Fact-Checking SKILL.md

This is the exact file agents read, including the credit header and licence notice. Save it as fact-checking/SKILL.md.

---
name: fact-checking
description: "Fact-check documents, decks, articles and AI outputs: extracting every claim, prioritising by risk, verifying against primary sources, recalculating numbers and units, confirming names, dates and quotes, spotting misleading-but-true framing, fixing wording and keeping a fact-check log. Use whenever someone needs to fact-check a report, article, pitch, social post or AI-generated text, verify statistics or quotes, check numbers before publishing or confirm that citations say what is claimed."
license: CC-BY-4.0
metadata:
  author: Afeez Adebayo
  author_url: https://afeezadebayo.com
  author_linkedin: https://www.linkedin.com/in/afeez-adebayo/
  author_x: https://x.com/za_icon
  source: https://afeezadebayo.com/skills-md/productivity/research-learning/fact-checking/
  category: productivity
  subcategory: research-learning
  version: "1.0.0"
  created: "2026-10-10"
  updated: "2026-10-14"
---

# Fact-Checking

One wrong number can undo a good report, pitch or article. Errors creep in through misread charts, outdated statistics, misquotes, figures copied from secondary sources, units confused, percentages calculated wrongly or AI-generated text that sounds right but is not. Professional fact-checkers work systematically: they list every checkable claim, verify each against the best available source and record what they found, rather than reading through and hoping errors stand out.

This skill fact-checks documents, presentations, articles, emails and AI outputs before they go out: extracting claims, prioritising by risk, verifying against primary sources, checking numbers and calculations, confirming quotes and names, correcting or softening claims, and keeping a record of what was checked.

## Operating principles
1. **List claims explicitly.** Every number, name, date, quote and factual statement is a claim to check.
2. **Prioritise by risk.** Claims that would damage trust, money or legal standing if wrong come first.
3. **Primary sources win.** Check against the original data, document, person or record.
4. **Recalculate numbers.** Verify sums, percentages, growth rates and units yourself.
5. **Fix wording, not only facts.** A true number can still be misleading if framed wrongly.
6. **Record the check.** Source, result and changes for every claim.

## Claim types and how to check them
| Claim type | Example | How to check |
|---|---|---|
| Statistic | "62% of SMEs use cloud accounting" | Original survey, sample, date, exact wording |
| Calculation | "Revenue grew 18% year on year" | Recalculate from source figures |
| Name and title | "Jane Doe, CFO of Acme" | Company website, recent filings or LinkedIn |
| Date and event | "The law took effect in March 2024" | Official gazette or regulator |
| Quote | "We expect growth to slow" | Original transcript, recording or article |
| Comparison | "Twice as fast as competitors" | Comparable data on both sides, same method |
| Causal claim | "The new pricing caused churn" | Evidence beyond timing; other explanations considered |
| Superlative | "The largest provider in Africa" | Authoritative ranking with date and definition |

## When to use, and when not to
Use it before publishing articles, sending reports or pitches, presenting to leadership or investors, posting on social media and relying on AI-generated research.

Don't apply full fact-checking to casual internal notes; scale the effort to the stakes.

## Inputs
- The document or output.
- Sources used to write it (if available).
- Audience and stakes (public, investors, legal).
- Deadline.

## Workflow

### Step 1. Extract claims
Go through the text and list each checkable claim in a table with its location.

### Step 2. Prioritise
Mark claims as critical (central to the argument or risky), important or minor.

### Step 3. Find the best source for each
Prefer primary sources; use the author's sources first, then independent ones.

### Step 4. Verify
Compare wording, numbers, dates, definitions and context. For numbers, recalculate.

### Step 5. Classify the result
Correct, correct with clarification, incorrect, outdated, unverifiable or misleading.

### Step 6. Fix
Correct the claim, add the source and date, soften wording to match evidence, or remove it.

### Step 7. Check the whole
Read again for consistency: the same figure used the same way throughout, totals matching parts, charts matching text.

### Step 8. Record
Keep the fact-check log with the document.

## Fact-check log template
```
| # | Location | Claim | Priority | Source checked | Result | Fix |
| 1 | p.1 para 2 | 62% of SMEs use cloud accounting | Critical | Original survey 2025, n=1,200 UK SMEs | Correct with clarification | "62% of UK SMEs surveyed in 2025" |
| 2 | Slide 4 | Revenue up 18% | Critical | Finance report | Incorrect (14.6%) | Corrected |
| 3 | p.3 | Largest provider in Africa | Important | No authoritative ranking | Unverifiable | Removed |
```

## Number checks
- Percentages: (new - old) / old; percentage points versus percent change.
- Totals: parts add up to totals in tables and charts.
- Units: thousands versus millions; currencies and exchange rates with dates.
- Rounding: consistent and not misleading.
- Base effects: large percentage changes from small bases are labelled.
- Averages: mean versus median stated where it matters.

## Misleading-but-true patterns
| Pattern | Fix |
|---|---|
| Cherry-picked period | Show the longer trend |
| Relative risk without absolute numbers | Add absolute figures |
| Correlation presented as cause | Use "associated with" or add evidence |
| Survey of customers presented as all users | State who was surveyed |
| Old data presented as current | Add the year |

## Fact-checking AI outputs
- Treat every specific fact, number, quote and citation as unverified.
- Open each cited link and confirm it says what is claimed.
- Watch for invented studies, misattributed quotes and plausible but wrong numbers.
- Ask the AI to list its claims with sources, then check them yourself.

## Quote and name checks
- Match quotes word for word with the original; mark cuts with ellipses and keep the meaning.
- Confirm the speaker, date and context of each quote.
- Check spelling of names, titles and organisations against official sources.
- Confirm current job titles; people change roles often.

## Team fact-checking process
For important documents, separate writing and checking: the writer lists sources for every claim, a second person checks critical claims against them, and disagreements go to the document owner. A shared log means nobody has to repeat checks when the document is reused.

## Corrections after publishing
If an error is found after publishing, correct it promptly, note the correction and date where readers will see it and tell anyone who relied on the wrong figure. Quiet edits damage trust more than visible corrections.

Re-check figures whenever the document is reused later.

Keep the log for future audits.

Schedule checks before deadlines, not after.

## Edge cases and judgement calls
| Situation | What to do |
|---|---|
| Source unavailable or paywalled | Mark as unverified; find another source or remove |
| Sources conflict | Use the most authoritative and recent; note the conflict if material |
| Claim about a person | Extra care; avoid unverified allegations; check legal risk |
| Tight deadline | Check critical claims only; flag the rest |
| Author disagrees with a correction | Show the evidence; escalate to the editor or owner |
| Live data that changes | State the date and time of the figure |

## Handoffs to related skills
- Source credibility: `source-evaluation`
- Finding sources: `desk-research`
- AI research tools: `ai-assisted-research`
- Reports and summaries: the Workplace Writing skills
- Data presentation: the Communication and Presentations `data-storytelling` skill

## Worked example
Before an investor update went out, the fact-check extracted 34 claims. Two critical errors appeared: year-on-year revenue growth was stated as 18% but the finance report showed 14.6%, and a market-size figure came from a blog that misquoted a 2019 report. A superlative ("largest in West Africa") had no authoritative source and was removed. Three statistics gained dates and samples. The founder sent a corrected update, and an investor later praised its accuracy when comparing it with the company's data room.

## Common mistakes
- Reading through instead of listing claims.
- Checking against the same secondary source the writer used.
- Not recalculating numbers.
- Ignoring misleading framing of true facts.
- Trusting AI citations without opening them.

## Output format
```
1. Claims list with priority
2. Fact-check log with sources and results
3. Corrected text or tracked changes
4. Remaining unverified items and risks
```

## Self-check before you answer
- Was every checkable claim listed?
- Were critical claims verified against primary sources?
- Were numbers recalculated and units checked?
- Is there a log of checks and fixes?

---
Created by Afeez Adebayo, Marketing, SEO & Growth Expert
Website: https://afeezadebayo.com
LinkedIn: https://www.linkedin.com/in/afeez-adebayo/
X: https://x.com/za_icon
Source: https://afeezadebayo.com/skills-md/productivity/research-learning/fact-checking/

Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). You may copy, share and adapt this skill, including for commercial use, as long as you credit Afeez Adebayo and link to the source URL above. Keep this notice in any copy or adapted version.