Choose one of three approaches. The steps for that option appear below.
Choose an option above. On a phone, swipe the tabs left or right to see all options →
The qualified statistician decides whether and how pooling is defensible; missing inputs block pooling. Elicit does first-pass extraction with citations.
Check inputs, add papers
Extract design, statistics
Assemble the compatibility table
Statistician drafts memo
Reviewer checks trail
Use: You
Have ready: study results, outcome definitions, analysis plan, bounded question and permitted-use limits. Confirm everything is complete and readable; write down the question and limits. Add the supplied study papers to an Elicit project; confirm each paper loaded.
Missing or unreadable inputs are flagged. Medical or other high-stakes implications require the relevant qualified professional; evidence research only. Check every supplied study is present; a missing paper blocks its row in the compatibility table.
Use: Elicit
Run data extraction over the studies for design, outcome definition, effect measure, measurement time, sample-overlap clues and required statistics, with sentence-level citations.
What to paste into the tool
Task: Study results, outcome definitions and analysis plan; Bounded research question and permitted-use limits; Confirmed study set with scope notes; Project containing the supplied studies; Extracted fields per study with sentence-level citations. For each study extract: design, outcome definition, effect measure, measurement time, sample overlap clues and reported statistics. Leave a field empty where the paper does not state it. Keep the received originals and earlier results alongside this step output; do not treat carried items as newly generated or checked.
Use: You
Open the cited sentence behind each extracted field; assemble the compatibility table with proposed groups, missing statistics left empty and source locations attached.
Compare effect measures, outcome timing, design, missing statistics and sample-overlap clues against original methods; groups stay provisional for the statistician.
Use: The qualified statistician
Assess proposed groups for heterogeneity and dependent observations; draft the pooling-readiness memo listing compatible groups and missing statistics.
Use: The named reviewer
Each group has justified outcome and design compatibility; missing inputs block pooling; the evidence trail is confirmed before the brief is treated as complete.
The qualified statistician decides whether and how pooling is defensible; the analyst fills the table by hand from checked citations.
Check inputs, upload studies
Ask per-study questions
Fill the table by hand
Check table, export sheet
Statistician drafts memo
Reviewer checks trail
Use: You
Have ready: study results, outcome definitions, analysis plan, question and limits. Confirm completeness; create a notebook and upload the papers as sources.
This record does not establish that upload is permitted; confirm with the data owner.
Use: Gemini Notebook
Ask, per study, for the design, outcome definition, effect measure, measurement time and reported statistics; answers cite direct quotes from the uploaded sources.
What to paste into the tool
Task: Study results, outcome definitions and analysis plan; Bounded research question and permitted-use limits; Notebook containing the supplied studies; Cited per-study extraction answers. For the named study, state its design, outcome definition, effect measure, measurement time and reported statistics, quoting the passages you relied on. Where the paper does not state a field, say so. Keep the received originals and earlier results alongside this step output; do not treat carried items as newly generated or checked.
Use: You
Open each citation in context, checking dense statistical tables with extra care; fill the table by hand, missing statistics left empty and visible.
Proposed compatibility groups remain provisional for the statistician; check each flagged difference against the papers before sharing.
Use: Gemini Notebook
Ask the notebook to check the hand-filled table against the papers and export it to Google Sheets: table first tab, citations second.
What to paste into the tool
Task: Check my hand-filled compatibility table against the uploaded papers and export it for the statistician. My hand-filled compatibility table with source locations and unresolved comparability notes is pasted below. For each row, check the supplied value against the cited passage in the uploaded sources, list any value that differs or any citation that does not support it, and keep missing statistics empty and visible. Do not rebuild the table from scratch and do not fill gaps by guessing. Then create a Data Table with the same columns - study_id, outcome_definition, effect_measure, design, measurement_time, sample_overlap_group, needed_statistic, supplied_value, proposed_compatibility_group, unresolved_issue - and export it to Google Sheets with the table on the first tab and citations on the second tab. Keep the received originals and earlier results alongside this step output; do not treat carried items as newly produced by this step.
Use: The qualified statistician
Assess proposed groups for heterogeneity and dependent observations; draft the pooling-readiness memo; missing inputs block pooling; unresolved comparability notes stay visible.
Use: The named reviewer
Each flagged difference was checked against the papers before sharing; the check is confirmed before the brief is treated as complete.
The qualified statistician decides whether and how pooling is defensible; you read each paper and build the table by hand.
Check inputs, fix question
Locate studies, surface statistics
Copy passages, build table
Statistician drafts memo
Reviewer checks trail
Use: You
Have ready: study results, outcome definitions, analysis plan, bounded question and permitted-use limits. Confirm everything is complete and readable; write down the question.
Flag missing or unreadable inputs. This artefact is evidence research, not a professional decision.
Use: Consensus
Find each supplied study, recording absences as gaps; use Chat with Full Text to surface design, outcome definition, effect measure, measurement time and reported statistics.
What to paste into the tool
Task: Study results, outcome definitions and analysis plan; Bounded research question and permitted-use limits; Confirmed study set with scope notes; Located papers, with absences recorded. Search for each supplied study by its exact title, then confirm authors, year and DOI against the supplied paper. Open the matching paper page and record its source ID and link. Mark a study NOT LOCATED if no matching record is found; use the supplied PDF for the later manual extraction rather than substituting a different study. Do not infer pooling compatibility at this search step. Retain the received source packet alongside this output. Do not send, publish or change records. Task: Study results, outcome definitions and analysis plan; Bounded research question and permitted-use limits; Confirmed study set with scope notes; Located papers, with absences recorded; Exact passages copied from the papers with locations. Where in this paper are the design, outcome definition, effect measure, measurement time and reported statistics stated? Point to the sections. Keep the received originals and earlier results alongside this step output; do not treat carried items as newly generated or checked.
Use: You
Copy exact passages from the paper text itself, not the chat answer; build the compatibility table by hand, missing statistics empty and visible.
The quoted corpus size does not prove the supplied studies are indexed; absence stays an explicit gap. Proposed compatibility groups remain provisional for the statistician.
Use: The qualified statistician
Assess proposed groups for heterogeneity and dependent observations and draft the pooling-readiness memo listing compatible groups and missing statistics.
Use: The named reviewer
Each group has justified outcome and design compatibility; missing inputs block pooling; the evidence trail is complete before the brief is treated as complete.
Best when: Structured extraction with sentence-level citations to check
What it does: Elicit extracts fields; the statistician decides pooling
Strong at: Each field traces to a cited sentence
Limit: Extraction accuracy on your studies is unchecked
Free plan: Yes, limited (limited usage for Research Agent and Research Reports)
Best when: Extraction tied to exactly the supplied papers
What it does: Gemini Notebook quotes answers; you fill the table
Strong at: Sheet export carries the table and citations
Limit: Dense statistical tables in PDFs are a weak spot
Free plan: Yes, limited (50 sources per notebook)
Best when: Reading each paper yourself with a guide
What it does: Consensus points to sections; you extract by hand
Strong at: Full-text chat points to methods and statistics
Limit: Supplied studies may be absent from its index
Free plan: Yes, limited (10 Pro messages, 3 Deep reviews a month)
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Each group has justified outcome and design compatibility.
Missing inputs block pooling.
Proposed compatibility groups remain provisional for the qualified statistician.
Source locations are retained for every supplied value.
Check the result against your original inputs. Approve it before sharing or acting.