Research from supplied material

Answer a focused research question from documents you supply, trace each finding to a passage, show conflicts and unknowns, and write a concise answer within the evidence.

Sample results: runs and scores on this page are illustrations while testing is set up. How we test

Bring: Research question, Material to use

Use This TemplatePrepare a prompt for your own AI chat.

Who it is for: Anyone preparing a briefing or answer from a defined packet of material.
Time: 20 minutes

Open in the Editor

Other Formats and Files

The Step 1 chat prompt under Use This Template is the quick start. Choose a full prompt chain, checklist, Claude skill or n8n file when that destination fits your work.

Choose one format

Prompt chain
Copy each step into ChatGPT or Claude, one after another.
Checklist
Follow the steps and check off the checks by hand.
Claude skill
A folder Claude loads, so it follows the steps by itself.
n8n file
Import into n8n and connect your own AI account.

Using a download:How to use this in n8n How to install a Claude skill

The same job, with and without the template

Sample

This comparison uses the Typical sample. Try an input below for a result guide matched to each sample.

One plain prompt

Did the Saturday workshop pilot improve attendance and cost per attendee? The three notes are pasted below.

The Saturday pilot was a success. It attracted 80 people compared with 60 on weekdays and reached more unique residents. Although the program spent more, it was more efficient overall.

Problem:The plain answer claims more unique people were reached, although the notes only count attendances and include repeat visitors.

With this template

Answer

The Saturday pilot logged more attendances across four sessions, 80 in June versus 60 in April. Cost per attendance fell from $12 ($720 divided by 60) to $10 ($800 divided by 80). These counts include repeat visitors, so they do not show whether more unique people attended.

Evidence

The Spring log, April says "Four weekday sessions had 60 attendees. Total spent was $720." The Pilot log, June says "Four Saturday sessions had 80 attendees. Total spent was $800." The Staff note, June says "Attendance counts include repeat visitors. Unique visitors were not counted."

What remains unknown

Unique reach is unknown, and the packet does not explain whether the change came from the day of the week or another factor.

Check before using

Confirm that the logs count attendance the same way and that the packet contains the final spending figures.

Check:Each document row has a topic and a limit: All three titled documents have a topic and a limit.
Check:Every evidence passage is in the supplied material: Every quoted passage is copied from the supplied logs or note.
Check:The assessment names unknowns: Unknown unique reach is kept separate from the supported attendance comparison.
Check:The answer includes an unknowns section: The final answer has a What remains unknown section.
Check:Each material finding traces to the evidence table: The cost calculations show their source operands and the attendance claim keeps its qualifier.
For your read:You inspect key passages before sharing: Check log definitions and final spending before sharing.

Runs and scores

Sample

Claude, Sonnet model

October 12, 2026, typical sample

Check:Each document row has a topic and a limit
Three titled documents have a topic and a limit.
Check:Every evidence passage is in the supplied material
The attendance, spending and repeat-visitor passages match the packet.
Check:The assessment names unknowns
Unique reach stays unknown.
Check:The answer includes an unknowns section
The answer has a What remains unknown section.
Check:Each material finding traces to the evidence table
Cost per attendance uses the stated spending and counts.
For your read:You inspect key passages before sharing
Confirm counting definitions and final figures.

Fixed: Nothing needed fixing in this illustrative run; unique reach remains unknown.

Met 5 of 5, plus 1 for your read

1 check is for your read: the tool hands it to you.

ChatGPT, GPT model

October 13, 2026, easy sample

Check:Each document row has a topic and a limit
The garden notice is the single supplied source.
Check:Every evidence passage is in the supplied material
The opening-time passage is quoted from the notice.
Check:The assessment names unknowns
Closing time is left unknown.
Check:The answer includes an unknowns section
The response names the notice and says what it does not cover.
Check:Each material finding traces to the evidence table
The answer uses only the notice’s 8 am and May 4 facts.
For your read:You inspect key passages before sharing
Inspect the garden notice.

Fixed: Nothing needed fixing in this illustrative run; no closing time is invented.

Met 5 of 5, plus 1 for your read

1 check is for your read: the tool hands it to you.

n8n, import format 2.20

October 14, 2026, tricky sample

Check:Each document row has a topic and a limit
Both sheets and the coordinator note are mapped.
Check:Every evidence passage is in the supplied material
The 48 and 44 counts are quoted from different sheets.
Check:The assessment names unknowns
The counts remain conflicting because no correction is supplied.
Check:The answer includes an unknowns section
The answer states that the usable count is unresolved.
Check:Each material finding traces to the evidence table
Neither count is selected without support.
For your read:You inspect key passages before sharing
Reconcile the source sheets.

Fixed: Nothing needed fixing in this illustrative run; the source conflict stays visible.

Met 5 of 5, plus 1 for your read

1 check is for your read: the tool hands it to you.

Use This Template

Add your material below. Logic Lab prepares Step 1; you copy it into your own AI and check the answer there. The site does not run the workflow.

Example materialSample

Choose a sample

Typical: Saturday workshop pilotEasy: Community garden hoursTricky: Two conflicting inventory records

A notice answers one narrow question directly with a title and date.

Research question (needed)Material to use (needed)What the answer should cover (optional)

Step 1 Chat Prompt

Review the exact prompt, then paste it into your own AI. For later steps, use the chat chain in Other Formats and Files.

You are a research assistant who inventories a document packet.

Your task is to map the material that can answer this question: What time does the garden open on Saturdays from May 4?

Material:
Garden notice, April 20: From May 4, the community garden opens at 8 am on Saturdays. Weekday hours remain unchanged.

Rules:
- Give one row per titled document in the order supplied. Keep its title as written.
- State the relevant topic in plain words and one limit, such as a missing measure or unclear definition.
- Do not answer the question yet. Do not invent a source or a publication date.

Return one markdown table: | Document | Relevant topic | Limit |

Send to AICopy and open ChatGPTCopies this result and opens a blank chat. Paste and review it there.

Before Step 2, check: Each document row has a topic and a limit.

What This Sample Should Produce

The notice says the garden opens at 8 am on Saturdays from May 4. The answer should cite the notice and avoid inventing a closing time.

A result guide for this invented input, not an AI reply.

Want to see where the materials, steps and checks fit? Watch this template being built.

The steps, and the check on each

Step 1: Map the supplied documents
Record what each document can answer and where the packet is thin.
Hands back: A document map with relevant topics and limits.
Check:Each document row has a topic and a limit
A plain rule. If it fails, run the step again (the n8n file does this for you).
Step 2: Extract evidence for the question
Pull relevant facts and quote the source words behind each one.
Hands back: A source-linked evidence table for the question.
Check:Every evidence passage is in the supplied material
A plain rule. If it fails, run the step again (the n8n file does this for you).
Step 3: Sort findings and unknowns
Separate supported findings, conflicts and missing information before drafting.
Hands back: Findings and unresolved questions with their evidence limits.
Check:The assessment names unknowns
A plain rule. If it fails, run the step again (the n8n file does this for you).
Step 4: Write the sourced answer
Answer in plain words and keep evidence limits visible.
Hands back: A concise sourced answer with limits and a human review cue.
Check:The answer includes an unknowns section
A plain rule. If it fails, run the step again (the n8n file does this for you).

After the last step

Check:Each material finding traces to the evidence table
A second look by the AI. If it fails, run the step again (the n8n file does this for you).
Check:You inspect key passages before sharing
You check this one. If it fails, it is flagged for you.

What you need first

Research question(needed)
Ask one answerable question and name the period or group if it matters.
Material to use(needed)
Paste each document with a short title and date. Only these words will be used for the answer.
What the answer should cover(optional)
Add a useful limit, such as a location, audience or format. Leave blank to answer the question as written.

Ways to adapt it

This works only from the packet you supply. Add a new document to the material and rerun the template if the evidence changes.
Give each document a title and date so findings can be checked by a person.
For a large packet, run a focused question against one section at a time, then compare the evidence tables.
A plausible statement without a passage remains unknown until you add evidence.

Change any step or check in the Editor, then download it again.

The scoring guide

Sample

Each template check has a criterion marked met or not met; a person must still make the handed checks. (The guides in Quality rate a whole piece of work on three levels.) Under each one is how the three sample runs scored.

Download the scoring guide (Markdown)

Each document row has a topic and a limit
A plain rule
Meets:Each supplied document has a named topic and a meaningful limit.
Does not meet:A supplied document is omitted or has an empty topic or limit.
Claude:Met
ChatGPT:Met
n8n:Met
Every evidence passage is in the supplied material
A plain rule
Meets:Each evidence passage appears word for word in the supplied material.
Does not meet:A passage is paraphrased, invented or taken from outside the packet.
Claude:Met
ChatGPT:Met
n8n:Met
The assessment names unknowns
A plain rule
Meets:The assessment separates supported, conflicting and unknown points.
Does not meet:It guesses an unknown or silently resolves a source conflict.
Claude:Met
ChatGPT:Met
n8n:Met
The answer includes an unknowns section
A plain rule
Meets:The answer includes a section on what remains unknown.
Does not meet:It omits the unknowns section or treats a missing measure as proven.
Claude:Met
ChatGPT:Met
n8n:Met
Each material finding traces to the evidence table
A second look by the AI
Meets:Each material finding traces to a passage or a shown calculation from the packet.
Does not meet:A material claim lacks evidence, or a conflict disappears.
Claude:Met
ChatGPT:Met
n8n:Met
You inspect key passages before sharing
You check this one
Meets:You check the key passages and packet completeness before sharing.
Does not meet:You share the answer without checking a material passage or missing source.
Claude:For your read
ChatGPT:For your read
n8n:For your read

Claude: met 5 of 5, plus 1 for your read

ChatGPT: met 5 of 5, plus 1 for your read

n8n: met 5 of 5, plus 1 for your read

Learn from this template

The document map tells the extraction step which sources matter. The evidence table then gives the assessment and final answer a visible passage for each finding, so a reader can inspect the answer rather than trust its tone.

Where to see it: step 1 (Map the supplied documents); step 2 (Extract evidence for the question); step 3 (Sort findings and unknowns); step 4 (Write the sourced answer).

Read the guide: Hand-offs Take the lesson: Hand-offs (lesson 5)

More to read:Adapting a template

Practice:Spot the mistake: research summary Scoring guides

Which one fits?

Use Pull key facts into a table when you need named values copied into rows from one text instead of an answer to a research question.

Related templates

All Analyze templates

Analyze

Turn meeting notes into decisions and actions5 min
Every action has an owner and a date, or is flagged; every item quotes its note

Analyze

Find themes in comments or interview notes15 min
Each theme has a count and two quotes

Analyze

Sort changes into keep, fix or drop15 min
Each keep, fix or drop decision is linked to feedback and the goal

Last run: October 14, 2026: n8n, import format 2.20 (sample run)
Tools and versions: Claude, Sonnet model; ChatGPT, GPT model; n8n, import format 2.20
What was checked: The files, against each tool’s import format, and 3 sample runs, each check’s finding recorded.
Version: 1.0
What changed: First release.

Report a problem with this template How we test

Research from supplied material

How it runs

The same job, with and without the template

One plain prompt

With this template

Answer

Evidence

What remains unknown

Check before using

Runs and scores

Claude, Sonnet model

ChatGPT, GPT model

n8n, import format 2.20

Use This Template

Step 1 Chat Prompt

The steps, and the check on each

Step 1: Map the supplied documents

Step 2: Extract evidence for the question

Step 3: Sort findings and unknowns

Step 4: Write the sourced answer

After the last step

What you need first

Ways to adapt it

The scoring guide

Learn from this template

Which one fits?

Related templates

Analyze

Analyze

Analyze