What GLP-1 Users Actually Report: Side-Effect Mentions Across 185 Community Reports

Our first two pieces in this series covered what clinical sources say about GLP-1 side effects: which effects have enough evidence to calibrate a statistical model, and the pooled clinical rate for each. This one is about the other track of Magistra's real-world evidence database: what people actually taking these medications talk about, counted rather than anecdotally quoted.

The measure is a reporting frequency: of the 185 distinct community reports in our corpus (as of 24 August 2026), what share mention each effect at all. One report = one source URL — a Reddit thread or a Drugs.com patient review, hosted on the platform itself. A report that mentions three effects counts once in each of those three rows, so the column does not sum to 100%; across all 15 tracked effects the 185 reports carry 394 mentions, so on average a report mentions about two of them.

What this number is not: it is not an incidence rate, and we never present it as one. A mention is not a diagnosis, and the share of *reports that talk about* nausea is a different quantity from the share of *patients who get* nausea. The two tracks are published side by side and never averaged, divided, or "converged" — the clinical rates article has the incidence numbers.

All 15 reporting frequencies (as of 24 August 2026)

Every row is served live at the link beside it (field `reportingFrequency` — mentions, distinct reports, share, and the current platform list). The predictor tool's real-world track serves the same shares with a Wilson 95% interval around each.

What stands out

Abdominal pain out-mentions nausea. Clinical trial literature reports nausea most consistently of any GLP-1 side effect — it tops our clinical evidence table by a wide margin. But in community reports, abdominal pain comes up more often (27.6% vs 24.9%). One plausible reading — an interpretation, not something we have measured — is that trials record nausea, vomiting and diarrhoea as separate standard adverse-event endpoints, while "stomach pain" is a catch-all in how patients describe the same cluster. Either way, the gap between what trials tabulate and what patients lead with is itself a finding.

The most talked-about effect with no clinical number at all is fatigue. 17.8% of community reports mention it — seventh of fifteen — yet our corpus currently holds zero eligible clinical rate points for it: the trial abstracts and registries we collect from rarely state a numeric fatigue rate, so the API falls back to a labelled published-literature range (6–14%) rather than a corpus-derived estimate. Emotional blunting is the same pattern at 8.6% mentions and zero eligible clinical points. Per our methodology, a high reporting frequency alongside a thin clinical base is a *pointer* — a signal that something matters to patients and deserves study — not evidence that the true incidence is high.

Why we count mentions instead of averaging what posts claim

Because averaging what individual posts claim produces confident-looking nonsense, and we know because we did it. Until 14 August 2026 our real-world track averaged self-reported percentages scraped from posts — and one X/Twitter post claiming "76%" hair loss pulled a two-source average to 39%. Averaging what one anecdote claims is not a measurement no matter how the average is weighted, so we replaced the computation with the reporting frequency above and published the correction. A second correction followed on 17 August: 47 of what were then 232 "community reports" turned out to be Google News search results *about* social posts rather than posts on the platform itself; removing them moved the nausea figure from 32.5% to 24.9%. Both corrections are kept publicly in our llms.txt correction history — we publish corrections rather than quietly overwrite them.

Honest limitations

The pool is a dated snapshot, not a live stream. The denominator has stood at 185 distinct reports since 17 August 2026 — Reddit currently blocks automated collection, so the community corpus grows only when a new collection route works. Cite these figures with their date.

Self-selection. People post about experiences that bother them. That skews mentions toward negative and dramatic experiences, in ways we cannot quantify — one more reason this is a reporting frequency, not an incidence estimate.

Two platforms. Today's reports come from Reddit and Drugs.com patient reviews (the live list is the `platforms` field at every API link above). Those communities are not a random sample of GLP-1 patients.

The clinical track — 143 eligible stated rates from 65 distinct sources as of 24 August 2026 — is the other half of the picture: see the pooled rates, or query everything yourself at magistra.health/en/data-api.