ChatGPT ties ten-plus years of symptoms to an MTHFR mutation

ChatGPT (OpenAI consumer app), given five uploaded Function Health lab PDFs and a written symptom timeline

o3, according to the author.

Exact release not established.

OpenAI

ChatGPT interpreted laboratory reports alongside the user’s symptom history.

Not assessed

Not assessed

Not assessed

The user already held a genetic result (homozygous A1298C MTHFR from a Function Health panel), five PDFs of laboratory results and a written symptom timeline. They gave all of it to the model rather than a description of symptoms alone. A paid reasoning model (o3, by the user's account) was asked to reconcile B12 values in the laboratory's normal range with persistent neurological symptoms. It proposed a mechanism and a supplement rather than a bare label. A treating doctor willing to review the model's reasoning and the gene report and to approve a methylated B-complex before it was started.

A doctor reviewed the interpretation and confirmed the supplement.

The mutation was not found by ChatGPT. The user wrote that a Function Health panel showed the homozygous A1298C MTHFR variant. The AppSo account says the same (“在功能医学检查中发现”). What ChatGPT (o3, per the user) added was the inference. Given about five paragraphs of history and five PDFs of results, it concluded the symptoms were “on par with the mutation”. It concluded that B12 values that looked normal on paper might not be utilized, so supplementation was warranted. The doctor confirmed the reading and the supplement. The user then improved. The counterfactual is unproven. The same results were available to the doctor. The user had earlier been told the symptoms were anxiety and given an SSRI and Xanax. Whether the improvement came from the supplement, from regression, or from expectation is unknown. Livemint (“The AI flagged a possible genetic mutation”) and the Monitor's summary (“identification of genetic mutations missed by doctors”) overstate the model's role. The ACMG guideline and the ifeng piece also put the MTHFR premise itself in doubt.

ChatGPT interpreted laboratory results that already showed an MTHFR variant and suggested that vitamin B12 supplementation might help. The user reported that a doctor agreed and that symptoms improved afterward.

Not assessed

A Reddit user says ChatGPT tied 10+ years of symptoms to an MTHFR A1298C result and B12 use; doctor agreed; symptoms largely resolved.

Improved Physical Health

Not assessed

People.

Not assessed

Not assessed

Not assessed

Not assessed

Yes

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

The account does not establish that the supplement caused the improvement or that ChatGPT found a mutation that clinicians had missed. The cited ACMG guidance also questions the clinical significance of the MTHFR premise.

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Developer: OpenAI. Operator: OpenAI.

Developer: United States. Operator: United States.

Not assessed

Not assessed

Not assessed

Not assessed

A clinician reviews the model's inference and approves any supplement or treatment before it is started. Here the doctor did, by the user's account. The genetic result comes from an actual test. The model interprets; it does not genotype. The ACMG guideline says MTHFR polymorphism testing has minimal clinical utility, so the premise itself is contested. The benefit is self-reported and pseudonymous. Any record of this kind must carry the absence of a clinician statement and the published medical doubt (DXY via ifeng) as a standing caveat.

Not assessed

8 July 2025. Four days after the post, DXY reported that a doctor dismissed the case as of doubtful authenticity and probably functional-medicine promotion, cited the ACMG guideline against routine MTHFR testing, and noted that the post now showed as deleted by its author; this is a published dispute and deletion, not a documented reversal of the reported recovery.

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

AIBD-0141-source-0001–0009

Numbered references below.

Not assessed

Beneficiary, Independent

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

People.

People: 2.

People: By the user's own account, ten-plus years of near-constant tingling, tinnitus and chronic fatigue were “largely resolved” within months of a doctor-approved methyl B complex. Corroborated, that would rate one level higher. It is pseudonymous and self-reported. No clinician has spoken publicly. A DXY-sourced ifeng piece reports doctors doubting its authenticity.

Not assessed

Not assessed

Not assessed

Not assessed

Independently studied

Not assessed

Not assessed

Consumer health (self-directed interpretation of laboratory and genetic results). Consumer.

ChatGPT is OpenAI's general-purpose assistant, built to answer questions over typed text and uploaded documents. Reading lab PDFs and a symptom history and proposing an explanation is an instance of that general use, not a designed diagnostic function. So the benefit falls within what the system was built to do.

Date basis unassessed: 2025 (precision unassessed). Updated: 21 October 2026.

Not assessed

SHA-256: 93eb1e0326c6…

oecd_aim: 2025-07-06-2ff8.

Not assessed

Not assessed

6 revisions. Latest: 21 October 2026.