Real example · Study, summaries & research
Used NotebookLM to extract data for a systematic review
What they did
A team of gastroenterology researchers loaded 57 articles into NotebookLM notebooks, used structured prompts to fill evidence tables and draft 70 candidate consensus statements, then had two reviewers check every output against the full texts.
How it went
Of 1,710 extracted data cells, about 9% were flagged and 4 cells (0.23%) had major factual errors. The workflow took 17.7 person-hours against a projected 165.1 for conventional extraction.
Worth knowing
A preprint that has not been peer reviewed. The comparison time is a projection, not a measurement, the main weakness was incomplete extraction, and human reviewers were essential to catching errors.
Try it yourself with Gemini Notebook (formerly NotebookLM)
Replace anything in [brackets] with your details.
From the attached paper, fill this table: [columns, e.g. study design, population, sample size, intervention, main outcome, effect size, follow-up].
Write 'not reported' when the paper does not say. For every cell, give the page or section where you found the value so a reviewer can check it. At the end, list any cells where you are unsure.Read the original
Performance of Google NotebookLM for AI-assisted data extraction and consensus statement generation in a heterogenous systematic review on inflammatory bowel disease, obesity, and cardiometabolic comorbidities: A Methodological Report ↗
pmc.ncbi.nlm.nih.gov · · Peer-reviewed study
Attributed to the source linked below; not our own test.
Share Share on X Share on WhatsApp
