{"course":{"slug":"ai-for-students","title":"AI for Students","url":"https://addaly.com/learn/ai-for-students","level":"beginner","track":"work"},"lesson":{"slug":"search-grounded-answers","title":"When it searches the web","url":"https://addaly.com/learn/ai-for-students/search-grounded-answers","position":43,"of":73,"minutes":8,"answer":"Retrieval replaces invented sources with real sources attached to claims they do not make, so the check is unchanged: open the link, find the sentence, and follow reporting back to the original.","keyIdea":"Retrieval replaces invented sources with real sources attached to claims they do not make, so the check is unchanged: open the link, find the sentence, and follow reporting back to the original."},"body":{"format":"markdown","text":"## A different failure, not no failure\n\nMost assistants now retrieve web pages and cite them. This is a genuine improvement: the model is working from text it just fetched rather than compressed memory, so pure invention drops sharply.\n\nWhat replaces it is subtler and easier to trust by mistake.\n\n**The link is real, the claim is not in it.** The most common failure. A citation attached to a sentence the page does not support, sometimes because the page is about the topic generally and the specific claim was generated around it. Independent evaluations of search-grounded assistants keep finding this in a substantial minority of citations.\n\n**The source is weak.** Retrieval optimises for what ranks, not for what is reliable. A content-farm article, an SEO page, a five-year-old forum post and a Cochrane review are all just retrieved text.\n\n**The summary distorts.** Even from a good page, conditions and hedges drop out. *May reduce risk in some populations* becomes *reduces risk*.\n\n**Only part was read.** Retrieval usually fetches a portion. The model can answer confidently about a document it saw a fragment of.\n\n**Circularity.** An increasing share of the web is itself model-generated. A claim can be confirmed by three sources that all originate from the same unverified generated text. Multiple agreeing pages are weaker evidence than they used to be.\n\n## The check, which has not changed\n\nOpen the link. Find the sentence. That is the whole method, and it is why this course keeps repeating it: every generation of this technology changes what fails and leaves the check identical.\n\nWhen you open it, ask three things:\n\n1. **Does this page say what the answer said?**\n2. **Is this page a source or a repeater?** A news article about a study is not the study. Follow it back.\n3. **Would I have cited this if I had found it myself?** A page you would not have chosen does not become citable because a machine surfaced it.\n\n## Deep research modes\n\nSeveral tools now offer a longer, multi-step research mode that runs many searches and returns a report with dozens of citations. These are more thorough, and they concentrate the risk rather than removing it: a twenty-page report with sixty citations invites acceptance precisely because checking it looks expensive.\n\nUse them as a **map**, which is what they are good for: what the areas are, which names recur, which terms to search. Then do your own reading of the sources that matter. Never submit a report of this kind, or its reference list, as your own work, and note that a bibliography assembled by a tool you did not verify is a bibliography of sources you have not read, which is the misconduct finding from two lessons ago.\n\n## Wikipedia, used correctly\n\nWikipedia is an excellent starting point and not a citable source, and both halves are true for the same reason: it summarises, and it tells you what it summarised from.\n\nThe correct use is to **read the article and then use the reference list at the bottom.** Those are the real sources; go to them, read them, cite them. The article's talk page is also unusually informative: it is where disputes about the content are argued out, so you can see immediately whether a claim is contested.\n\nAnd for offline work, **Kiwix** packages Wikipedia for local reading, which is free and useful with intermittent data.\n\n## The one place it is unambiguously better\n\nRetrieval is a real improvement for anything time-bound: this year's entry requirements, the current version of a library, a deadline, an exam board specification, who currently holds an office. A model without search has no route to any of these and will answer from whenever its training stopped, often without saying so. If your question has a date in it, or implicitly has one, switch search on and open the page.\n\n## Building the habit\n\nA short discipline for any answer with links:\n\n- Open every link you intend to rely on. Not the ones you are ignoring.\n- Search within the page for the specific term from the claim.\n- If the page is reporting somebody else's work, follow it back to the original.\n- If you cannot find the claim in the page, the claim is unsupported. Delete it or find a real source. Do not keep it because it sounded right.\n\nThat last instruction is the one people break. A claim that survived because it was plausible, attached to a source that does not contain it, is the single most common way a careful student ends up in an integrity meeting."},"check":{"question":"Why does the growth of AI-generated web content weaken the practice of confirming a claim across several independent-looking pages?","options":[{"text":"Because several pages may all derive from the same unverified generated text, so agreement is not independence.","correct":true},{"text":"Because generated pages are more likely to contain factual errors than human-written ones, raising the error rate of any sample.","correct":false,"tempting":"Plausible on average and beside the point; the failure described is correlation between sources, not the error rate of each."},{"text":"Because generated pages rank lower in search, so retrieval increasingly returns only a narrow slice of the web.","correct":false,"tempting":"Makes an unsupported ranking claim and would, if true, reduce rather than create the problem."},{"text":"Because generated content lacks citations of its own, so the trail back to a primary source is broken at the first step.","correct":false,"tempting":"Describes a real inconvenience, but generated pages frequently do carry citations, which is part of why they are convincing."}]},"prev":{"slug":"what-counts-as-evidence","title":"What counts as evidence","url":"https://addaly.com/learn/ai-for-students/what-counts-as-evidence"},"next":{"slug":"notes-that-survive","title":"Notes that survive the term","url":"https://addaly.com/learn/ai-for-students/notes-that-survive"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'When it searches the web', in AI for Students. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-students/search-grounded-answers (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"When it searches the web","courseTitle":"AI for Students","module":{"index":5,"slug":"research-and-evidence","title":"Research, sources and evidence"},"answer":null,"keyIdea":"Retrieval replaces invented sources with real sources attached to claims they do not make, so the check is unchanged: open the link, find the sentence, and follow reporting back to the original.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"a-different-failure-not-no-failure","runs":[{"text":"A different failure, not no failure"}]},{"t":"p","runs":[{"text":"Most assistants now retrieve web pages and cite them. This is a genuine improvement: the model is working from text it just fetched rather than compressed memory, so pure invention drops sharply."}]},{"t":"p","runs":[{"text":"What replaces it is subtler and easier to trust by mistake."}]},{"t":"p","runs":[{"text":"The link is real, the claim is not in it.","b":true},{"text":" The most common failure. A citation attached to a sentence the page does not support, sometimes because the page is about the topic generally and the specific claim was generated around it. Independent evaluations of search-grounded assistants keep finding this in a substantial minority of citations."}]},{"t":"p","runs":[{"text":"The source is weak.","b":true},{"text":" Retrieval optimises for what ranks, not for what is reliable. A content-farm article, an SEO page, a five-year-old forum post and a Cochrane review are all just retrieved text."}]},{"t":"p","runs":[{"text":"The summary distorts.","b":true},{"text":" Even from a good page, conditions and hedges drop out. "},{"text":"May reduce risk in some populations","i":true},{"text":" becomes "},{"text":"reduces risk","i":true},{"text":"."}]},{"t":"p","runs":[{"text":"Only part was read.","b":true},{"text":" Retrieval usually fetches a portion. The model can answer confidently about a document it saw a fragment of."}]},{"t":"p","runs":[{"text":"Circularity.","b":true},{"text":" An increasing share of the web is itself model-generated. A claim can be confirmed by three sources that all originate from the same unverified generated text. Multiple agreeing pages are weaker evidence than they used to be."}]},{"t":"h2","id":"the-check-which-has-not-changed","runs":[{"text":"The check, which has not changed"}]},{"t":"p","runs":[{"text":"Open the link. Find the sentence. That is the whole method, and it is why this course keeps repeating it: every generation of this technology changes what fails and leaves the check identical."}]},{"t":"p","runs":[{"text":"When you open it, ask three things:"}]},{"t":"ol","start":1,"items":[[{"text":"Does this page say what the answer said?","b":true}],[{"text":"Is this page a source or a repeater?","b":true},{"text":" A news article about a study is not the study. Follow it back."}],[{"text":"Would I have cited this if I had found it myself?","b":true},{"text":" A page you would not have chosen does not become citable because a machine surfaced it."}]]},{"t":"h2","id":"deep-research-modes","runs":[{"text":"Deep research modes"}]},{"t":"p","runs":[{"text":"Several tools now offer a longer, multi-step research mode that runs many searches and returns a report with dozens of citations. These are more thorough, and they concentrate the risk rather than removing it: a twenty-page report with sixty citations invites acceptance precisely because checking it looks expensive."}]},{"t":"p","runs":[{"text":"Use them as a "},{"text":"map","b":true},{"text":", which is what they are good for: what the areas are, which names recur, which terms to search. Then do your own reading of the sources that matter. Never submit a report of this kind, or its reference list, as your own work, and note that a bibliography assembled by a tool you did not verify is a bibliography of sources you have not read, which is the misconduct finding from two lessons ago."}]},{"t":"h2","id":"wikipedia-used-correctly","runs":[{"text":"Wikipedia, used correctly"}]},{"t":"p","runs":[{"text":"Wikipedia is an excellent starting point and not a citable source, and both halves are true for the same reason: it summarises, and it tells you what it summarised from."}]},{"t":"p","runs":[{"text":"The correct use is to "},{"text":"read the article and then use the reference list at the bottom.","b":true},{"text":" Those are the real sources; go to them, read them, cite them. The article’s talk page is also unusually informative: it is where disputes about the content are argued out, so you can see immediately whether a claim is contested."}]},{"t":"p","runs":[{"text":"And for offline work, "},{"text":"Kiwix","b":true},{"text":" packages Wikipedia for local reading, which is free and useful with intermittent data."}]},{"t":"h2","id":"the-one-place-it-is-unambiguously-better","runs":[{"text":"The one place it is unambiguously better"}]},{"t":"p","runs":[{"text":"Retrieval is a real improvement for anything time-bound: this year’s entry requirements, the current version of a library, a deadline, an exam board specification, who currently holds an office. A model without search has no route to any of these and will answer from whenever its training stopped, often without saying so. If your question has a date in it, or implicitly has one, switch search on and open the page."}]},{"t":"h2","id":"building-the-habit","runs":[{"text":"Building the habit"}]},{"t":"p","runs":[{"text":"A short discipline for any answer with links:"}]},{"t":"ul","items":[[{"text":"Open every link you intend to rely on. Not the ones you are ignoring."}],[{"text":"Search within the page for the specific term from the claim."}],[{"text":"If the page is reporting somebody else’s work, follow it back to the original."}],[{"text":"If you cannot find the claim in the page, the claim is unsupported. Delete it or find a real source. Do not keep it because it sounded right."}]]},{"t":"p","runs":[{"text":"That last instruction is the one people break. A claim that survived because it was plausible, attached to a source that does not contain it, is the single most common way a careful student ends up in an integrity meeting."}]}],"check":{"question":[{"text":"Why does the growth of AI-generated web content weaken the practice of confirming a claim across several independent-looking pages?"}],"options":[{"text":[{"text":"Because several pages may all derive from the same unverified generated text, so agreement is not independence."}],"correct":true,"tempting":null},{"text":[{"text":"Because generated pages are more likely to contain factual errors than human-written ones, raising the error rate of any sample."}],"correct":false,"tempting":[{"text":"Plausible on average and beside the point; the failure described is correlation between sources, not the error rate of each."}]},{"text":[{"text":"Because generated pages rank lower in search, so retrieval increasingly returns only a narrow slice of the web."}],"correct":false,"tempting":[{"text":"Makes an unsupported ranking claim and would, if true, reduce rather than create the problem."}]},{"text":[{"text":"Because generated content lacks citations of its own, so the trail back to a primary source is broken at the first step."}],"correct":false,"tempting":[{"text":"Describes a real inconvenience, but generated pages frequently do carry citations, which is part of why they are convincing."}]}],"order":[2,1,0,3],"why":null},"prev":{"slug":"what-counts-as-evidence","title":"What counts as evidence","minutes":8},"next":{"slug":"notes-that-survive","title":"Notes that survive the term","minutes":8}}}