Sign in Start free

Verify Claims Before Publish

Fact checking usually degrades into confirming that a number appears somewhere on the internet, which is how a misquoted figure survives ten years of citation. The real work is tracing claims to a primary source and checking the source still says what everyone thinks it says. In regulated sectors an unverified claim is a compliance issue, not just an accuracy one.

CATEGORY
Content production
FORMAT
verify-claims-before-publish.md
STEPS
6
PRICE
Free - no account
WHEN TO REACH FOR THIS

Use when a draft contains statistics, quotes or factual assertions that will be visible to customers or regulators.

The skill file

verify-claims-before-publish.md
---
name: verify-claims-before-publish
description: Use when a draft contains statistics, quotes or factual assertions that will be visible to customers or regulators.
---

# Verify Claims Before Publish

Fact checking usually degrades into confirming that a number appears somewhere on the internet, which is how a misquoted figure survives ten years of citation. The real work is tracing claims to a primary source and checking the source still says what everyone thinks it says. In regulated sectors an unverified claim is a compliance issue, not just an accuracy one.

## What you need first

- the draft with every factual claim marked
- access to primary sources or the ability to request them
- a record of which claims were checked and when

## Method

1. Separate claims into three types: verifiable facts, opinions, and predictions. Only the first needs sourcing; the other two need attribution instead, which is a different fix.
2. Trace each statistic back through citations to the original study or dataset. Stop only when you reach the body that produced the number - a news article citing a report citing a survey is three removes from evidence.
3. Check the date and the population of the original. A figure that was true for one country in one year is routinely republished as a universal constant.
4. For any claim you cannot trace to a primary source, either cut it or restate it as attributed opinion. Do not keep it with a weaker verb.
5. Check claims about your own product against what the product currently does, not what the roadmap says. Marketing pages describing unreleased behavior are the most common source of complaints.
6. Record source URL, publication date and access date next to each claim in a working file, so the next refresh does not repeat the entire trace.

## What this produces

A draft where every retained factual claim has a primary source logged with its date, and untraceable claims are cut or attributed.

## Where this goes wrong

- Accepting a figure because several sites repeat it - repetition across sites usually traces to one original, often misread
- Leaving statistics undated on the page, so a correct 2019 figure reads as a current one and quietly becomes false
- Checking the headline number but not the methodology, then repeating a survey of 200 self-selected respondents as an industry benchmark

---

From the QuQi skill library - https://www.quqi.io/skills/verify-claims-before-publish
Free to download · no account, no email

What you need first

  • the draft with every factual claim marked
  • access to primary sources or the ability to request them
  • a record of which claims were checked and when

Method

  1. 01 Separate claims into three types: verifiable facts, opinions, and predictions. Only the first needs sourcing; the other two need attribution instead, which is a different fix.
  2. 02 Trace each statistic back through citations to the original study or dataset. Stop only when you reach the body that produced the number - a news article citing a report citing a survey is three removes from evidence.
  3. 03 Check the date and the population of the original. A figure that was true for one country in one year is routinely republished as a universal constant.
  4. 04 For any claim you cannot trace to a primary source, either cut it or restate it as attributed opinion. Do not keep it with a weaker verb.
  5. 05 Check claims about your own product against what the product currently does, not what the roadmap says. Marketing pages describing unreleased behavior are the most common source of complaints.
  6. 06 Record source URL, publication date and access date next to each claim in a working file, so the next refresh does not repeat the entire trace.

What this produces

A draft where every retained factual claim has a primary source logged with its date, and untraceable claims are cut or attributed.

Where this goes wrong

  • Accepting a figure because several sites repeat it - repetition across sites usually traces to one original, often misread
  • Leaving statistics undated on the page, so a correct 2019 figure reads as a current one and quietly becomes false
  • Checking the headline number but not the methodology, then repeating a survey of 200 self-selected respondents as an industry benchmark

Use this skill in your own AI

The download is a plain markdown file with the name and trigger in its frontmatter. Where an assistant supports skills it can load itself, that frontmatter is what it reads to decide this one applies.

Claude Code Save it as ~/.claude/skills/verify-claims-before-publish/SKILL.md and Claude loads it on its own when what you are doing matches the trigger line. Put it in .claude/skills inside a project instead if the whole team should have it.
Claude Upload the file in the skills section of your settings. Once it is there it applies itself in any conversation where the trigger fits, so you do not have to remember it exists.
ChatGPT There is no skills format to install into, so paste the file contents into a Project instruction or a Custom GPT instead. It then applies to every chat in that project rather than only the one you paste it into.
Anything else Paste the markdown into the chat before your question. It works in any assistant, it just has to be pasted again each time.

Questions about this skill

When do I need a full trace rather than a quick search to confirm a figure?

When the page is customer-facing or the sector is regulated. Confirming a number appears somewhere online is how a misquoted figure survives a decade of citation, since repetition across sites usually traces back to one original that was misread. For low-stakes internal copy the quick check is defensible; for a claim carrying commercial or compliance weight it is not.

What do I need before starting, and what happens without it?

The draft with every factual claim already marked, access to primary sources, and a working file to record what you checked. Without the marking pass you verify whatever catches your eye and miss the confident sentence that contains no number. Without the record, the next refresh repeats the entire trace, which is why second checks rarely happen.

What comes out of this, and which part gets used later?

A draft where each retained fact carries a primary source with its publication date, and untraceable claims are cut or restated as attributed opinion. The dated log beside each claim is the part with a second life: it survives publication, tells the next editor what was checked and when, and reduces the annual refresh to spot checks.

What is the mistake that costs the most here?

Checking the headline number and not the methodology. A survey of two hundred self-selected respondents, repeated as an industry benchmark, is wrong in a way no URL check catches and is exactly what a competitor or a regulator picks up. Leaving statistics undated on the page runs it close: a correct 2019 figure quietly becomes a false current one.

More in Content production