Verdict № 004 · Sep 5, 2026 · updated Sep 8, 2026

Antidepressants, Dissected.

AI-researched · adversarially reviewed · human-checked · corrections public
BEFORE YOU READ — HOW DO YOU RULE?
The Verdict For severe depression, antidepressants genuinely help and can be part of saving a life — the effect is real, replicated, and grows with severity. For mild, everyday misery, the average benefit over placebo is small enough that researchers argue about whether patients can feel it — yet that is exactly where much of the prescribing happens. The drugs were marketed for decades on a "chemical imbalance" story the evidence never supported; withdrawal is real and was long downplayed; and stopping should be planned with a clinician, never done abruptly.
Conditional
MODERATE EVIDENCE
§ What it actually does

Where the evidence says yes

§ What it doesn't do

Where the marketing outruns the data

§ The Money Map

Who profits, stage by stage

The golden ageSSRIs were pharma's biggest franchise for two decades. The "chemical imbalance" narrative was a marketing invention that outlived its evidence — it made a prescription feel like insulin for the brain.
The evidence machineIndustry-funded trials, ghost-written papers, and selective publication (documented in the Turner NEJM analysis) inflated apparent efficacy for years.
Today's pipelinePatents expired; generics are cheap. The new money is in telehealth subscription prescribing (questionnaire → refill), and in patented successors (esketamine et al.) priced like the old days.
The other side's moneyAudit the critics too: "get off your meds" content sells courses, supplements, and books. Withdrawal-scare monetization is real, and it kills people who quit abruptly.
This pageNo pharma, no telehealth, no "natural mood" supplements. See "How we make money."
§ The other side

The strongest case against this verdict

The strongest case against our verdict\'s skeptical half: averages hide responders. A drug with a modest average effect can still transform the meaningful minority who respond strongly, and clinicians cannot identify them in advance — so restricting access to "severe only" would deny real help to some mild-moderate patients. Small average effects on symptom scales can also understate functional benefit (getting to work, staying alive). And placebo response in trials is partly the ritual of care — the real-world comparison isn\'t drug vs elaborate placebo ceremony, it\'s drug vs a six-minute GP visit and nothing.
§ Symmetric harms

Harms of using it — and of avoiding it

If you use it

  • Sexual dysfunction is common, dose-related, and under-disclosed; a subset report it persisting after stopping (PSSD) — acknowledged by EU regulators.
  • Withdrawal/discontinuation symptoms are real and can be severe and prolonged after long-term use; tapering needs to be slow and planned.
  • Emotional blunting reported by a substantial minority.
  • In under-25s: an activation/suicidality signal early in treatment (the FDA black-box warning) — needs close monitoring, not casual telehealth refills.
  • Elderly: falls, hyponatremia, bleeding interactions.

If you avoid it

  • Untreated severe depression is lethal: suicide, and a mortality/morbidity burden across every organ system. This is the strongest harm-of-avoiding on this page.
  • Refusing continuation after response roughly doubles relapse risk.
  • Quitting abruptly because of something you read — including this page — is the worst of all options: withdrawal + relapse together.
§ The Minority Report

Who gets hurt when it goes right for everyone else

Both tails get hurt: people medicated who barely benefit, and people scared off medication who badly needed it.

How to tell if you might be in this group: Ask: how severe is my depression, formally scored? Am I under 25 (monitoring plan)? Any bipolar screening? What is the taper plan for the eventual exit, and what functional change will tell us in 8 weeks whether this is working?
§ The alternatives ledger

Everything else, judged by the same standards

strong Psychotherapy (CBT and related)
Comparable to drugs for mild-moderate depression, with benefits that persist after treatment ends — the durability drugs lack. Access and cost are the honest obstacles.
moderate Exercise
Likely a genuine antidepressant for mild-moderate cases in meta-analyses — unprofitable, so nobody markets it to you.
strong Lithium, ECT, ketamine — for severe/refractory illness
The heavy artillery has strong evidence exactly where SSRIs run out: lithium prevents suicide; ECT works in severe depression; ketamine acts in hours (and is being commercialized accordingly).
moderate St. John's Wort
May work for mild depression (German trials) — and interacts dangerously with many drugs, including birth control and SSRIs themselves. "Natural" is not "inert."
§ Take it to a human

Questions for your doctor

§ Sources, COI-flagged

Every claim, checkable

  1. meta Cipriani A, et al. (2018). Comparative efficacy and acceptability of 21 antidepressant drugs: systematic review and network meta-analysis. Lancet. Academic (Oxford); some authors disclose industry honoraria. 522 trials: drugs beat placebo; effect sizes mostly modest.
  2. meta Kirsch I, et al. (2008). Initial Severity and Antidepressant Benefits: FDA data meta-analysis. PLoS Medicine. Academic, built on FOIA-obtained FDA data (including unpublished trials): benefit concentrated in severe depression.
  3. meta Turner EH, et al. (2008). Selective Publication of Antidepressant Trials and Its Influence on Apparent Efficacy. NEJM. Academic. The publication-bias receipt: ~1/3 of FDA-registered trials unpublished or spun positive.
  4. meta Fournier JC, et al. (2010). Antidepressant Drug Effects and Depression Severity. JAMA. Academic patient-level meta-analysis: the severity-response gradient in its clearest form.
  5. meta Moncrieff J, et al. (2022). The serotonin theory of depression: a systematic umbrella review. Molecular Psychiatry. Academic; lead author is a prominent critic of psychiatric drug models (an intellectual-position COI worth flagging in both directions). Found no consistent support for the low-serotonin story.
  6. observational Rush AJ, et al. (2006). Acute and Longer-Term Outcomes in Depressed Outpatients Requiring One or Several Treatment Steps (STAR*D). Am J Psychiatry. NIMH-funded real-world sequence; later re-analyses argue original reporting overstated cumulative remission — flagged both ways.
  7. meta Davies J, Read J (2019). A systematic review into the incidence, severity and duration of antidepressant withdrawal effects. Addictive Behaviors. Academic; authors are known critics (position COI). Withdrawal more common/severe than official lines long claimed; methodology debated — held to "real and under-acknowledged," no further.
§ What we don't know

The honest uncertainty list

§ Corrections

Updates & corrections — public, dated, proud

No corrections yet. Found an error? Challenge it — substantiated challenges are corrected within 48 hours and logged here permanently. Submit a challenge →

§ We dissect ourselves too

How WE make money

This pageNo pharmaceutical advertising, no telehealth referral fees, no supplement affiliates. Health.AI earns from optional Pro subscriptions. If that changes, this box changes first.

Demand the next dissection

Any disease, drug, supplement, or practice. The queue decides what we dissect next.

If you're struggling right now: this page is about evidence, not about you — and no verdict is a reason to face a crisis alone. India: iCall 9152987821 · Tele-MANAS 14416. US: call or text 988. Elsewhere: your local emergency number. If you're on medication and want to stop, do it with a clinician — never abruptly.

How Dissected works

What is Dissected?

A free treatment auditor on health.ai. Type any drug, supplement, therapy, or procedure and get an honest verdict: what the evidence shows, what it doesn't, who gets hurt even when it works — and who profits at every stage. Pharmaceutical and natural treatments are held to the same standards: our homeopathy verdict and our stent verdict are equally blunt.

How to use it

  • Search any treatment from the top bar. If we've dissected it, you'll land on its verdict.
  • No verdict yet? Hit “Demand it.” The queue is public and demand decides what we dissect next — leave an email and you'll get the verdict the moment it publishes.
  • Rule first. Verdict pages ask for your call — Works / Overhyped / Doesn't work — before the reveal. It's optional (“Just show me” skips it), but it's the honest way to notice what the marketing has already done to you.
  • Play the daily game. Placebo or Not? — ten treatments, three buttons, one score. Same deck for everyone each day; share your 🟩🟥 grid and challenge someone.

How to read a verdict

  • The badge (Works, with catches Conditional Placebo) — our one-line finding.
  • The evidence meter — how solid the science behind the verdict is (strong / moderate / weak / unknown). A “Placebo” verdict can have strong evidence: we're very sure it doesn't work.
  • Tier chips on each claim — where it comes from: meta-analysis > trial > observational > opinion. Higher tier, harder to argue with.
  • 💰 The Money Map — who profits at every stage, including who funded the studies we cite. The wellness seller's margin gets audited alongside pharma's.
  • The Minority Report — who gets hurt even when the treatment works for most people, and how to tell if you might be in that group. We never let the 5% disappear into the 95%.
  • The other side — the strongest honest case against our own verdict, written to win. If we can't survive our own counterargument, we shouldn't publish.
  • Sources — every claim links to its study, and every study carries a 💰 COI flag telling you who paid for it.

Why trust it?

  • AI-researched, human-checked. Five adversarial research passes, then a person reads every verdict before it publishes.
  • No ads. No affiliates. Ever. We sell nothing on any verdict page, and every page discloses how we make money (optional Pro subscriptions to health.ai's AI tools).
  • Corrections are public. Substantiated errors get fixed within 48 hours and logged permanently on the page. Found one? Use the challenge link on any verdict.
  • Evidence-weighted, never popularity-weighted. Reader votes are compared against verdicts, never used to change them.

What Dissected is not

Not medical advice. Verdicts describe evidence about populations — they can't know your history, your medications, or your situation. Never start or stop a treatment because of this site: bring the “questions for your doctor” from any verdict to a qualified clinician instead.