{"course":{"slug":"ai-and-you","title":"AI, Safety and What Goes Wrong","url":"https://addaly.com/learn/ai-and-you","level":"beginner","track":"start"},"lesson":{"slug":"the-second-source-habit","title":"How professional checkers actually work","url":"https://addaly.com/learn/ai-and-you/the-second-source-habit","position":23,"of":73,"minutes":7,"answer":"Judging a claim by reading it closely fails; professionals leave the page immediately, extract one checkable claim at a time, and read what independent primary sources say about it.","keyIdea":"Judging a claim by reading it closely fails; professionals leave the page immediately, extract one checkable claim at a time, and read what independent primary sources say about it."},"body":{"format":"markdown","text":"## The counter-intuitive finding\n\nIn a study at Stanford, historians, undergraduates and professional fact-checkers were given unfamiliar websites and asked to judge their reliability. The historians and students did what educated people do: they read the site carefully, examined its design, looked at its About page, weighed its arguments. They were frequently fooled.\n\nThe fact-checkers were fast and accurate, and the reason was that they barely read the site at all. Within seconds they had left it, opened new tabs, and were reading what other sources said about the organisation. The researchers called this **lateral reading**, as opposed to reading vertically down the page.\n\nScrutinising a fluent paragraph for signs of falsehood is the historians' strategy, and it fails for the same reason: a well-constructed page and a well-constructed paragraph both look exactly like a good one. Leave, and check elsewhere.\n\n> Figure: Reading down the page against reading sideways. Vertical reading: the historians and undergraduates: Read the site carefully, top to bottom; Examined the design and the About page; Weighed the arguments on their merits; Frequently fooled. Lateral reading: the professional fact-checkers: Left the page within seconds; Opened new tabs on the organisation and the claim; Read what independent sources said about it; Fast, and accurate. The historians and students were fooled; the fact-checkers were fast and right. A fluent paragraph and a well-built page both look exactly like good ones, which is why judging them by reading harder does not work.\n\n## The core move, in four steps\n\n**Pull out the checkable atom.** A paragraph is not checkable; a claim is. \"The scheme covered 1.2 million households by March 2023\" is an atom. Extract them one at a time, because fluent text glues true and false atoms together and reading for overall impression cannot separate them.\n\n**Go sideways, not down.** Search the specific claim, in its own words, plus a term that would bring up the original: the scheme's name, the ministry, the journal. Do not search for confirmation of the sentence: search for the entity and read what it says.\n\n**Reach the primary document.** Court filing, statute, annual report, dataset, press release, published paper. This is one or two clicks in most cases and it settles most disputes immediately.\n\n**Check what the claim omits.** Correct-but-misleading is more common than false. A real figure from a real report describing a different population, year or definition.\n\n## Tools, all free\n\n- **Reverse image search**: Google Lens, TinEye, Bing Visual Search, Yandex. For any photograph presented as evidence, this is the first move. An image that has been online for six years is not from yesterday's event.\n- **The Wayback Machine**: what a page said before it was edited, and whether it existed at all. Also the answer when a source has been quietly deleted.\n- **Frame-level video checking**: take a screenshot of a distinctive frame and reverse-search that. Repurposed old footage is by volume a far more common problem than deepfakes.\n- **Site registration data**: a domain registered three weeks ago and presenting itself as a decades-old institute is settled with one lookup.\n- **Statistical offices and official registries**: India's data portal, Eurostat, the ONS, the World Bank, and national company registries. They publish the tables underneath the headlines and they are free.\n- **Semantic Scholar, PubMed, arXiv and Google Scholar**: for whether a paper exists at all, and for its abstract, which contains the population and the effect size.\n\n## Where this fits with AI use\n\nA workable routine for a claim that matters:\n\nFirst, ask *what kind of claim is this?* Recent, niche, numerical, or about a named person: those are the four high-risk categories from the hallucination lesson, and they deserve checking before anything else.\n\nSecond, extract the atoms and check each with a lateral search.\n\nThird, if two independent sources disagree, do not average them. Find out why they disagree, because the reason is usually a definition (different years, different inclusion criteria, different geography) and understanding it is more valuable than either number.\n\nAnd fourth, know when to stop. Verification cost should track consequence. Two minutes for something you will say in a meeting; twenty for something going into a document with your name on it; more if someone could be harmed. The triage from the first module is exactly the schedule for this.\n\n## The habit worth building\n\nSpeed comes from doing it in a fixed order rather than from working harder. Atom, sideways search, primary source, omission check. Four steps, most of the time under three minutes."},"check":{"question":"You are sent a striking photograph said to show flooding in a named city yesterday. What is the highest-value first check?","options":[{"text":"Examine it for signs of AI generation, such as warped signs or odd shadows and hands","correct":false,"tempting":"Looks for artefacts in the pixels, which is slow, unreliable, and misses the far commoner case of a genuine old photograph."},{"text":"Check whether the account that posted it has a history of accurate reporting","correct":false,"tempting":"Reputation is weak evidence about one item. Accounts that post real images also forward misattributed ones."},{"text":"Reverse image search it, since reused old photos are far commoner than fakes","correct":true},{"text":"Ask a model whether the image is authentic and fits the event being reported","correct":false,"tempting":"Hands the check to a system that cannot verify where an image came from, and will sound sure either way."}],"why":"Read sideways first: find where the picture has appeared before. A real photo from another flood is the likeliest story."},"prev":{"slug":"search-results-are-not-sources","title":"When a summary launders a source","url":"https://addaly.com/learn/ai-and-you/search-results-are-not-sources"},"next":{"slug":"numbers-that-do-not-add-up","title":"Checking a number without looking it up","url":"https://addaly.com/learn/ai-and-you/numbers-that-do-not-add-up"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Source and assessment review continues. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'How professional checkers actually work', in AI, Safety and What Goes Wrong. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-and-you/the-second-source-habit (Accessed: 21 November 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"How professional checkers actually work","courseTitle":"AI, Safety and What Goes Wrong","module":{"index":3,"slug":"truth-and-verification","title":"Truth, sources and verification"},"answer":null,"keyIdea":"Judging a claim by reading it closely fails; professionals leave the page immediately, extract one checkable claim at a time, and read what independent primary sources say about it.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"caption":"The historians and students were fooled; the fact-checkers were fast and right. A fluent paragraph and a well-built page both look exactly like good ones, which is why judging them by reading harder does not work.","left":{"title":"Vertical reading: the historians and undergraduates","items":["Read the site carefully, top to bottom","Examined the design and the About page","Weighed the arguments on their merits","Frequently fooled"]},"right":{"title":"Lateral reading: the professional fact-checkers","items":["Left the page within seconds","Opened new tabs on the organisation and the claim","Read what independent sources said about it","Fast, and accurate"]},"title":"Reading down the page against reading sideways","type":"compare"},"alt":"Figure: Reading down the page against reading sideways. Vertical reading: the historians and undergraduates: Read the site carefully, top to bottom; Examined the design and the About page; Weighed the arguments on their merits; Frequently fooled. Lateral reading: the professional fact-checkers: Left the page within seconds; Opened new tabs on the organisation and the claim; Read what independent sources said about it; Fast, and accurate. The historians and students were fooled; the fact-checkers were fast and right. A fluent paragraph and a well-built page both look exactly like good ones, which is why judging them by reading harder does not work."}},"blocks":[{"t":"h2","id":"the-counter-intuitive-finding","runs":[{"text":"The counter-intuitive finding"}]},{"t":"p","runs":[{"text":"In a study at Stanford, historians, undergraduates and professional fact-checkers were given unfamiliar websites and asked to judge their reliability. The historians and students did what educated people do: they read the site carefully, examined its design, looked at its About page, weighed its arguments. They were frequently fooled."}]},{"t":"p","runs":[{"text":"The fact-checkers were fast and accurate, and the reason was that they barely read the site at all. Within seconds they had left it, opened new tabs, and were reading what other sources said about the organisation. The researchers called this "},{"text":"lateral reading","b":true},{"text":", as opposed to reading vertically down the page."}]},{"t":"p","runs":[{"text":"Scrutinising a fluent paragraph for signs of falsehood is the historians’ strategy, and it fails for the same reason: a well-constructed page and a well-constructed paragraph both look exactly like a good one. Leave, and check elsewhere."}]},{"t":"h2","id":"the-core-move-in-four-steps","runs":[{"text":"The core move, in four steps"}]},{"t":"p","runs":[{"text":"Pull out the checkable atom.","b":true},{"text":" A paragraph is not checkable; a claim is. “The scheme covered 1.2 million households by March 2023” is an atom. Extract them one at a time, because fluent text glues true and false atoms together and reading for overall impression cannot separate them."}]},{"t":"p","runs":[{"text":"Go sideways, not down.","b":true},{"text":" Search the specific claim, in its own words, plus a term that would bring up the original: the scheme’s name, the ministry, the journal. Do not search for confirmation of the sentence: search for the entity and read what it says."}]},{"t":"p","runs":[{"text":"Reach the primary document.","b":true},{"text":" Court filing, statute, annual report, dataset, press release, published paper. This is one or two clicks in most cases and it settles most disputes immediately."}]},{"t":"p","runs":[{"text":"Check what the claim omits.","b":true},{"text":" Correct-but-misleading is more common than false. A real figure from a real report describing a different population, year or definition."}]},{"t":"h2","id":"tools-all-free","runs":[{"text":"Tools, all free"}]},{"t":"ul","items":[[{"text":"Reverse image search","b":true},{"text":": Google Lens, TinEye, Bing Visual Search, Yandex. For any photograph presented as evidence, this is the first move. An image that has been online for six years is not from yesterday’s event."}],[{"text":"The Wayback Machine","b":true},{"text":": what a page said before it was edited, and whether it existed at all. Also the answer when a source has been quietly deleted."}],[{"text":"Frame-level video checking","b":true},{"text":": take a screenshot of a distinctive frame and reverse-search that. Repurposed old footage is by volume a far more common problem than deepfakes."}],[{"text":"Site registration data","b":true},{"text":": a domain registered three weeks ago and presenting itself as a decades-old institute is settled with one lookup."}],[{"text":"Statistical offices and official registries","b":true},{"text":": India’s data portal, Eurostat, the ONS, the World Bank, and national company registries. They publish the tables underneath the headlines and they are free."}],[{"text":"Semantic Scholar, PubMed, arXiv and Google Scholar","b":true},{"text":": for whether a paper exists at all, and for its abstract, which contains the population and the effect size."}]]},{"t":"h2","id":"where-this-fits-with-ai-use","runs":[{"text":"Where this fits with AI use"}]},{"t":"p","runs":[{"text":"A workable routine for a claim that matters:"}]},{"t":"p","runs":[{"text":"First, ask "},{"text":"what kind of claim is this?","i":true},{"text":" Recent, niche, numerical, or about a named person: those are the four high-risk categories from the hallucination lesson, and they deserve checking before anything else."}]},{"t":"p","runs":[{"text":"Second, extract the atoms and check each with a lateral search."}]},{"t":"p","runs":[{"text":"Third, if two independent sources disagree, do not average them. Find out why they disagree, because the reason is usually a definition (different years, different inclusion criteria, different geography) and understanding it is more valuable than either number."}]},{"t":"p","runs":[{"text":"And fourth, know when to stop. Verification cost should track consequence. Two minutes for something you will say in a meeting; twenty for something going into a document with your name on it; more if someone could be harmed. The triage from the first module is exactly the schedule for this."}]},{"t":"h2","id":"the-habit-worth-building","runs":[{"text":"The habit worth building"}]},{"t":"p","runs":[{"text":"Speed comes from doing it in a fixed order rather than from working harder. Atom, sideways search, primary source, omission check. Four steps, most of the time under three minutes."}]}],"check":{"question":[{"text":"You are sent a striking photograph said to show flooding in a named city yesterday. What is the highest-value first check?"}],"options":[{"text":[{"text":"Examine it for signs of AI generation, such as warped signs or odd shadows and hands"}],"correct":false,"tempting":[{"text":"Looks for artefacts in the pixels, which is slow, unreliable, and misses the far commoner case of a genuine old photograph."}]},{"text":[{"text":"Check whether the account that posted it has a history of accurate reporting"}],"correct":false,"tempting":[{"text":"Reputation is weak evidence about one item. Accounts that post real images also forward misattributed ones."}]},{"text":[{"text":"Reverse image search it, since reused old photos are far commoner than fakes"}],"correct":true,"tempting":null},{"text":[{"text":"Ask a model whether the image is authentic and fits the event being reported"}],"correct":false,"tempting":[{"text":"Hands the check to a system that cannot verify where an image came from, and will sound sure either way."}]}],"order":[3,0,2,1],"why":[{"text":"Read sideways first: find where the picture has appeared before. A real photo from another flood is the likeliest story."}]},"prev":{"slug":"search-results-are-not-sources","title":"When a summary launders a source","minutes":7},"next":{"slug":"numbers-that-do-not-add-up","title":"Checking a number without looking it up","minutes":8}}}