable GLP-1 receptor agonists given with oral drugs. Across all four biopharmaceutics classes — highly soluble and permeable drugs such as warfarin and acetaminophen, poorly permeable ones such as ACE inhibitors, poorly soluble ones such as statins, and drugs poor at both such as digoxin — the pattern was the same: peak concentration unaffected or reduced, time to peak delayed, and area under the curve without clinically significant change.[6]
The size of that peak effect, and how quickly it wears off, was measured directly with acetaminophen. After a first 5 mg dose of tirzepatide, peak acetaminophen concentration fell 55% and the peak arrived an hour later, while total exposure over 24 hours was unchanged. By week six, at 15 mg, there was no meaningful impact on either.[1] The underlying phase 1 work found the same shape in participants with and without type 2 diabetes, with the delay diminishing after multiple doses in healthy participants and a residual delay persisting in those on an escalating schedule.[7] That is why the risk window in this class tracks the dose-escalation calendar rather than the whole course of treatment.
The exception a label turns into an instruction
Tirzepatide with a combined oral contraceptive is the one place the numbers were large enough to change behavior. With a single 5 mg dose, peak concentrations of ethinyl estradiol, norgestimate and norelgestromin fell by 59%, 66% and 55%, total exposure fell by 20%, 21% and 23%, and time to peak was pushed back 2.5 to 4.5 hours.[1] The four-week backup instruction that follows, and why it reopens at every dose step, is set out in the contraception article.
Thyroid replacement is the mirror image — an increase rather than a decrease, and a monitoring note rather than an instruction — and it has its own page. Those two medicines are the entire named list.
Modeling points at drugs nobody has dosed
The clinical program tested the drugs a diabetes trial population happened to be taking. A 2025 physiologically based pharmacokinetic analysis asked what the same motility delay would do to drugs that were never tested, simulating a virtual cohort of 1,000 adults with obesity using gastric emptying data from capsule endoscopy in liraglutide-treated patients.[8]
Predicted exposure increases were 64% for rosuvastatin, 90% for valsartan, and 205% for dabigatran — the last of these a narrow therapeutic index anticoagulant. These are model outputs, not measurements: no person in that analysis took a dose of anything. The authors say so, and call for clinical validation and pharmacovigilance. What the analysis establishes is where the untested territory lies, which is a different and more useful claim than a risk estimate.
The real-world signals are not about absorption
Warfarin is the drug the labels flag, and the largest real-world look at it found an effect that has nothing to do with the stomach. Among 1,021 patients screened from a research network and followed six months either side of starting a GLP-1, time in therapeutic INR range fell from 64.2% to 62.1%, a change of −2.1% (95% CI, −3.7% to −0.6%; p = 0.01). Mean INR did not move at all (0.00; 95% CI, −0.02 to 0.03; p = 0.79), while time above the therapeutic range rose 1.3% (p = 0.04).[9]The proposed mechanism is reduced vitamin K intake from eating less, not altered drug absorption.
A 2025 case series makes the same point about a different drug. Three patients on stable lithium regimens started semaglutide; two developed toxicity with rising levels despite stable renal function and no other medication changes, and the third stayed elevated even after a preemptive dose reduction. The mechanisms the authors propose are dehydration from reduced intake, vomiting or diarrhea, and altered kidney function alongside delayed emptying.[10] Lithium is not an antidepressant, and the separate question of SSRIs, SNRIs and bupropion has its own data. What both examples share is that the interaction runs through eating and hydration rather than through the tablet.
What the evidence does not cover
The systematic review names its own limits plainly: the studies enrolled healthy subjects, and there is insufficient data in conditions that change pharmacokinetics, kidney dysfunction among them.[6] Reduced renal clearance is exactly the setting where a 30% exposure shift stops being academic, and it is covered in the renal article. Beyond warfarin, no narrow-therapeutic-index drug has been formally studied with these agents in people.
There is also nothing at all on compounded formulations. Compounded semaglutide and tirzepatide are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before dispensing, and no compounded preparation appeared in any interaction trial described here. The delayed emptying is a property of the molecule and should behave the same way; the dose delivered by a given vial is a separate question addressed in the article on vial contents.
What is actually actionable
Three things follow from the measured record. A complete medication list at intake is the only input that makes any of this checkable, including over-the-counter and supplement use. Drugs judged by a blood level — warfarin, lithium, thyroid replacement, antiepileptics, immunosuppressants — are the ones where a level ordered around a dose change costs little and catches the shift. And the weeks after starting and after each escalation are where the effect concentrates, because that is where the emptying delay is largest before it adapts.
None of that is a reason to stop a medicine, and no website is positioned to adjust one. The standard these readings are held to is described in the methodology.