AI Fact Checker Workflow: Verify Claims Step by Step

Tools:

• An AI assistant with web search (ChatGPT, Claude, Gemini or Perplexity)

• A normal search engine

• Reverse image search (Google Lens or similar)

• The official website or verified social accounts of whoever the claim is about

A message lands in your family WhatsApp group: "URGENT: new government policy from Monday, forward to everyone." It has a logo, a date and a confident tone. Half the group has already shared it. Is it true?

Pasting it into an AI fact checker and asking "Is this true?" feels like the obvious move, and it can help. But AI is only as reliable as the way you use it. Chatbots can misread a claim, rely on outdated information, or invent a source that sounds official. Used the wrong way, AI can make a false claim look even more convincing.

This workflow treats AI as a research assistant, not a judge. You use it to break claims apart, plan your checks and summarise what you find, while you confirm the important facts yourself. By the end of this workflow, you will have a clear verdict on any claim (true, false, misleading or unproven), the evidence behind it, and a short, polite correction you can share.

Before You Start: Why an AI Fact Checker Gets Things Wrong

Know why AI makes mistakes. Language models predict likely text; they do not look facts up unless they are using search, and even then they can misquote or misread pages. Our guide to what AI hallucination is explains this in plain language and is worth reading first.

Also, use a tool with live web search for anything recent. A model without search cannot know about a policy announced last week. Our comparison of Perplexity vs Google search explains when an AI answer engine helps and when a traditional search is better.

Step 1: Capture the Claim Exactly

Copy the full message, not your memory of it. Note where you saw it, when, and who sent it. Screenshot any images. Small wording details matter: "JAMB has cancelled" and "JAMB is considering" are very different claims.

Do not forward it "just in case" while you check. That is how false claims spread.

Here is a message I received. Do not tell me whether it is true yet.
Just tell me:
1. Who or what organisation it claims to speak for
2. What date or timeframe it refers to
3. What it asks people to do (forward, pay, click, register)
4. Any emotional or urgency language it uses

Message:
[paste the full text]
Where I saw it: [WhatsApp group / X / Facebook / TikTok]
Date received: [date]

Checkpoint: You have the exact claim saved, plus a note of its source, its date and what it wants you to do.

Step 2: Break It Into Checkable Claims

Most viral messages mix several claims together: a real organisation, a made-up policy, a genuine-looking date and a fake link. Each needs checking separately. This is where AI saves real time: it is good at claim extraction.

Break this message into separate, specific factual claims that could
each be checked. Number them. For each, say:
- what type it is (event, number, date, quote, image, link, person)
- which official or primary source would confirm or deny it
- whether it is opinion rather than a checkable fact

Message:
[paste message]

Checkpoint: You have a numbered list of three to eight checkable claims, each with the kind of source that could settle it.

Step 3: Ask for a Plan, Not a Verdict

Instead of "Is this true?", ask the AI what would prove or disprove each claim. This keeps it from guessing and gives you a checklist.

Pay attention to red flags it points out, such as fees for "registration", links that are not the organisation's real domain, deadlines that create panic, or a "leaked" document nobody official mentions.

For each numbered claim, tell me:
1. What evidence would prove it true
2. What evidence would prove it false
3. Where to look first (official site, verified social account,
   reputable news outlet, government gazette)
4. Any red flags that suggest a scam or hoax

Do not give a final verdict. If you do not know something, say so.

Checkpoint: You have a short checking plan for each claim and a list of red flags to look out for.

Step 4: Read Laterally

Professional fact-checkers use a habit called lateral reading. Instead of staying on the page or message and judging how official it looks, they open new tabs and check what other reliable sources say about it and about who is behind it.

For each main claim, search for it yourself. Look for the organisation's official website and verified social accounts, and for coverage by established news outlets. If a major policy really changed, you would normally see it in more than one reliable place. If you find it only in forwards, blogs with no named authors, or sites with odd domains, be cautious.

Then use AI to help you compare what you found.

I searched for these claims myself. Here is what I found:
[paste short notes or text from each source, with source names]

Compare the sources. Which claims are supported by an official or
reputable source? Which are contradicted? Which are only found in
unverified sources? Quote the exact sentences that support each point.

Checkpoint: For each claim, you know whether official or reputable sources support it, contradict it, or say nothing.

Step 5: Ask for Sources, Then Open Them Yourself

If you use an AI fact checker with search, ask it for sources. Then open every link and check three things: does the page exist, is it the organisation it says it is, and does it actually say what the AI claimed?

This step catches a common failure. AI tools sometimes cite a real page that says something different, an old page from a previous year, or a link that does not exist at all.

Using web search, find the most authoritative sources on claim
[number]. For each source give: the publisher, the page title, the
publication date and the exact sentence that is relevant.
Prefer official and primary sources. If you cannot find an official
source, say "no official source found" rather than guessing.

Checkpoint: Every source you rely on has been opened by you, is from a trustworthy publisher, is current, and says what was claimed.

Step 6: Check Numbers, Dates, Quotes and Images

These details are where false claims often break down.

Numbers. Are figures in the right units, from the right year, and from a named source? A real number from five years ago is often recycled as "new".

Dates. Check whether the date has already passed, falls on a weekend, or matches a real calendar event. Old news shared as new is very common.

Quotes. Search for the exact quote in quotation marks. If no reliable outlet carries it, treat it as unconfirmed.

Images and audio. Run images through reverse image search to see where they first appeared. Our guide on how to tell if an image is AI-generated covers the signs to look for. Voice notes "from" officials or family members can be cloned; our guide to AI voice cloning scams explains how to check.

Check this message for internal problems:
- Do the dates and days of the week match? (Today is [date].)
- Are any numbers inconsistent or implausible?
- Does the writing style, spelling or formatting differ from how
  [organisation] normally communicates?
- Does any link use a domain other than the official one: [official domain]?
List each problem with the exact text that caused it.

Checkpoint: You have checked every number, date, quote and image, and noted anything that does not add up.

Step 7: Reach a Verdict and Record Your Confidence

Now decide. Use one of four labels: True (confirmed by an official or multiple reliable sources), False (contradicted by official sources), Misleading (contains some truth but distorts it), or Unproven (not enough evidence either way). "Unproven" is an honest answer; it is not a failure.

Write down the evidence in a few lines. If the topic matters for work or study, our workflow to turn research into a clear brief helps you write up your findings properly.

Based only on the evidence below, help me write a fact-check summary:
- Verdict: [True / False / Misleading / Unproven]
- The claim in one sentence
- What the official or reliable sources say (with publisher and date)
- Any remaining uncertainty

Evidence I verified myself:
[paste notes and source names]
Do not add any new facts I have not provided.

Checkpoint: You have a verdict, a short evidence summary based only on sources you checked, and a note of what remains uncertain.

Step 8: Share a Calm Correction

If the claim is false or misleading and it is spreading in your groups, a short, respectful reply helps more than an argument. Point to the official source by name and avoid making the person who forwarded it feel foolish.

Write a short, friendly WhatsApp reply (under 70 words) explaining
that this message is [false/misleading/unproven]. Name the official
source that clarifies it. Warm tone, no blame. Suggest people check
[organisation]'s official channels before forwarding.

Checkpoint: You have shared a correction, or decided not to, and you have not forwarded the original message.

Worked Example: A Viral JAMB Message and a Fake Job Ad

Ibrahim, a teacher in Abuja, receives two forwards in one evening. The examples below are illustrative, not real announcements.

The JAMB message. It says: "JAMB announces that from next year, all UTME candidates must pay an extra ₦7,500 'verification fee' through the link below before 15 November. Forward to all parents." It carries a JAMB logo.

Ibrahim pastes it into an AI assistant and runs Steps 1 and 2. The AI extracts four claims: a new fee, its amount, a deadline, and a payment link. In Step 3 it flags red flags: payment through an unfamiliar link, urgency, and the instruction to forward.

He then reads laterally. He checks JAMB's official website and verified social media accounts and finds no such announcement. Searching the exact wording shows the same message, with a different amount, circulated last year. In Step 5 he asks the AI for sources; it cites a news article, but when Ibrahim opens it, the article is about a different, older issue. The AI had linked something related, not something that supported the claim.

The link domain in the message is not JAMB's official domain. Verdict: False, and likely a phishing attempt. He posts a calm reply in the parents' group naming JAMB's official channels.

The job ad. The second forward advertises "Federal Ministry recruitment, 3,000 vacancies, salary ₦350,000 monthly, apply with ₦5,000 processing fee". The AI fact checker extracts the claims and flags the application fee as a common scam sign. Ibrahim cannot find the vacancy on the ministry's official site. Our guide to spotting AI job scams and fake recruiters lists the same warning signs. He suggests his nephew look for checked listings on Nexframe Opportunities instead.

Total time: about 25 minutes for both. Neither message was forwarded again from his groups.

Common Mistakes

Asking "Is this true?" and stopping there. A single chatbot answer is not a fact-check. Use AI to plan and organise, then verify yourself.

Trusting citations you have not opened. AI can link to real pages that do not support the claim. Always open and read them.

Using a model without web search for recent news. It may not know about anything recent and may guess.

Treating "looks official" as evidence. Logos, letterheads and formal language are easy to fake, with or without AI.

Forwarding while you check. If you are not sure, do not share it.

Confusing "no evidence found" with "false". Sometimes the honest verdict is "unproven".

Next Steps

Save these prompts in your notes app so you can run a quick check from your phone. Start with the claims that ask you to pay, click or forward, because those carry the most risk. If you want to understand why AI makes these errors in the first place, revisit our hallucination guide.

Finally, make the habit social. When someone in your family or class group shares a claim you have checked, reply with the official source rather than a simple "this is fake". Over time, people start checking before they forward, and your groups become a little harder for scammers to exploit.

Frequently Asked Questions

Can AI fact-check information on its own?

Not reliably. AI can extract claims, suggest where to look and summarise sources, but it can also misread pages or invent citations. You need to confirm important facts with official or reputable sources yourself.

Which AI is best for fact-checking?

Use a tool with live web search and visible citations, such as Perplexity, ChatGPT with search, Gemini or Claude with web search turned on. The method matters more than the tool.

What is lateral reading?

Lateral reading means leaving the page or message and checking what other reliable sources say about the claim and the people behind it, rather than judging it by how it looks.

How do I check if a WhatsApp forward is fake?

Find the original source, check the organisation's official website and verified accounts, search the exact wording, check the dates and any links, and be wary of fees, urgency and instructions to forward.

Can AI tell if an image is fake?

AI detectors are not fully reliable. Reverse image search, checking the original source and looking for visual inconsistencies are more dependable.

What if I cannot find any evidence either way?

Label it unproven and do not share it. You can revisit it later if an official source comments.

Is it safe to paste private messages into an AI tool?

Remove names, phone numbers and personal details first. Paste only the text you need checked.