How I Check Whether Research Papers Actually Agree Using Elicit
A citation is not a consensus. Elicit's extraction tables turned my literature review from vibes about what studies found into a sortable table of what they measured.
Nothing slows a literature review like the gap between citation and consensus. Ten papers cited as supporting a claim often contain three agreements, five partial ones, and two studies using methods that make them barely comparable. Elicit is the tool I now use to find that out before I write the sentence, and the extraction table is the mechanism.
Start with a real research question, not a topic. Elicit works best when phrased the way a reviewer would ask it: what is the effect of remote work on measured productivity in controlled studies, not remote work productivity. Run the search, screen the results by title and abstract with Elicit's relevance columns, and be ruthless: forty relevant papers you will never read are worth less than twelve you will. The screening columns are where the hours get saved.
Then the step that makes Elicit different from a smarter search engine: extraction. Add columns for the things that decide whether studies agree, sample size, study design, outcome measure, effect direction, and population studied. Run extraction across your screened set and you get a sortable table of what each paper actually did and found. In my remote work review, the table exposed immediately that the disagreement in the literature was mostly a disagreement in outcome measures, self-report versus logged output, which no amount of abstract-reading had surfaced.
Sort and slice the table before believing any summary. Filter by study design and watch the claim shrink or hold, because effects that survive in randomized designs and evaporate in cross-sectional surveys tell you the honest state of the field. Export the table and keep it as your evidence matrix, the document reviewers actually want, and update it as Elicit finds newer work with the same extraction columns.
The discipline that keeps this honest: spot-check the extraction. I pull the actual PDF for any finding I cite directly, two minutes each, and I have caught enough mischaracterizations, mine and the model's, to know this step is not optional. Elicit is the research assistant who reads faster than you and never gets bored of tabulating; you are still the one who decides what the table means.