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 →
Gemini Notebook: builds the map and drafts, you check
Gemini Notebook builds the cited task-to-evidence map and drafts the package from your checked map. You confirm every row first.
- You
Add the sources
- Gemini Notebook
Build the map
- You
Check the map
- Gemini Notebook
Draft the package
- You
Finish the copy
Have ready: the role tasks and materials
Use: You
The approved role tasks, permitted sample materials and candidate time limit. Add the authorised sources to one Gemini Notebook.
Check
Use fictional or licensed data; never obtain unpaid usable production work or infer protected traits.
Build the task-to-evidence map
Use: Gemini Notebook
Ask for a task-to-evidence map with scope and resource assumptions, a citation on each row and an explicit gaps list.
Details
What to paste into the tool
Work only from the uploaded authorized sources. Build the task-to-evidence map with scope and resource assumptions as a table: one row per item, asking for a source citation and effective date on each row (the reviewer rejects any row without a citation), and an explicit 'not stated' entry where evidence is missing. Do not add anything that is not in the sources. Use fictional or licensed data; never obtain unpaid usable production work or infer protected traits.
Check every map row and citation
Use: You
Check each row against the original sources; reject any row without a citation and mark gaps before drafting.
Check
Every row traces to a cited source passage; gaps stay listed.
Draft the brief, rules and observation sheet
Use: Gemini Notebook
From the checked map only, draft the work-sample brief, permitted-assistance rules and reviewer observation sheet with an unresolved-question section.
Details
What to paste into the tool
Using only the reviewer-confirmed rows of the task-to-evidence map with scope and resource assumptions in this notebook, draft the work-sample brief, permitted-assistance rules and reviewer observation sheet. Mark it DRAFT, ask for a citation next to each statement (the reviewer rejects any statement without one), and end with an unresolved-question section listing every open item. Use fictional or licensed data; never obtain unpaid usable production work or infer protected traits.
Prepare the final copy and record the decision
Use: You
Copy the saved note into a review document, re-check every reference against the originals and write the issue log.
Check
A role specialist confirms the exercise matches real work within the stated time. Keep it as preparation only. Your step: assemble the review copy, verify each reference survived the move out of the notebook, log issues, and write down the final decision and who approved it.
ChatGPT: drafts only from a hand-built checked map
You build the map by hand; only the checked map and limits go to ChatGPT for drafting.
- You
Build the map by hand
- ChatGPT
Draft the package
- You
Compare and finish
Build the task-to-evidence map by hand
Use: You
Have ready: the approved role tasks, permitted sample materials and candidate time limit. Read the sources yourself and build the map with locations and gaps.
Check
Your step: read the authorised sources directly and build the task-to-evidence map with scope and resource assumptions by hand, one row per item, with source locations, dates and explicit gaps.
Draft the work-sample package from the map
Use: ChatGPT
Upload only the checked map, supplied limits and unresolved questions; ask for the brief, rules and observation sheet, marked as draft.
Details
What to paste into the tool
Draft the work-sample brief, permitted-assistance rules and reviewer observation sheet using only the uploaded reviewer-checked task-to-evidence map with scope and resource assumptions and the supplied policy, role or process limits. Mark it DRAFT, keep the map's row IDs next to each statement, and list every unresolved question at the end. Do not add facts that are not in the uploaded map. Use fictional or licensed data; never obtain unpaid usable production work or infer protected traits.
Compare the draft with the map and finish
Use: You
Compare the draft line by line with the checked map, apply corrections, write the issue log and record the final decision.
Check
Keep it as preparation only. Your step: verify the draft against the checked map, assemble the review copy, log issues, and write down the final decision and who approved it.
ChatGPT: builds and drafts, LanguageTool checks the language
ChatGPT builds the map and drafts from your checked map. LanguageTool runs a language-only pass on your edited draft.
- You
Have ready: evidence
- ChatGPT
Build the map
- You
Check each row
- ChatGPT
Draft the package
- You
Check the draft
- LanguageTool
Language pass
- You
Accept fixes, finish
Have ready: the authorised evidence
Use: You
Have ready the approved role tasks, permitted sample materials and candidate time limit, as authorised evidence to paste.
Check
Use fictional or licensed data; never obtain unpaid usable production work or infer protected traits.
Build the task-to-evidence map
Use: ChatGPT
Ask ChatGPT to build the task-to-evidence map with scope and resource assumptions from the pasted evidence, with open items listed.
Details
What to paste into the tool
Build a task-to-evidence map from the evidence I paste: one row per approved role task, the permitted sample material and source location for each row, scope and resource assumptions, and the candidate time limit. List anything the evidence does not cover as an open item. Do not add tasks or requirements that are not in the evidence. Use fictional or licensed data; never obtain unpaid usable production work or infer protected traits. Preparation only: no autonomous hiring, employee rating, pay or access change, record update, external message, purchase or legal decision. Named humans retain authority. Evidence: [paste the authorised evidence here].
Check every map row against the sources
Use: You
Check every map row against the source locations and drop any row that does not trace. Do not ask for the draft before this check.
Check
Check the scope, resource assumptions and the candidate time limit are realistic.
Draft the package from the checked map
Use: ChatGPT
Ask ChatGPT to draft the work-sample brief, permitted-assistance rules and reviewer observation sheet from the checked map only, marked DRAFT.
Details
What to paste into the tool
Draft a work-sample package from my checked map only: a work-sample brief, permitted-assistance rules and a reviewer observation sheet. Mark it DRAFT. Beside each item, name the map row it comes from. Do not add tasks, rules or scoring items that are not in the map. Include an unresolved-question section listing the open items. Use fictional or licensed data; never obtain unpaid usable production work or infer protected traits. Preparation only: no autonomous hiring, employee rating, pay or access change, record update, external message, purchase or legal decision. Named humans retain authority. Checked map: [paste it here].
Check the draft against the map
Use: You
Compare each item with the checked map and edit the draft; a role specialist confirms the exercise matches real work within the stated time.
Check
Every item traces to a map row; open items stay listed.
Check grammar and spelling only
Use: LanguageTool
Paste the edited draft into LanguageTool for a grammar and spelling pass; content and source fidelity stay with you.
Details
What to paste into the tool
Check this human-written draft for grammar and spelling only. Do not add content, facts or reasons. Use fictional or licensed data; never obtain unpaid usable production work or infer protected traits.
Accept language fixes and finish the copy
Use: You
Accept or reject each language suggestion, finalise the review copy, write the issue log and record the decision and who approved it.
Check
The reviewer accepts or rejects each suggestion; no content, reason or commitment is added by the tool. Keep it as preparation only. Your step: assemble the final review copy, log issues, and record the human disposition.
Compare the three approaches
Option A: Gemini Notebook: builds the map and drafts, you check
Best when: You want citations beside each statement to check against originals
What it does: Gemini Notebook builds the map and drafts; you check
Strong at: Answers questions grounded in your sources
Limit: Creative requests may be refused or need rephrasing
Free plan: Yes
Option B: ChatGPT: drafts only from a hand-built checked map
Best when: You want the raw sources to stay with people only
What it does: You build the map by hand; ChatGPT drafts from it
Strong at: Writes, rewrites and summarises text
Limit: Drafting from a checked map not vendor-described; check each line
Free plan: Yes, limited (uploads, images, voice, analysis)
Option C: ChatGPT: builds and drafts, LanguageTool checks the language
Best when: You want one tool for map and draft, plus a language pass
What it does: ChatGPT builds and drafts; LanguageTool checks language; you check
Strong at: Checks grammar and spelling in 30+ languages
Limit: ChatGPT Free has usage limits on file uploads; paste text instead
Free plan: ChatGPT: Yes, limited; LanguageTool: Yes, limited
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Final check
Never paste credentials or full payment details.
Hiring manager and HR reviewer: Check each draft or result against your original inputs. Use fictional or licensed data; never obtain unpaid usable production work or infer protected traits.
Hiring manager and HR reviewer: A role specialist confirms the exercise matches real work and can be attempted within the stated time. Confirm all changes and the authorised audience; any communication, record update or employment decision happens separately under human authority.
Preparation only: no autonomous hiring, employee rating, pay or access change, record update, external message, purchase or legal decision. Named humans retain authority.
Speed and quality are relevant user priorities, not measured performance claims; no product binding or benchmark is asserted.
Use only material authorised for this purpose and audience; cut down employee and candidate personal data.
Keep source references, effective dates and unresolved facts; do not infer protected traits or fabricate missing evidence.
Check the result against your original inputs. Approve it before sharing or acting.