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 →
Claude builds the register, proposes duplicate pairs and drafts the plan. You check it against the originals. No files change.
Get documents ready
Draft the filing plan
Check and approve
Use: You
Have ready: backlog folder, filing taxonomy and retention schedule. Give each file an ID and keep an ID key with the originals.
Check your organisation's permission and account settings allow this upload.
Use: Claude
Upload the files and taxonomy. Ask Claude for a classification register, duplicate pairs, then a draft filing plan with folder and retention rule per file.
What to paste into the tool
Task: Classification register with file IDs, party tokens, periods, source locations and unclear reasons. Use one row per file ID; record document type, party token and period with source page or text location; use UNCLEAR with a reason where evidence is absent; keep all file IDs; classify only, without changing files Task: Duplicate-pair list with file IDs and matching evidence. Pair file IDs using matching reference, period and content evidence; distinguish identical-file candidates from similar documents; include both source locations; flag uncertain pairs for opening side by side; do not remove files Task: Draft mapping table with folder, retention rule citation and unclear reasons. For every file ID return class, target folder and retention class only from the supplied rules; quote the rule and include duplicate-pair IDs from the supplied list; mark UNCLEAR with a reason when no rule applies; produce a table for local review; assemble the results into a draft filing plan table for local review
Use: You
Check every file mapping and duplicate pair against the originals. Approve the draft plan.
Check rule citations and unclear rows against the originals. Open both originals of every duplicate pair. Each file needs a class and target, or an unclear reason. Pair duplicates; delete nothing; stop at the approved plan.
You build a file list. Gemini proposes classes and duplicates from metadata, then folders. You correct from the originals.
Build metadata sheet
Propose classes
Correct from content
Propose folders, retention
Check and assemble
Use: You
Have ready: backlog folder, filing taxonomy and retention schedule. Build a file list with stable file IDs, names, dates and sizes, plus rule ranges.
Check your organisation's permission and account settings allow this upload.
Use: Gemini
Enter the task text as the AI function prompt with the metadata range as context; select Generate and Insert.
What to paste into the tool
Task: Proposed class and period columns and duplicate-pair candidates with UNCLEAR reasons. Use only the supplied row range and rule definitions; propose type and period from descriptive name tokens; flag likely duplicate pairs from matching name tokens, dates and sizes, with both file IDs; return UNCLEAR with a reason for missing content; enter the Task text as the AI function prompt with the explicit context range, then select Generate and Insert
Use: You
Open the documents and correct every proposed class from the content. Check or reject each proposed duplicate pair.
Treat filename-based classes as proposals: the AI has not read the documents. Check party and period, record a source location, and open both originals of every duplicate pair. Leave unclear items with reasons.
Use: Gemini
Use explicit context ranges: corrected classes, duplicate-pair IDs and taxonomy rules. Ask for target folder, retention class and rule citation per file.
What to paste into the tool
Task: Draft folder and retention columns with rule citations. Include class rows, duplicate-pair IDs and taxonomy rules in the explicit optional range; do not rely on a separate unprovided tab; return target folder, retention class, exact rule citation and UNCLEAR reasons as text; keep every file ID and duplicate-pair ID
Use: You
Check row coverage and rule citations against the original schedule. Assemble the filing plan without changing any files.
Each file needs a class and target, or an unclear verdict. Check party and period against the originals. Pair duplicates without removing files.
You prepare one database row per document. Basic Autofill proposes class, folder and retention. Duplicate pairing stays with you.
Prepare Notion rows
Propose properties
Check and approve
Use: You
Have ready: backlog folder, taxonomy, retention schedule. Note duplicate pairs. Create one Notion row per document with page text and retention rules; leave properties empty.
Read each original for type, party and period. Open duplicate candidates side by side; note paired file IDs. Record unclear reasons and source locations. Leave originals unchanged, no attachments. Check organisation permission and account settings allow uploading these hand-built rows.
Use: Notion AI
In Basic Autofill choose a custom instruction for the class, folder and retention properties. It reads each row page only.
What to paste into the tool
Task: Proposed class, target-folder and retention properties with cited rule text. In Basic Autofill choose a custom instruction for the class, target-folder and retention text properties; use only this row page and its supplied rule text; return file ID, class, folder, retention class and cited rule or UNCLEAR with reason; no web search, Notion Agent or Custom Agent is part of this path
Use: You
Check every proposed class, folder and retention class against the originals and schedule. Check or reject each duplicate pair; approve the filing draft.
Check a sample first. Basic Autofill reads one row at a time, so duplicate pairing stays with you. Keep unclear reasons. Stop before moving or deleting files.
Best when: Document content is needed to propose classes and duplicates.
What it does: Claude reads, classifies and pairs; you check and approve.
Strong at: Cites source locations for each classification.
Limit: Document-format support does not establish filing classification.
Free plan: Yes, limited (usage capped in 5-hour and weekly windows)
Best when: File names and dates are enough for a first pass.
What it does: Gemini proposes from metadata; you correct from the originals.
Strong at: Works in Sheets without reading document content.
Limit: Categorisation does not establish retention-rule compliance.
Free plan: Yes, limited
Best when: You want proposals inside a Notion database.
What it does: Autofill proposes properties; you pair duplicates and approve.
Strong at: Cites the retention rule text for each proposal.
Limit: Property tagging does not establish correct retention assignments.
Free plan: Paid plan needed (Free and Plus get a trial)
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Deciding between Claude and Gemini for this? See the comparison.
Never paste credentials or full payment details.
Every file needs a class and target or an unclear reason.
Duplicates remain paired for removal review without file changes.
Check the result against your original inputs. Approve it before sharing or acting.