Complete AI Training
Sign inGet my AI kit

Your job's AI kit

Get your AI kit

Tell us who you are and what you do. We show you your kit right away and email you the link: skills, prompts, AI agents, MCP servers and courses for your job.

500+ jobs ready, and we make a kit for any other job. No payment needed to look.

Share

AI agent for historians

Exhibit Label Fact Review Agent

Make every claim on every exhibit label traceable to a cited source.

Quality controlScience and research
See how it worksGet access
Exhibit Label Fact Review Agent: what goes in, what the agent does and what you get

What it does

Exhibit labels are short and confident, which makes unchecked claims easy to hide. This agent reads each label and splits it into individual claims: dates, names, quantities, causes and quotations. It checks each against the cited sources and the curator's research file, and marks it supported, partly supported or unsupported. For anything unsupported, it proposes safer wording, such as adding about, according to, or removing a causal claim. It records every change and rechecks the revised label against the sources, since a rewrite can introduce a new claim. It never edits the master text. The curator approves the final text. Edge case: a quotation is checked word for word against the source, including spelling.

How it works

Follow the arrows from top to bottom. The orange dashed arrow is the loop: when a check fails, the agent goes back and tries again.

Start and resultWhat it doesA check on its own workWaits for your OKGoes back and retries
Yes, continueYes, continueApprovedNoNo 1 STARTS WHEN Label drafts submitted 2 USES A TOOL Split each label into claims 3 USES A TOOL Find the support for each claim in the sources 4 CHECKS THE RESULT Is each claim stated in a named source? If not: search the wider research file and markremaining claims unsupported. Back to step 3. 5 DOES Propose safer wording for unsupported or partialclaims 6 USES A TOOL Check quotations word for word 7 DOES Produce a revised label with a change list 8 CHECKS THE RESULT Does the revised text contain any new unsupportedclaim? If not: re-split the revised label and check the newclaims. Back to step 2. 9 YOU APPROVE Curator approves the final text 10 RESULT Reviewed labels with source notes
Read the steps as a list
  1. Label drafts submitted
  2. Split each label into claims
  3. Find the support for each claim in the sources
  4. Is each claim stated in a named source?If not: search the wider research file and mark remaining claims unsupported. Back to step 3.
  5. Propose safer wording for unsupported or partial claims
  6. Check quotations word for word
  7. Produce a revised label with a change list
  8. Does the revised text contain any new unsupported claim?If not: re-split the revised label and check the new claims. Back to step 2.
  9. Curator approves the final textThe agent waits here for your OK.
  10. Reviewed labels with source notes

How it decides

A claim is supported only if a named source states it. Where sources conflict or only imply it, the agent proposes hedged wording rather than a firm statement.

  • Mark a claim unsupported if no cited source states it
  • Hedge causal claims unless a source states the cause
  • Require exact quotes to match the source including spelling
  • Keep label length within the word limit after changes

Make it yours

Every agent is a starting point. You choose these settings for your own situation.

  • Word limit per label (default 60)
  • House hedging phrases
  • Source ranking
  • Whether to check images and captions too

What keeps you in control

It always asks you first

  • Revised label text
  • Any removal of a claim

Hard limits

  • Never edit the master label file
  • Quote the source line for each check

It stops when

  • Done: all claims supported or hedged and approved
  • Stop: key claim cannot be sourced and the curator must decide

Set it up

We guide you through the set-up, step by step

Members get the full set-up guide for this agent. No technical skills needed: you copy, paste and upload.

10 minto set it up in your AI
5 AIsChatGPT, Claude, Copilot, Gemini, Grok
  • One set of instructions to paste into your AI, with the clicks for ChatGPT, Claude, Microsoft 365 Copilot, Gemini and Grok
  • The agent then walks you through connecting your own data, one source at a time
  • A downloadable copy with the flow chart, the rules and the full guide
Get access to this agent

An example run

What happensA 60-word label said a bridge was built in 1872 by 400 workers and caused the town to boom. The agent found 1872 in the city record, 380 workers in a payroll source, and no source for the boom. It rewrote it as about 380 workers and removed the causal claim. A recheck of the new text raised nothing. The curator approved it.

More agents for historians

Quality control · Writers

Nonfiction Claim Evidence Agent

A draft where every factual claim is backed by a source passage at the right strength.

8 steps1 check6 jobs
Creating · Writers

Nonfiction Outline Evidence Agent

An outline where every chapter is backed by available evidence.

7 steps1 check3 jobs
Research · Writers

Writer Research Question Agent

A focused research trail and a writing brief.

8 steps1 check4 jobs
Data cleanup · Science and research

Specimen Catalogue Reconciliation Agent

Bring catalog entries, labels and storage locations into agreement for a drawer, cabinet or collection.

10 steps1 check2 jobs