in collaboration with
Principle 07 · Before you share · Accuracy
The depth of checking matches the cost of being wrong, and every source relied on is opened and confirmed.
Key question
What would it cost if this were wrong?
Benefit
Your reader can trace a figure to the original data or find the passage behind a claim.
AI writes just as fluently when it’s wrong. It can invent a reference, misquote a source, drop a condition or turn “expected Friday” into “guaranteed.” NIST calls the invented part confabulation: false content stated with confidence, sometimes with made-up citations that make it look justified. Long reports make these errors harder to spot, and they have reached clients and courts.
You can’t check everything, so match the depth of the check to the cost of being wrong.
| Level | For | What you do |
|---|---|---|
| A glance | Internal drafts, ideas, notes to yourself | Read it through. Does it answer the question you asked? |
| Spot checks | Anything you share with colleagues | Check the names, dates, figures and claims that matter most. |
| Full verification | Anything published, signed, sent to a client or used to decide | Open every source, recalculate, confirm quotes and dates, and ask a second person when the stakes are high. |
Open the cited source and find the passage that supports the claim. Check its date and whether it applies to your country and market. A second AI can share the first one’s blind spots, so its agreement doesn’t replace opening the source.
Specialist tools reduce errors without removing them. A Stanford study published in 2025 found that leading AI legal research tools still hallucinated on 17% to 33% of test queries. By September 2026 a public database had logged more than 2,000 court and tribunal decisions involving AI-invented content, most of them fake citations.
A familiar face, voice or writing style can be convincing even when someone has imitated it with AI. In early 2024 an employee in Hong Kong transferred about HK$200 million after a video call in which the other participants, including the finance chief, were fakes. Read the case study.
Treat any unusual request for money, credentials or confidential data as unverified until you confirm it through a channel you start yourself: call back on a number you already have, or walk to the person’s desk. The FBI and Spain’s national cybersecurity institute both advise this.
Compare the two prompts and check which step you need to take yourself.
Research
Before
Write a paragraph on the growth of the European AI market, with statistics.
After
Using only the three reports I’ve attached, write a paragraph on the European AI market. After each number, give the report and page. List any claim you couldn’t support.
Your part
Check every number against its page. Numbers you can’t find come out.
Why it works: A report title and page number beside each statistic let you check it against the original.
Contracts and policies
Before
Summarize the new EU rules on AI chatbots.
After
Summarize Article 50 of the EU AI Act from the official text I’ve attached, in plain English. Quote the sentence behind each point, and flag anything that depends on dates or exceptions.
Your part
Check each quoted sentence in the official text, and ask your legal team before you act on it.
Why it works: Quotes tie every point to the text, and the flags show where the detail matters.
Data and numbers
Before
What’s the total cost of the three quotes?
After
Extract the price, quantity and any extra charges from each of the three attached quotes into a table, with the page for each figure. Then total them and show the calculation.
Your part
Check every figure against its page and redo the total yourself. Look for charges hidden in the terms.
Why it works: The page references let you check the figures against each supplier’s quote. Adding them yourself gives you a separate check on the total.
Agree which outputs get a glance, which get spot checks and which get full verification, so everyone checks in proportion.
Team toolVerification levelsThese sources support the facts on this page. We explain what each one establishes and where its evidence ends. The recommendations are ours. See every source we use.
Founderz (2026). Principle 7: Verification proportionate to the stakes. The RUAI Standard, 2026 edition. Developed by Founderz in collaboration with Microsoft. https://responsibleai.founderz.com/toolkit/principles/proportionate-verification
Licensed under CC BY 4.0: share and adapt with attribution.