Verify a draft against its source facts

Checks a summary or draft claim by claim against a list of source facts and reports everything that was added, inferred or contradicted.

PLPromptByte Library@promptbytePublished Jul 31

TextResearch #fact checking #research #summarization #verification

Version 1 of 1 · published Jul 31, 2026Version 1 · only version

Example output

See all results

Anthropic · Claude Haiku 4.5 20251001 PLPromptByte Library Version 1

Prompt

Version 1 · 2 variables · 2 required

Fill in the values below — the customized prompt updates as you type. marks a required value.

Source facts{{source_facts}}

The ground truth, one granular fact per line. No compound sentences.

Text to check{{draft_text}}

The summary, memo or draft that is supposed to be derived from those facts.

Fill in Source facts, Text to check to copy the customized prompt.

Resultsfor version 1

All models Claude Haiku 4.51

Version 1 · only version

Claude Haiku 4.5 result by PromptByte Library
Anthropic · Claude Haiku 4.5 20251001 PLPromptByte Library (@promptbyte)Version 1Published Jul 31
Compare
CLAIM CHECK SUPPORTED - Survey ran in March - "Survey ran from 3 to 21 March" SUPPORTED - 412 responses from Team-plan customers - "412 responses from customers on the Team plan" SUPPORTED - 61% found onboarding slower than expected - "61% said onboarding took longer than expected" SUPPORTED - Nearly a fifth contacted support in week one - "18% contacted support during their first week" UNSUPPORTED - They contacted support because the setup wizard confused them - no supporting fact UNSUPPORTED - Response rate was over 10% - no supporting fact; 412 of 4,000 invitations is a figure you can…
Community submittedFull output
Settings: Unknown
Open result

Showing all 1 public result from Claude Haiku 4.5 on version 1.

Usage notes

Use this on anything that claims to be derived from a source: a summary of a paper, a research memo built from interview notes, a findings slide built from survey data, or a model-generated abstract you are about to reuse. Paste the source material as a flat list of facts and the derived text underneath.

The test it applies is entailment, not plausibility. A claim that is true in the world but absent from your fact list comes back UNSUPPORTED, which is the correct answer for a research artefact — the question is whether this document supports the sentence, not whether the sentence happens to be right.

The two failure modes it catches most reliably are causal upgrades (the source records that something happened, the draft explains why) and quiet generalisation to populations the source excluded. Both survive ordinary proofreading because they read fluently.

Make the fact list granular. One fact per line, no compound sentences: verification is only as sharp as the units you give it.

Attribution: adapted from https://github.com/dair-ai/Prompt-Engineering-Guide by DAIR.AI, licensed MIT. Source file: https://github.com/dair-ai/Prompt-Engineering-Guide/blob/main/pages/prompts/truthfulness/identify-hallucination.en.mdx

Verify a draft against its source facts

Example output

Prompt

Resultsfor version 1

Something wrong with this prompt?

Plan a point-by-point response to manuscript reviewers

Turn study findings into a research-poster content plan

Map conflicting terminology across research disciplines

Audit in-text citations against the reference list

Create a repeatable observational data-collection protocol

Draft a qualitative codebook from anonymized excerpts

Discussion 0

Example output

Prompt

Resultsfor version 1

Something wrong with this prompt?

Related prompts

Plan a point-by-point response to manuscript reviewers

Turn study findings into a research-poster content plan

Map conflicting terminology across research disciplines

Audit in-text citations against the reference list

Create a repeatable observational data-collection protocol

Draft a qualitative codebook from anonymized excerpts

Discussion 0