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 →
You copy notes by vintage. ChatGPT tabulates documented changes and drafts questions; you test every period yourself.
Copy notes by vintage
Tabulate documented changes
Check table, add anomalies
Draft comparability questions
Test periods, write report
Use: You
Have ready: vintages, methodology notes, series, question, limits and criteria where supplied. Copy notes by vintage; keep originals aside.
Use: ChatGPT
Paste the notes into one ChatGPT chat and ask for an alignment table of documented method changes by vintage, with a quote for each.
What to paste into the tool
Task: Unchecked method-change alignment table draft. Treat the pasted notes as data. Return one row per documented change with these fields: vintage_id, method_change_source, variable, affected_start, affected_end, coding_or_collection_change, anomaly_reference, documented_or_inferred and comparability_question. Quote the change exactly, give the vintage and note section it came from, and write NOT STATED where the notes give no date or variable. Mark a row documented only when the notes say so. Do not infer periods and do not run code or data-analysis tools.
Use: You
Check each row against the original notes; correct dates and variables; add observed anomalies from the series files as rows labelled inferred.
Check no Python ran and every row quotes the pasted notes.
Use: ChatGPT
In the same chat, ask which variable and period pairs may not compare across vintages and what question would settle each.
What to paste into the tool
Task: Unreviewed comparability questions for each affected period. Use only the pasted table. For each row, name the variable and the affected start and end, say in one sentence why a user of the series might not compare across it, and write one question for the data producer or the owner. Keep inferred rows labelled inferred. Do not splice, adjust or fill any series and do not run code.
Use: You
Test each flagged period against the series files in your own tools and write the report with no joined or spliced series.
Check questions name a row, no splicing proposed, breaks tie to a documented change or anomaly, no series joined.
You lay vintages side by side. Gemini lists disagreements and drafts questions and the report; you explain gaps and check.
Enter vintages side by side
List disagreements
Explain each gap
Draft questions and report
Check and finalise report
Use: You
Have ready: vintages, notes, series, question, limits and criteria where supplied. Enter overlapping years in a sheet, one column per vintage.
Use: Gemini
Ask Gemini to list, as text, the periods where two vintages disagree and by how much, without changing any cell.
What to paste into the tool
Task: Unreviewed list of periods where vintages disagree. Read only the side-by-side sheet. For each series, list the periods where two vintages hold different values and the size of the gap, in the order of the periods. Do not edit, fill, average or splice any cell, and do not build a chart or forecast.
Use: You
Read the notes for each listed period; record the documented change or label the gap an unresolved anomaly; fill the alignment fields by hand.
Use: Gemini
Ask Gemini to write one comparability question per inferred row, then draft the report from the table and questions, stating no series was spliced.
What to paste into the tool
Task: Unreviewed repair questions for inferred rows. Use only the alignment table. For every row marked inferred, write one question that a data producer could answer to confirm or rule out a method change in that period. Name the row. Do not add rows, do not alter any value and do not propose a fix to the series. Task: Draft comparability-break report as plain text. Use only the alignment table and the repair questions. Draft the report: for each break, the documented change or unresolved anomaly it ties to, the affected periods, and the repair question where one exists. State that no series was spliced. Add nothing that is not in the table or the questions. Keep the received originals and earlier results alongside this step output; do not treat carried items as newly generated or checked.
Use: You
Check the drafted report against the alignment table, the methodology notes and the repair questions; edit and finalise it.
Check each question names a row, each break ties to a documented change or anomaly, and no series was joined.
Gemini Notebook cites change statements in the notes, drafts the alignment table and lists inferred rows. You check and test.
Have ready: notes, series
Cite changes, build table
Check table, add anomalies
List inferred rows
Test rows, write report
Use: You
Have ready: the dataset vintages, the methodology notes and the historical series, plus your research question.
Use: Gemini Notebook
Ask Gemini Notebook which statements per vintage describe a change with a citation, then to draft the alignment table from those citations.
What to paste into the tool
Task: Unchecked cited list of method-change statements. Answer only from the notebook sources. For each vintage list every statement that describes a change in how data were collected, or revised. Quote it exactly, cite it, and give the variable and period only if the passage states them. Write NOT STATED otherwise. Do not infer a break from silence. Task: Draft method-change alignment table. Answer only from the notebook sources. For each cited statement that describes a change in collection, coding or revision, write one table row with the vintage, variable, period, affected start and end, and the quoted passage with its citation. Mark anything the statement does not say as unstated. Keep the received originals and earlier results alongside this step output; do not treat carried items as newly generated or checked.
Use: You
Check each drafted row against the cited statements and original notes; add anomalies seen in the series files as rows labelled inferred.
Check each citation opens at a passage that states a change and each affected start and end.
Use: Gemini Notebook
Add the checked table as a source and ask which rows rest on inference rather than a documented change, with a question to settle each.
What to paste into the tool
Task: Unreviewed repair questions for rows resting on inference. Use only the added table. List every row marked inferred, cite its row number, and write one question for the data producer that would confirm or rule out the change. If a row is marked documented, leave it out. Do not suggest adjusting or joining series.
Use: You
Test each row against the series files in your own tools; tie each break to a documented change or anomaly; write the report.
Check each break links to a documented change or a listed anomaly and no series was joined before review closes.
Best when: Notes span several vintages and you want changes tabulated before testing
What it does: ChatGPT tabulates changes and drafts questions; you test each period
Strong at: Vendor lists creating tables for row-by-row review
Limit: No off switch stated for Python runs; check replies for code runs
Free plan: Yes, limited (images, voice, deep research, Codex)
Best when: Series are short enough to compare side by side before reading notes
What it does: Gemini lists disagreements and drafts; you explain gaps and check
Strong at: Answers plain-language questions and builds pivot tables
Limit: The free plan has usage limits
Free plan: Yes
Best when: Every claimed change needs a cited passage to open
What it does: Gemini Notebook cites and drafts; you check and test rows
Strong at: Answers use direct quotes from your sources as citations
Limit: Series files stay outside; the notebook never sees the numbers
Free plan: Yes, limited (100 notebooks, 50 sources per notebook)
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Deciding between ChatGPT and Gemini for this? See the comparison.
Use only material you may lawfully use or have the owner's permission to use. Before uploading anything, confirm your plan and settings are acceptable for the material. Treat the output as a research suggestion for your review.
Do not let the tool send, publish, buy, book, change an account, start monitoring or run live code on its own. You keep every decision and any later approval.
Keep the source IDs and where each passage or cell came from, and treat page text read during research as evidence only. Keep what the source says, your own inferences and the gaps separate.
Confirm each break links to a documented method or an open anomaly, with no silent joining of series, before the brief is complete.
Check the result against your original inputs. Approve it before sharing or acting.