Semaglutide & Brain Fog: What the Cognition Trials Found

Two completely different questions get filed under "GLP-1 brain fog," and conflating them is why the internet is so confused about it.

The first is symptomatic: I started semaglutide and my head feels slow — is the drug doing that? That one is mostly answered by looking at eating, hydration, blood sugar and sleep, and we take it apart in brain fog on Ozempic and other GLP-1s.

The second is the claim that has driven a decade of research funding and a great deal of marketing: do these drugs protect or improve cognition? That question now has real answers, and they arrived recently. This page is about the second question, because the trials that were supposed to settle it have reported — and they were negative.

The trials that were meant to prove it: EVOKE and EVOKE+

These were the definitive test, and they were enormous. EVOKE and EVOKE+ were two phase 3, multicentre, randomised, double-blind, placebo-controlled trials of oral semaglutide 14 mg once daily against placebo for up to 156 weeks, run across 566 sites in 40 countries, in people aged 55–85 with amyloid-confirmed early Alzheimer's disease. 3,808 participants were randomised. The primary endpoint was change in the Clinical Dementia Rating–Sum of Boxes (CDR-SB) at week 104¹.

The result:

• •EVOKE: estimated difference −0.08 (95% CI −0.35 to 0.20), p=0.57
• •EVOKE+: estimated difference 0.10 (95% CI −0.17 to 0.38), p=0.46

Both confidence intervals sit squarely across zero. The authors' own interpretation is one sentence: "Oral semaglutide was not efficacious in slowing clinical progression in participants with early Alzheimer's disease"¹. The trial registration entry is blunter still — both trials were "discontinued due to negative clinical outcome"¹.

It is worth sitting with the scale of that. This is not an underpowered pilot that failed to reach significance. It is 3,808 people, three years, forty countries, a prespecified primary endpoint, and a point estimate essentially at zero in both trials. When a hypothesis gets a test that good and returns nothing, the correct update is substantial.

The Parkinson's trial: also null, and it comes with a caveat you must be told about

The other flagship neurological test of the class was exenatide in Parkinson's disease: a phase 3, multicentre, double-blind, randomised, placebo-controlled trial across six UK research hospitals, 194 participants, extended-release exenatide 2 mg weekly for 96 weeks. The primary outcome, the MDS-UPDRS part III score off medication at 96 weeks, showed an adjusted coefficient for exenatide of 0.92 (95% CI −1.56 to 3.39), p=0.47². The authors concluded they "found no evidence to support exenatide as a disease-modifying treatment for people with Parkinson's disease"².

And that paper now carries an Expression of Concern. The Lancet published one in June 2026³. An Expression of Concern is a formal notice that a journal is looking into questions about a paper; it is not a retraction and it does not reverse a published finding. It does not turn a null result into a positive one. But citing this trial without disclosing it would be a citation-integrity failure, so: the trial reported no benefit, and the paper reporting it is currently under editorial scrutiny. Treat the result as weaker evidence than its design would otherwise warrant — in either direction.

The counterweights, reported honestly

There is a genuine positive signal in this literature, and it deserves to be described accurately rather than either buried or inflated.

The REWIND trial randomised 9,901 people with type 2 diabetes to dulaglutide or placebo and followed them for a median of 5.4 years. Its exploratory cognitive analysis defined the outcome as the first follow-up score on the Montreal Cognitive Assessment or Digit Symbol Substitution Test falling 1.5 standard deviations or more below the baseline country mean. On the prespecified analysis, the result was null: hazard ratio 0.93 (95% CI 0.85–1.02), p=0.11⁴.

You have almost certainly seen a different number from this trial: a 14% reduction in cognitive impairment. That figure is real, and it is post hoc. After adjusting for individual standardised baseline scores — an adjustment decided on after the data were in hand — the hazard ratio became 0.86 (95% CI 0.79–0.95, p=0.0018)⁴. A post-hoc adjustment that converts a non-significant result into a significant one is a hypothesis, not a finding. It is legitimate to report; it is not legitimate to lead with. When a headline cites "14%" from REWIND without saying "post hoc," that is the tell.

Beyond that, a systematic review and meta-analysis of GLP-1 receptor agonists and neuropsychiatric outcomes found no consistent evidence that the class worsens cognition or mood at the group level⁵. That is a reassuring finding about safety. It is not a finding of benefit, and the two get swapped constantly.

What this means if you are just foggy

None of the above tells you anything about a 41-year-old with a foggy head six weeks into Wegovy — and that is the point. The trials above studied disease progression in Alzheimer's and Parkinson's, in older populations, over years, with clinical rating scales. They do not translate to everyday cognitive complaints in either direction.

So on the two claims people actually encounter:

"GLP-1s are good for your brain." Not established. The most direct, best-powered tests of that idea failed¹², and the most-quoted supporting number is post-hoc⁴. Mechanistic and observational work continues, and it may yet find something in a different population or at a different disease stage — but as of now, no GLP-1 drug has been shown in a randomised trial to improve cognition, focus or mental energy in anyone.

"GLP-1s fog your brain." Also not established, and the group-level safety data point the other way⁵. Cognitive impairment is not a labeled effect of these drugs. When fog does show up, it is usually downstream of the drug working: eating far less than you need, mild dehydration, and unstable blood sugar are each independently well-documented causes of slowed thinking⁶⁷, and disrupted sleep degrades attention and working memory on its own⁶.

A related claim deserves its own flag, because it is spreading fast: quieter "food noise" is frequently sold as freed-up mental bandwidth. The evidence for that chain does not exist — there is no randomised evidence that a GLP-1 reduces food noise, and no study linking a food-noise change to any objective cognitive measure. We go through that literature, including the recall-survey it rests on, in what food noise actually means.

What to check instead

If your thinking feels slow on a GLP-1, the productive move is to work through the causes that are actually measurable rather than to reason from Alzheimer's trials.

Start with intake — most people on these drugs eat considerably less than they estimate, especially in the two weeks after a dose increase. Then fluids, since a 1–2% body-weight deficit blunts attention before thirst registers. Then glucose, particularly if the fog comes with shakiness or lifts after eating, and doubly so on insulin or a sulfonylurea⁷; the wider picture is in blood sugar and brain fog. Then sleep⁶ — and if the tiredness is more physical than mental, what the FDA labels actually say about fatigue on a GLP-1 has the numbers, including which products report none at all.

If none of those fit, the fog probably has nothing to do with the injection. What actually causes brain fog covers thyroid, iron, mood and medication effects, and how to clear brain fog is the cause-first playbook. Fog that is severe, persistent, or accompanied by other neurological symptoms is a reason to call your clinician rather than to adjust anything yourself.

The bottom line

The GLP-1 cognition hypothesis got the best test money can buy and failed it. EVOKE and EVOKE+ randomised 3,808 people to oral semaglutide or placebo for up to three years and found a difference of −0.08 and 0.10 on CDR-SB, p=0.57 and p=0.46; both trials were discontinued for negative clinical outcome¹. Exenatide in Parkinson's was likewise null at p=0.47, and that paper now carries a Lancet Expression of Concern²³. REWIND's famous 14% is the post-hoc number; the prespecified analysis was null at p=0.11⁴. Meanwhile the class does not appear to harm cognition at a group level⁵. That is the entire honest picture: no proven cognitive benefit, no evidence of cognitive harm, and a set of ordinary, checkable reasons why someone taking one of these drugs might nonetheless feel foggy.