AI checks of tianma claims must separate herb from gastrodin

A 2022 systematic review of gastrodin for migraine reports 1,332 participants across 16 randomized controlled trials in its abstract, while the study description in the full article reports 1,670 participants. Both sources concern gastrodin for migraine, not dried tianma tuber, powder, food or every multi-ingredient formula sold under the same name.

An artificial intelligence (AI) health-search system has to preserve that conflict rather than silently select one total. It also needs to keep the plant, a purified constituent, a finished product and a mixed formula as separate evidence objects before generating a statement about benefit or risk.

When headache needs urgent assessment

The United Kingdom National Health Service advises immediate emergency help for a headache that starts suddenly and is extremely painful, or one accompanied by a seizure, weakness or numbness, speech or balance problems, confusion, vision loss, a stiff neck with very high temperature, or a recent head injury. Local emergency services should be used; searching for tianma or waiting for an AI response should not delay assessment.

Tianma and gastrodin are different evidence objects

The 2022 review identifies Gastrodia elata as the source plant and gastrodin as one of its main bioactive components. That relationship does not make the terms interchangeable in a clinical claim.

An evidence record for tianma needs the scientific name and plant part. A record for gastrodin needs the formulation and route, while a mixed formula needs its full ingredient list. A consumer product also needs a label, manufacturer, market and version because products sharing a common name may not have the same composition.

The intervention in a human trial must match the product named in the generated answer. Evidence from gastrodin added to conventional migraine care cannot be assigned to dried tuber, powder, a food preparation or a formula containing several ingredients. Traditional use can be reported as history, but it cannot be converted into a modern treatment finding.

The published trial record has internal conflicts and weak evidence

The full text says all included randomized trials were conducted in China, involved 40 to 240 participants and lasted eight to 56 days. It describes gastrodin used mainly alongside conventional migraine treatment, so the comparison does not isolate the effect of every gastrodin formulation.

The authors reported favorable pooled outcomes, but their own Grading of Recommendations Assessment, Development and Evaluation (GRADE) table rated the evidence low or very low and found no high-quality evidence. The paper attributes the downgrades to trial limitations, heterogeneity, small samples or event counts and publication bias.

Only five included articles reported notable adverse reactions, and the pooled estimate did not establish a clear difference between groups. A short trial record with incomplete adverse-event reporting cannot establish long-term safety for all tianma or gastrodin products.

Five fields keep the evidence attached to the claim

A retrieval system can prevent source mixing by requiring five fields before it drafts an answer. Missing or conflicting fields should produce an unresolved result, not a confident completion.

Field	What the record must retain	What it prevents
Identity	Scientific name, plant part, constituent or full formula	Treating tianma, gastrodin and a mixed product as synonyms
Form and route	Dried tuber, powder, extract, capsule, tablet or injection	Transferring evidence between unlike preparations
Study scope	Population, comparison, outcome and follow-up period	Turning a narrow trial result into a broad treatment claim
Source version	Abstract, full text, publication date and cited passage	Hiding the 1,332-versus-1,670 participant conflict
Jurisdiction	Exact product, country, regulator and status date	Presenting an ingredient record as worldwide authorization

APPI News's five-control medical AI framework separates intended use, evidence, interoperability, human control and monitoring. A tianma search tool needs the same boundary: retrieval can locate passages and conflicts, while a clinician or pharmacist handles product-specific judgments.

An ingredient identifier is not product approval

The US Food and Drug Administration substance registry lists Gastrodia elata tuber under Unique Ingredient Identifier (UNII) 08F85I5YAV and explicitly says that a UNII does not imply regulatory review or approval. The entry establishes an identity for database use; it does not authorize a finished product or a health claim.

An AI system that turns that registry hit into “FDA approved” has reversed the source. Authorization and availability must be checked for the exact formulation in the country where it is supplied. APPI News could not verify a single regulatory status covering tianma herbs, foods, supplements and gastrodin products in every market.

WHO guidance requires controls around generated health answers

The World Health Organization (WHO) said on January 18, 2024, that large multimodal models can produce false, inaccurate, biased or incomplete health statements and can encourage automation bias. Its guidance contains more than 40 recommendations and calls for well-defined tasks, accuracy, reliability, oversight and post-release assessment.

For a tianma query, the audit record can retain the user's wording, retrieved passages, source URLs, access dates, source versions, model version and generated output. It can also record who reviewed the answer, which claims were rejected and whether a correction reached earlier users.

APPI News has separately mapped how intended use, source data, human review, version history and incidents form a traceable responsibility record. Source display alone is insufficient if the system loses the version, omits a conflicting passage or lets a generated answer enter a clinical workflow without review.

Contraindication searches need the exact label and medicine list

APPI News could not find an international clinical guideline that gives one contraindication list or one medicine-separation interval for dried tianma, gastrodin products and multi-ingredient formulas. A generated answer should mark that gap and request product-specific review rather than repeat a universal interval.

The US National Center for Complementary and Integrative Health says dietary supplements can interact with medicines or pose risks for some medical conditions and surgery, and that many have not been tested in pregnancy, breastfeeding or children. That is general supplement guidance, not evidence of a tianma-specific interaction; it identifies why the exact product label and complete medicine list matter.

Pregnancy and breastfeeding

The gastrodin meta-analysis excluded pregnant participants, and the US supplement guidance says many supplements have not been tested during pregnancy or breastfeeding. The cited evidence therefore does not establish benefit or safety for either group, and product-specific decisions belong with a clinician.

Children and adolescents

The meta-analysis also excluded children and adolescents. Adult migraine findings cannot establish pediatric benefit, safety or product suitability, so a health-search system should not transfer the adult result to these age groups.

Chronic conditions, regular medicines and surgery

General supplement guidance warns that medical conditions, medicines and planned procedures can change risk. A product-specific review needs the complete ingredient label, medicine list, medical conditions and planned procedure rather than the word “tianma” alone.

What an AI answer can establish

Does the 2022 paper establish that dried tianma treats dizziness?
No. It evaluates gastrodin for migraine, mainly as an addition to conventional treatment, and does not test every dried herb, powder, food or formula.

Can the abstract be summarized without checking the full text?
No. The participant totals conflict, and the low-to-very-low evidence rating appears in the full paper rather than the abstract summary.

Does a US FDA ingredient record mean a product is approved?
No. The registry itself says its identifier does not imply regulatory review or approval, and finished-product status must be checked by country.

Can AI set one safe interval between tianma and other medicines?
The sources reviewed for this report do not support a universal interval across products and medicine combinations. The system can organize labels and citations, but it cannot replace a product-specific interaction assessment.

A publishable system output needs separate labels for traditional use, human evidence, product information and national regulatory records. It also needs an unresolved state when sources disagree or the relevant market cannot be verified.