Research Tools / Stage 2 / Gemini Notebook

STAGE 2 · DOCUMENT ANALYSIS · HANDS-ON TOOL GUIDE

Gemini Notebook for SLR Document Analysis

Use Gemini Notebook (formerly known as NotebookLM) to ask questions about a selected set of full-text sources, inspect source-linked citations, and prepare a draft evidence extraction. The researcher checks the evidence and records the final decision.

Open Gemini Notebook ↗Return to Stage 2 extraction →Open Stage 2 prompt bank Build a Stage 2 review prompt

Gemini Notebook workflow: add included full-text PDFs, ask focused extraction questions, inspect source-linked citations, verify claims in the PDFs, and record verified evidence in the extraction matrix — Use Notebook as an evidence assistant: verify every source-linked claim in the original PDF before recording it in the extraction matrix.

1 · What this tool is for

Notebook-based chat can answer questions from uploaded sources, let you choose which sources to use, and provide citations linking back to quoted material. Open each citation and check the original text before recording a claim. Google: Use chat in Gemini Notebook ↗

Suitable Stage 2 tasks

Locate evidence in full texts; draft a structured extraction; compare methods, contexts and reported findings across selected included studies.

Keep human decisions

Do not use it as a screening vote, quality score, or authority for inclusion decisions. It supports evidence review; reviewers decide and document.

Suggested boundary: Use Notebook after records have been screened and the papers for analysis are selected. For appraisal, ask it to map evidence to criteria; the reviewer assigns and justifies each judgement.

2 · Login and create a notebook

Open notebook.google.com or Gemini Apps and sign in with your Google account.
If your account shows a Notebooks option, select New notebook. Availability and menu labels may vary by account, plan and rollout.
Name the notebook for a bounded analysis task, for example Stage 2 · Pilot extraction · AI feedback in programming education.
Select Add sources and upload a small pilot set. Supported sources include PDFs, documents, spreadsheets, images, audio, text/Markdown, Google Drive files, websites and copied text. Google: Add or discover notebook sources ↗
Wait for the sources to load. Check their titles and select only the sources relevant to your current question.
Ask a focused question. Review its citations and save a useful response as a note if needed.

Workshop setup: Use a laptop browser for the demonstration. Confirm that Gemini Notebook is available with the account participants plan to use before the session.

3 · Hands-on workflow for Stage 2

Action	Participant does	Evidence to retain
Prepare	Choose two or three included articles for a pilot and note their study IDs.	Study IDs and source files used.
Load	Create a notebook and add the pilot PDFs or practice documents.	Notebook title and source list.
Question	Ask for one field at a time or a compact extraction table aligned with the review questions.	Draft value plus Notebook citation.
Verify	Open the citation, read the surrounding context, and check the PDF page or section.	Verified quotation/paraphrase and page reference.
Record	Transfer only verified content to the extraction matrix and note uncertainties.	Completed pilot rows in the Stage 2 extraction template.
Compare	Ask for similarities and differences across selected sources; inspect each supporting citation.	Evidence-linked synthesis notes, not unsupported generalisations.

Download Stage 2 extraction matrix Open Stage 2 output worksheet Open quality appraisal worksheet

4 · Copyable prompts for Stage 2

The first three prompts cover common document-analysis tasks. The additional prompts below extend the bank across protocol alignment, full-text evidence, study methods, outcomes, limitations, research-question mapping, synthesis and audit. Replace bracketed fields, select only relevant sources, and verify every citation.

Prompt A · Extract evidence from one study

Using only source [STUDY ID / TITLE], prepare a draft extraction for these fields: [FIELD LIST]. For each field, provide (1) the extracted value, (2) a short supporting quotation or close paraphrase, and (3) the citation to the source passage. If the paper does not report a field, write “not reported”. Do not infer missing information or combine this paper with other sources. This is a draft for a human reviewer to verify.

Prompt B · Map evidence to appraisal criteria

Using only the selected source and the checklist below, map source evidence to each checklist item. Checklist: [PASTE CHECKLIST]. For each item, provide the exact supporting or conflicting passage and its Notebook citation. If evidence is absent or unclear, say “not reported” or “unclear”. Do not calculate an overall score and do not make the final appraisal judgement. A human reviewer will decide and record the rationale.

Prompt C · Compare evidence across studies

Compare the selected sources for Research Question [RQ]. Create a table with study ID, context, method, finding relevant to the RQ, and citation to the supporting source passage. Separate recurring observations from differences and contradictions. Do not claim causation or effectiveness unless the source directly supports it. Do not introduce outside information; identify where evidence is absent or incomparable.

Prompt D · Check protocol-to-paper alignment

Using only the selected source and the eligibility criteria below, prepare an evidence table for a human reviewer.

Eligibility criteria: [PASTE APPROVED CRITERIA]
Study ID: [ID]

For each criterion, give the source evidence with a Notebook citation and mark it as “appears met”, “appears not met”, “unclear”, or “not reported”. Do not make the final inclusion decision. Do not create or revise criteria. Flag which criterion needs human review.

Prompt E · Identify the study design and context

From source [STUDY ID / TITLE] only, extract the study design, setting, population or sample, duration, intervention or phenomenon, comparison (if any), data sources and analysis method. Return one row per field with the value, source citation and uncertainty. Use “not reported” when the paper does not state a field. Do not infer a design or context from the journal or title.

Prompt F · Extract outcome measures and reported results

For source [STUDY ID / TITLE], identify each outcome or measure relevant to [RESEARCH QUESTION]. For every result, report the outcome name, measure or instrument, comparison/group if applicable, numerical or qualitative result exactly as reported, and Notebook citation. Do not calculate unreported statistics, infer statistical significance, or describe an association as causal. Mark absent details as “not reported”.

Prompt G · Extract limitations and evidence gaps

Using only the selected source [STUDY ID / TITLE], list (A) limitations explicitly acknowledged by the authors and (B) information relevant to [RESEARCH QUESTION] that appears not to be reported in this source. For each item, provide a Notebook citation. Keep author-stated limitations separate from your observation that information is absent. Do not infer that the authors failed to do something unless the source supports that statement.

Prompt H · Map evidence to a research question

Map the selected source [STUDY ID / TITLE] to this research question: [RQ]. Return: relevance (direct, partial, unclear, or not relevant), the exact finding or information that addresses the RQ, source citation, context in which the result was obtained, and any boundary that limits interpretation. Use only the selected source. Do not stretch the finding beyond its stated context.

Prompt I · Compare methods across studies

Compare the selected sources for these method fields: [FIELDS, e.g. design, sample, setting, duration, intervention, measures]. Create a table with one row per study ID and a citation for each value. Mark missing information as “not reported”. After the table, summarize method similarities and differences that are directly visible in the sources. Do not rank study quality or infer why researchers chose a method.

Prompt J · Find contrasting or contradictory findings

For Research Question [RQ], compare the findings in the selected sources. Separate (1) similar findings, (2) differing findings and (3) findings that cannot be compared because the measures, context or methods differ. For each statement, list study IDs and Notebook citations. Do not resolve contradictions by guessing, average incompatible results, or claim causation beyond the evidence. State when the source set is insufficient.

Prompt K · Build a source-grounded RQ evidence matrix

Using only the selected sources, create an evidence matrix for these research questions: [LIST RQs]. Columns: RQ, study ID, relevant finding, context/method, source citation, and evidence limitation. Include a study only where its source contains relevant evidence. Do not fill gaps with outside knowledge. Identify any unanswered question and avoid presenting absence from this source set as absence from all literature.

Prompt L · Check a verified PRISMA count table

Check the arithmetic and category relationships in the count table below. These counts have been transcribed from our verified review log; do not search for or invent counts. Count table and category definitions: [PASTE VERIFIED COUNTS]. Show each calculation, identify an arithmetic mismatch or undefined category, and state which supplied count needs checking. Keep records and reports distinct. If the information is insufficient, state “cannot verify”.

Prompt M · Audit extraction traceability

Audit the extraction rows below against the selected sources. Check only whether each value appears supported by a source citation and whether the citation is relevant to the stated field. Extraction rows: [PASTE STUDY IDs, FIELDS, VALUES AND CITATIONS]. Return supported, citation needs checking, or not found in selected source for each row. Explain briefly and identify the source passage to inspect. Do not silently rewrite the extraction or introduce facts from outside the sources.

5 · Practise with synthetic documents

These short, fictional reports provide different methods and findings so participants can practise source-grounded extraction and comparison. They are training materials only and must never be cited as real studies.

SYN-01 · Hint-based pilotTwo-group pilot; immediate test and learner feedback.SYN-02 · Conversational assistantOptional use; no significant score difference; interview findings.SYN-03 · Student experienceTwo-site qualitative study; perceptions, no performance measure.

Download all three Markdown files, then add them as sources to a new practice notebook.
Ask Prompt A for SYN-01 and SYN-02 separately. Check the citations and compare what each report does and does not measure.
Ask Prompt C to compare the three sources. Check that the response does not present student perceptions as measured learning gains.
Record a verified pilot row in the Stage 2 extraction matrix. Mark missing information as “not reported”.

Synthetic data: IDs SYN-01, SYN-02 and SYN-03 and all article content are invented for training. Do not add them to a real SLR corpus or reference list.

6 · Privacy, copyright and verification

Use only documents you are permitted to upload and process. Follow institutional policies for licensed full texts, unpublished work and confidential documents.
Google’s current privacy information distinguishes notebook source files from notebook chats: source files are not used directly to train generative AI models, while chats are handled according to the relevant activity settings and privacy notice. Review the current notice and your account type before use.
Notebook citations help locate source passages, but generated summaries may omit context or be wrong. Verify every value and citation in the original source.
Do not paste AI-generated conclusions into a paper without checking them against the reviewed studies and your documented synthesis method.

Google: Gemini Apps Privacy Hub ↗

Output for the Stage 2 worksheet: verified extraction fields, pilot notes, synthesis approach and evidence-linked patterns. The complete record-level evidence belongs in your extraction matrix/review system.

← Full-text screening Quality appraisal Extraction and synthesis →All research tools