DOABLE · SAFETY CHECK
Spot the AI nonsense
AI can sound certain even when it is guessing. Use PAUSE before you trust, share, spend money, or make an important decision.
THE REAL-LIFE PROBLEM
What this guide helps you do
AI can deliver an outdated, generalized, or invented claim with the same confident tone it uses for a correct answer.
WHEN THIS HELPS
Use it when an AI claim affects money, health, safety, legal questions, reputation, or a fact you plan to repeat as true.
WHEN TO SKIP IT
Skip the full check for low-stakes brainstorming, drafting, or organizing material you already know and will revise yourself.
ONE MAIN PROMPT
Ask AI to challenge the claim before you trust it
COPY, PASTE AND PERSONALIZE
CONTEXT AI gave me this claim or recommendation: [paste the claim]. I may use it to [describe what you may decide, share, buy, or do]. ROLE Act as a skeptical fact-checking assistant. Help me identify what is known, what is uncertain, and what needs an outside source. INTERVIEW Ask me up to 3 short questions, one at a time, about where the claim came from, how current it needs to be, and what would happen if it were wrong. TASK 1. Restate the exact claim in one sentence. 2. Name the specific source supporting it, if one is available. 3. Flag anything that may be outdated, generalized, or based on an assumption. 4. Give me the strongest reason the claim could be wrong. 5. Tell me which reliable outside source or qualified person should confirm it before I act. Do not treat another AI answer as independent verification.
The PAUSE method
Use these five steps, in order, whenever AI gives you a fact, statistic, or claim you plan to act on.
Pin down the claim
Restate exactly what AI told you in one sentence. If you cannot state it simply, you do not understand it well enough to trust it yet.
Ask for the source
Reply: "Where did that come from? Give me the specific source." A real source is a named study, agency, or document. "General knowledge" or no answer is a signal to check further.
Update the facts
Ask: "Is this still current, or could it have changed recently?" AI can carry outdated information, especially for prices, policies, schedules, and anything that changes yearly.
Seek another reliable source
Check the claim against one outside source, ideally a government site, a primary source, or established reporting, not another AI chat.
Escalate when stakes are high
If the decision involves money, health, safety, or something legal, do not stop at a web search. Talk to a person qualified to advise you.
See it in action
You ask how long you have to return a refrigerator that does not fit through your doorway. AI confidently gives a specific number of days.
AI told me I have 15 days to return a major appliance, no questions asked.
AI says that is typical for major retailers, but cannot name the store or point to its policy.
AI admits return windows vary and large appliances can have shorter windows or restocking fees.
Your receipt and the retailer's policy show a different window, plus a fee AI never mentioned.
You call the store to confirm the exact deadline for your purchase before the window closes.
The AI was not lying. It gave a reasonable general answer. But reasonable and general is not the same as correct for your purchase, and only the check caught the difference.
Confident claim vs. what to check
| AI says this confidently | What to check before you act |
|---|---|
| This is the current rate, price, or rule. | Confirm it against the official current source. |
| Studies show... | Ask which study, by whom, and when. No named study means it is not verified. |
| This is generally safe. | Ask for exceptions and whether any could apply to you. |
| A specific number or statistic. | Verify any number tied to money, health, safety, or legal decisions. |
| Yes, that is correct, after a leading question. | Rephrase the question neutrally and ask again. |
When to trust AI
This is not a guide about distrusting AI. Most everyday uses do not need this level of scrutiny.
The rule of thumb is simple: the more a wrong answer would cost you, the more of PAUSE you should use.
BEFORE YOU ACT
What to verify yourself
What AI can do
- Give you a fast starting answer.
- Name sources when asked directly.
- Flag exceptions when prompted.
- Check a claim from another angle.
What you decide
- Decide whether the claim needs checking.
- Judge whether the source is reliable.
- Notice whether your situation fits the general answer.
- Know when to ask a qualified person.
Sources and details
- Stanford HAI, 2026 AI Index Report, Responsible AI chapter, published April 13, 2026. The chapter reports the 22% to 94% range and the measured effect of framing a false belief as the user's belief.
These are benchmark statistics, not a guarantee about any single response. PAUSE is Considered's practical editorial framework for applying established verification habits.
Last checked: August 3, 2026
AI can sound sure and still be wrong. A little checking, on the claims that matter, is what makes AI worth using.