Fact-Checking with AI

🔄 Quick Recall: In the last lesson, you learned to use AI for research — summarizing documents, building background dossiers, and discovering sources. Now let’s apply that same AI capability to one of journalism’s most critical functions: making sure what you publish is true.

Why AI Fact-Checking Matters

Every published error erodes trust. A wrong statistic, a misattributed quote, a name misspelling — these are the cracks that undermine credibility. Traditional fact-checking is thorough but slow. AI won’t replace your editorial standards, but it can catch errors faster and flag items that need deeper verification.

Think of AI fact-checking as a quality control layer that runs before your story reaches an editor. It catches the obvious problems so human fact-checkers can focus on the subtle ones.

The AI Fact-Check Workflow

Structure your fact-checking as a three-pass system:

Pass 1 — AI sweep. Feed your draft to AI and ask it to flag claims that need verification.

Pass 2 — Human verification. For each flagged item, trace it to a primary source. This is where you do the real fact-checking.

Pass 3 — Final check. Review the corrected draft for consistency. Did fixing one fact change the meaning of a paragraph? Did a number correction affect a calculation elsewhere?

Review this article draft for factual accuracy. For every factual claim (statistics, dates, names, quotes, historical references), provide:

1. The specific claim
2. Your confidence level (high / medium / low / cannot verify)
3. Potential issues (outdated data, common misquotation, ambiguous context)
4. Suggested verification approach (which primary source to check)

Flag ANY claim you're less than 90% confident about. It's better to over-flag than to miss something.

✅ Quick Check: Why is the three-pass system more effective than asking AI to fact-check and trusting the results?

Because AI can be wrong with high confidence. The three-pass system uses AI only for the first sweep — identifying claims that need checking. Humans do the actual verification against primary sources. This combines AI’s speed (scanning an entire article in seconds) with human judgment (evaluating source credibility and context).

Checking Statistics and Data

Statistics are where errors hide most easily. A number gets misremembered, a percentage is applied to the wrong base, or a study’s findings are overstated.

I'm fact-checking this statistic from my article: "[paste the specific claim]"

Please:
1. Is this statistic commonly cited? What's the original source?
2. Is the number correct, or is a slightly different number more accurate?
3. Is the context accurate? (Sometimes a true number is applied to the wrong group or time period)
4. When was this data collected? Is it still current?
5. Are there common ways this statistic is misused or misunderstood?

The most dangerous errors aren’t completely wrong numbers — they’re slightly wrong numbers that sound right. “Unemployment fell to 3.4%” when it actually fell to 3.7% is harder to catch than a completely fabricated statistic.

Verifying Quotes and Attributions

Misquotes are both common and damaging. AI can help you cross-reference:

Direct quotes: Paste the quote and ask AI to find the original source. Was it really said by this person? In this context? Is the full quote different from the excerpt you’re using?

Attributed positions: “Senator X has always supported Y” — AI can search for instances where Senator X may have held a different position, catching oversimplifications.

Historical references: “As Einstein once said…” — Many famous quotes are misattributed. AI can flag quotes with disputed origins.

Detecting Misinformation Patterns

AI can identify common misinformation tactics:

• Misleading framing: A true statistic presented in a way that implies a false conclusion
• Cherry-picked data: Selecting only the data points that support a narrative while ignoring contradicting ones
• Out-of-context quotes: Real quotes removed from their original context to change their meaning
• False equivalence: Treating a fringe position as equivalent to mainstream consensus
• Outdated information: Using old data to describe current situations

I received this claim from a source: "[paste claim]"

Analyze it for:
1. Is the core factual claim accurate?
2. Is it presented in appropriate context?
3. Does it use any common misinformation patterns (cherry-picking, false equivalence, misleading framing)?
4. What's the strongest counter-argument or complicating factor?
5. What additional reporting would I need to responsibly include this claim?

Pre-Publication Checklist

Before your story goes live, run this systematic check:

Names and titles: Ask AI to verify the spelling and current title of every person mentioned. People change jobs; titles become outdated.

Dates and timelines: Have AI review your chronology. Events sometimes get placed in the wrong order, especially in complex narratives.

Math and percentages: Ask AI to recalculate any math in your story. If you say “a 40% increase from $2 million,” have AI confirm the resulting number.

Institutional names: Organizations rename themselves, merge, or restructure. Verify that you’re using the current correct name.

Exercise: Fact-Check a Draft

Take an article draft you’re working on (or select a published article from a major outlet) and run the AI fact-checking workflow:

1. Feed the full text to an AI tool using the fact-check prompt above
2. Note every claim AI flags as less than 90% confident
3. For the top 3 flagged items, trace each to its primary source
4. Compare: Was AI right to flag them? Were any errors real?
5. Run the pre-publication checklist on names, dates, and math

Key Takeaways

• AI fact-checking is a first-pass filter that catches obvious errors and flags items for human verification
• The three-pass system (AI sweep → human verification → final check) combines speed with accuracy
• Statistics are the most common source of errors — always trace numbers to their original source and check methodology
• AI can detect misinformation patterns like cherry-picking, false equivalence, and misleading framing
• Pre-publication checks for names, dates, math, and institutional names prevent embarrassing corrections
• AI fact-checking supplements but never replaces editorial judgment — the journalist makes the final call

Up Next: In the next lesson, you’ll learn how AI can help you write — crafting stronger headlines, tighter leads, and better-structured stories without sacrificing your voice.