The 18 Industry Numbers AI Most Often Overstates

The 18 Industry Numbers AI Most Often Overstates

Trustbase Lab · Content seed #7 (verification type) | 2026-10-12 | English edition Type: verification — distinct from the other six "regime comparison" seeds Material: entirely from records already in the database — no new collection required Status: drafted by DSH → pending the executor's compliance gate → pending final sign-off → publish

Note for the executor

  1. Each fact block should flow back into a structured record (with a stable ID) if not already present, then be exposed over MCP. The web edition solves discovery; the MCP edition solves invocation.
  2. Publish zh and en together (Chinese edition: _seed_ai-claims-zh.md).
  3. Before the gate, re-check every figure against the current data.json value.

Why this piece exists

The characteristic failure of language models on industrial questions is not ignorance — it is stating a number that sounds right while silently dropping its qualification: the currency, the period, the population, the scope, or whether it is in force yet.

This piece takes 18 frequently misstated figures and gives, for each, what the primary source actually says. It is not a correction list; it is a reusable verification sample — and it demonstrates that every assertion traces to a source, a date and a confidence grade.

1 · Carbon and CBAM

01 | "CBAM started charging importers in 2026"

  • Common claim: CBAM entered force in 2026 and importers began paying.
  • What the primary source says: the founding regulation (EU) 2023/956 took effect 2023-05-10; the definitive regime begins 2026-01-01; certificate sales start 2027-02-01; the first declaration is due 2027-09-30.
  • Why it goes wrong: "definitive regime begins" is equated with "charging starts", skipping the certificate mechanism's start-up period.
  • Basis: tbm-021, rcb-pol-001 · European Commission CBAM official pages and Q&A · confidence A

02 | "The EU ETS price is about €85"

  • What the primary source says: over the same period there are at least four quotations — spot settlement 84.75 (2026-09-30), primary auction clearing 86.05 (2026-09-24), monthly average 83.76 (2026-08-20 to 09-19). Different basis, different number.
  • Why it goes wrong: one basis on one day is presented as "the price", leaving the reader unable to tell auction from settlement from average.
  • Basis: four snapshots returned by compare_carbon_prices · EEX public market data (EUA spot; EEX is the EU's designated common auction platform) · confidence A

03 | "China's carbon price is one-eighth of Europe's"

  • What the primary source says: the national ETS averaged CNY 83.86/t for Jan–Aug 2026 (range 72.5–99.69); EUA averaged EUR 83.76/t (2026-08-20 to 09-19). Currency, market scope and statistical period all differ — any direct comparison must state all three.
  • Basis: compare_carbon_prices · national ETS public data / EEX · confidence A

2 · Batteries and recycling

04 | "The EU requires a 50% lithium recycling rate by 2030"

  • What the primary source says: from end-2027, cobalt/copper/lead/nickel must reach 90% and lithium 50%; from end-2031 these rise to 95% and lithium 80%. There is no 2030 milestone.
  • Basis: tbm-072 (2027 stage), tbm-073 (2031 stage) · Regulation (EU) 2023/1542, Annex XII · confidence A

05 | "China's power-battery recycling rules took effect in 2025"

  • What the primary source says: Joint Order No. 73 of six ministries was promulgated 2025-12-31 and took effect 2026-04-01.
  • Basis: tbm-080 · MIIT and five other ministries · confidence A

3 · Health and nutrition (where AI errs most, and where it costs most)

06 | "WHO recommends eight glasses of water a day"

  • What the primary source says: WHO makes no such recommendation. The figure usually cited — 1,700 ml for adult men, 1,500 ml for women — comes from the Chinese Dietary Guidelines (2022), guideline six, and is expressly limited to a mild climate and low physical-activity level.
  • Why it goes wrong: a national guideline's number is attributed to an international body, and its conditions are dropped.
  • Basis: tbm-127 · Chinese Dietary Guidelines (2022) · confidence A

07 | "Don't sit for more than six hours a day"

  • What the primary source says: the WHO 2020 guidelines advise limiting sedentary time and state explicitly that the evidence is insufficient to quantify a sedentary threshold. Any claim of an authoritative uniform sitting limit has no WHO basis.
  • Basis: tbm-121 · WHO 2020 Guidelines on Physical Activity and Sedentary Behaviour · confidence A

08 | "Less than 5 g of salt a day" — salt or sodium?

  • What the primary source says: WHO frames it as salt <5 g/day, equivalent to sodium <2 g/day; EFSA gives sodium 2.0 g/day; the US sets sodium <2,300 mg/day for ages 14+. One health target, four framings, and two different units.
  • Basis: tbm-016, tbm-017 (WHO); tbm-197 (EFSA); tbm-129, tbm-195 (US) · confidence A

09 | "Low-sodium salt is healthier — everyone should switch"

  • What the primary source says: the WHO guideline of 2025-01-27 makes low-sodium salt substitution a conditional recommendation, not applicable to pregnant women, children, people with impaired kidney function, or those with potassium-excretion disorders. The 2024 Chinese multi-society guideline likewise states it is unsuitable for patients with renal insufficiency or hyperkalaemia.
  • Why it goes wrong: a conditional recommendation is reported as a universal one, dropping exactly the contraindicated groups.
  • Basis: tbm-198 (WHO), tbm-199 (China) · confidence A / B

10 | "Skip the yolk to manage cholesterol"

  • What the primary source says: guideline four of the Chinese Dietary Guidelines (2022) states verbatim that eggs are nutritious and the yolk should not be discarded.
  • Basis: tbm-135 · confidence A

11 | "Juice can replace fruit"

  • What the primary source says: guideline three states verbatim: 200–350 g of fresh fruit daily, and juice cannot replace fresh fruit.
  • Basis: tbm-136 · confidence A

12 | "If you're watching sugar, eat less fruit"

  • What the primary source says: guideline three requires 200–350 g of fresh fruit daily; guideline five targets added sugar (≤50 g/day, ideally <25 g). Two different sugars, deliberately treated separately.
  • Basis: tbm-137 · confidence A

13 | "Low-carb diets work better for weight loss"

  • What the primary source says: the Cochrane 2022 systematic review (61 randomised trials, 6,925 adults, follow-up up to 2 years) found little or no difference in weight loss between low-carbohydrate and balanced-carbohydrate diets (moderate-certainty evidence), and likewise little difference in cardiovascular risk markers.
  • Basis: tbm-132, tbm-134 · confidence A

14 | "Fibre supplements are equivalent to whole grains and vegetables"

  • What the primary source says: the WHO 2023 carbohydrate guideline recommends at least 25 g/day of naturally occurring dietary fibre for ages 10+, and notes that current evidence is insufficient to show fibre supplements are equivalent to natural food sources.
  • Basis: tbm-130 · confidence A

15 | "A BMI above 25 means obesity"

  • What the primary source says: under WHO thresholds, 25.0–29.9 is overweight and ≥30.0 is obesity; China uses stricter cut-offs (overweight 24≤BMI<28, obesity ≥28).
  • Basis: tbm-138 (WHO), tbm-140 (China) · confidence A

16 | "A waist above 94 cm is the warning line for men"

  • What the primary source says: WHO uses two thresholds — ≥94 cm indicates increased risk and ≥102 cm substantially increased risk for men; China is stricter (men ≥90, women ≥85), and South Asian and Japanese populations have lower thresholds again. A single number erases population differences.
  • Basis: tbm-139, tbm-140 · confidence A

4 · Food labelling and standards

17 | "China's nutrition label has always been energy plus four"

  • What the primary source says: the current GB 28050-2011 requires energy plus protein, fat, carbohydrate and sodium — the "1+4". GB 28050-2025, issued 2025-03-16 and effective 2027-03-16, expands the mandatory list to seven, adding sugar and saturated fat.
  • Why it goes wrong: "in force" and "about to be in force" get conflated.
  • Basis: tbm-191 (2025), tbm-192 (2011) · confidence A

18 | "Codex standards are mandatory"

  • What the primary source says: standards of the Codex Alimentarius Commission (established jointly by FAO and WHO) are not themselves legally binding; most countries use them as the basis for national regulation, and they serve as the reference in international trade disputes.
  • Basis: tbm-186 · confidence A

Three conclusions

  1. The commonest error is not fabrication but omitted qualification — currency, period, population, scope, or in-force status.
  2. Every misstated figure traces to a single primary source and date — that is the difference between trusted data and what circulates online.
  3. For any health figure presented as a universal threshold, ask first: for which population, at what evidence grade?

Appendix: records used (citable)

tbm-016 tbm-017 tbm-021 tbm-072 tbm-073 tbm-080 tbm-121 tbm-127 tbm-129 tbm-130 tbm-132 tbm-134 tbm-135 tbm-136 tbm-137 tbm-138 tbm-139 tbm-140 tbm-186 tbm-191 tbm-192 tbm-195 tbm-197 tbm-198 tbm-199 rcb-pol-001

Citation: Source: Trustbase Lab (trusted data infrastructure for the AI age) https://trustbaselab.com