Google Antigravity · NHANES
What Google Antigravity’s demo teaches about checking an AI-assisted NHANES draft
Google showed a roughly one-hour cheminformatics workflow. An NHANES example shows where to start timing a manuscript draft and which results to check.
On November 3, Google DeepMind showed Antigravity tracing compounds and literature, running matched molecular pair analysis and proposing candidates for follow-up. The reported hour applies to that workflow. An NHANES project can borrow the discipline of recording sources, calculations and the evidence behind each conclusion.
Trace the reported results
The table uses a public RigorMed research demonstration. The frozen analysis included 4,816 adults taking antihypertensive medications. The interaction interval crosses 1, so these data leave the between-group change uncertain. Before drafting, verify the survey cycles, variable definitions and weights. Check that the tables, figures, prose and references all point to the same result.
| Check | Example result |
|---|---|
| Period interaction | OR 0.91 · 95% CI 0.64–1.30 · p=0.597 |
| Difference in standardized prevalence changes | −2.24 percentage points · 95% CI −10.72–6.24 |
Three checks for an NHANES draft
Trace the sample count to eligibility
Open the linked analysis and compare cycle, medication use, cancer history and BP outcomes. Verify that 4,816 is the analyzed count, distinct from weighted population estimates. Record data files, variable definitions and exclusions.
Trace the odds ratio to its model and scale
Match the period-by-cancer-history interaction OR of 0.91 and 95% CI of 0.64–1.30 to the table. The null is 1. Separately check the difference in standardized prevalence changes of −2.24 percentage points, whose interval crosses zero. These are different scales and neither establishes equal changes.
Trace every conclusion to material
Map each abstract result to its table and model, each background claim to a primary source and passage, and each discussion implication to its applicable population. A reader should be able to trace a claim back to survey cycle, weighting and source.
Try going from research question to manuscript draft in 20 minutes
Work through the research question, literature search, public data retrieval and analysis, figures and first draft in one project. The timing below covers the drafting stage.
The drafting stage took 10 minutes and 17 seconds in our test. The study used two NHANES cycles to examine food insecurity and hypertension care, with BMC Public Health as the target journal. Timing ran from submission of independently checked aggregate results until the English draft and forest plot were both downloadable. Topic selection, data retrieval, analysis and author review took place outside that measurement.
What should a timing record include?
What should a timing record include?
Log time separately for question approval, retrieval, analysis, file-ready drafting and author review. The 10 minutes and 17 seconds reported here starts from completed analysis and covers delivery of the draft and forest plot.
Verified sources
The official sources below were checked on 2026-11-10. Access conditions and update dates are rechecked before publication.
Public-data research
Explore the NHANES example
Open the complete example to inspect its sample, weighting and results. For your own question, check whether the population, exposure and outcome are available in the same survey cycle.
