{"course":{"slug":"ai-for-teachers","title":"AI for Teachers","url":"https://addaly.com/learn/ai-for-teachers","level":"beginner","track":"work"},"lesson":{"slug":"why-it-cannot-look-it-up","title":"Why it cannot look it up","url":"https://addaly.com/learn/ai-for-teachers/why-it-cannot-look-it-up","position":4,"of":57,"minutes":8,"answer":"A model on its own has no way to consult anything at answer time; when a tool adds web search it fetches pages and summarises them.","keyIdea":"A model on its own has no way to consult anything at answer time; when a tool adds web search it fetches pages and summarises them, which fixes staleness but not judgement about which page to trust."},"body":{"format":"markdown","text":"## Two very different things wearing the same interface\n\nWhen you type a question into a chatbot, one of two things happens, and the box looks the same either way.\n\n**Mode one: the model answers from what it learned.** Nothing is consulted. No page is opened. The answer is produced word by word from the patterns absorbed during training, which finished months ago. This is what a language model is. It cannot check, because there is nothing to check against: the training text is not stored as a library it can open, it is dissolved into billions of numerical weights.\n\n**Mode two: the tool searches first.** Many products now bolt a web search onto the model. Your question is turned into a search, some pages are fetched, and the model is handed their text along with your question. It then writes an answer that summarises what it was given. This is a genuine improvement for anything recent or specific, and it changes what you should worry about.\n\n## How to tell which happened\n\nLook for sources. A searched answer shows links, citations, or a line like \"Searched the web\". An unsearched answer shows nothing. If you see confident specifics and no sources, you are in mode one, and the specifics are reconstructed from old patterns.\n\n> Figure: What actually happens when the tool searches. Steps: 1. You type a question (The box looks identical whether or not a search follows.); 2. The tool writes a search query (Often not your question. A vague question makes a vague query.); 3. Pages are fetched and ranked (A confident content farm ranks alongside the exam board.); 4. The page text goes to the model (Along with your question, inside the same memory window.); 5. The model summarises (Exceptions and conditions fall out first; they are the least predictable words.); 6. You get an answer with links (The links are the gain. Open them.). Six steps, and the model only enters at step four. Every judgement about which pages to trust is made by pattern, at step three and step five. That is why a searched answer is a research assistant's note rather than a source.\n\nMost free tiers search automatically for questions that look like they need it, and not otherwise. You can force it: \"Search the web before answering, and show me the pages you used.\" You can also forbid it, which is useful when you want the model's general explanation rather than a summary of whatever ranked first on the web this morning.\n\n## What search fixes\n\n- **Staleness.** The training cutoff stops mattering for the question at hand. The new syllabus, this year's dates, the latest results are reachable.\n- **Local specifics, sometimes.** If your district's population is on a page the search finds, you get it. If no page has it, you are back to guessing, and the model may not tell you that it found nothing useful.\n- **Checkability.** You now have a page to open. This is the real gain. Open it.\n\n## What search does not fix\n\nThe model still has to decide which pages to trust, and it decides the way it decides everything: by pattern. A confident, well-formatted page ranks with an accurate one. A content farm that copied an old syllabus looks like the exam board. If three pages disagree, the model may blend them into a fourth answer none of them said.\n\nIt also summarises, and summaries lose things. The exceptions, the footnote that says \"for the 2024 cohort only\", the conditions: these fall out first, because they are the least predictable part of a sentence.\n\nAnd the search itself can go wrong. A vague question produces a vague search, which fetches irrelevant pages, which produce an answer built on the wrong material. When an answer seems oddly off, look at what it searched for. Often the search query was not your question.\n\n## The teacher's rule\n\nFor anything that must be right (a date, a figure, a rule from the board) the model's job is to find the page, and your job is to read it. Treat a searched answer as a research assistant's note: \"I think it's on this page, and it seems to say this.\" Then open the page.\n\nFor anything conceptual (an explanation, an analogy, a set of likely misconceptions) search usually makes the answer worse, because it replaces the model's broad synthesis with a summary of two websites. Turn it off for those.\n\n## An example that comes up every year\n\nA teacher asks for the marks allocation in this year's Grade 12 physics paper. Without search, the model produces a plausible table from previous years. With search, it may find the board's circular, or it may find a tutoring site's guess from last year, and it will present either with the same tone. The only safe version of this question is: \"Find the official circular from the board's own website and give me the link.\" Then you read the circular.\n\nThat is not the tool failing. That is the tool doing exactly what it is: a very good guesser, sometimes with a web page in hand."},"check":{"question":"A teacher asks a chatbot for this year's exam timetable and gets a confident list of dates with no links or sources shown. What is the most likely explanation?","options":[{"text":"The model looked the timetable up on the exam board's website in the background and is reporting what it found there accurately.","correct":false,"tempting":"Assumes search happened when there is no evidence of it; a search-backed answer shows the pages it used, and an answer with no sources is generated from training text."},{"text":"The exam board published the timetable early enough to be in the model's training data, so the dates are accurate up to the cutoff.","correct":false,"tempting":"The current year's timetable almost certainly post-dates the training cutoff, and even if it did not, precise dates are rare specifics the model reconstructs rather than recalls."},{"text":"The model declined to search because exam timetables count as restricted information, and fell back on its own knowledge.","correct":false,"tempting":"Invents a refusal mechanism; models do not decline to search on this basis, and the answer was given, not withheld."},{"text":"The dates were generated from the pattern of previous years' timetables, and none of them should be trusted.","correct":true}]},"prev":{"slug":"where-it-is-reliable-and-where-it-is-not","title":"Where it is reliable, and where it is not","url":"https://addaly.com/learn/ai-for-teachers/where-it-is-reliable-and-where-it-is-not"},"next":{"slug":"the-conversation-has-a-memory-limit","title":"The conversation has a memory limit","url":"https://addaly.com/learn/ai-for-teachers/the-conversation-has-a-memory-limit"},"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) 'Why it cannot look it up', in AI for Teachers. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-teachers/why-it-cannot-look-it-up (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":"Why it cannot look it up","courseTitle":"AI for Teachers","module":{"index":1,"slug":"understanding-the-machine","title":"Understanding the machine you are about to use"},"answer":null,"keyIdea":"A model on its own has no way to consult anything at answer time; when a tool adds web search it fetches pages and summarises them, which fixes staleness but not judgement about which page to trust.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"type":"flow","steps":[{"label":"You type a question","detail":"The box looks identical whether or not a search follows."},{"label":"The tool writes a search query","detail":"Often not your question. A vague question makes a vague query."},{"label":"Pages are fetched and ranked","detail":"A confident content farm ranks alongside the exam board."},{"label":"The page text goes to the model","detail":"Along with your question, inside the same memory window."},{"label":"The model summarises","detail":"Exceptions and conditions fall out first; they are the least predictable words."},{"label":"You get an answer with links","detail":"The links are the gain. Open them."}],"title":"What actually happens when the tool searches","caption":"Six steps, and the model only enters at step four. Every judgement about which pages to trust is made by pattern, at step three and step five. That is why a searched answer is a research assistant’s note rather than a source."},"alt":"Figure: What actually happens when the tool searches. Steps: 1. You type a question (The box looks identical whether or not a search follows.); 2. The tool writes a search query (Often not your question. A vague question makes a vague query.); 3. Pages are fetched and ranked (A confident content farm ranks alongside the exam board.); 4. The page text goes to the model (Along with your question, inside the same memory window.); 5. The model summarises (Exceptions and conditions fall out first; they are the least predictable words.); 6. You get an answer with links (The links are the gain. Open them.). Six steps, and the model only enters at step four. Every judgement about which pages to trust is made by pattern, at step three and step five. That is why a searched answer is a research assistant’s note rather than a source."}},"blocks":[{"t":"h2","id":"two-very-different-things-wearing-the-same-interface","runs":[{"text":"Two very different things wearing the same interface"}]},{"t":"p","runs":[{"text":"When you type a question into a chatbot, one of two things happens, and the box looks the same either way."}]},{"t":"p","runs":[{"text":"Mode one: the model answers from what it learned.","b":true},{"text":" Nothing is consulted. No page is opened. The answer is produced word by word from the patterns absorbed during training, which finished months ago. This is what a language model is. It cannot check, because there is nothing to check against: the training text is not stored as a library it can open, it is dissolved into billions of numerical weights."}]},{"t":"p","runs":[{"text":"Mode two: the tool searches first.","b":true},{"text":" Many products now bolt a web search onto the model. Your question is turned into a search, some pages are fetched, and the model is handed their text along with your question. It then writes an answer that summarises what it was given. This is a genuine improvement for anything recent or specific, and it changes what you should worry about."}]},{"t":"h2","id":"how-to-tell-which-happened","runs":[{"text":"How to tell which happened"}]},{"t":"p","runs":[{"text":"Look for sources. A searched answer shows links, citations, or a line like “Searched the web”. An unsearched answer shows nothing. If you see confident specifics and no sources, you are in mode one, and the specifics are reconstructed from old patterns."}]},{"t":"p","runs":[{"text":"Most free tiers search automatically for questions that look like they need it, and not otherwise. You can force it: “Search the web before answering, and show me the pages you used.” You can also forbid it, which is useful when you want the model’s general explanation rather than a summary of whatever ranked first on the web this morning."}]},{"t":"h2","id":"what-search-fixes","runs":[{"text":"What search fixes"}]},{"t":"ul","items":[[{"text":"Staleness.","b":true},{"text":" The training cutoff stops mattering for the question at hand. The new syllabus, this year’s dates, the latest results are reachable."}],[{"text":"Local specifics, sometimes.","b":true},{"text":" If your district’s population is on a page the search finds, you get it. If no page has it, you are back to guessing, and the model may not tell you that it found nothing useful."}],[{"text":"Checkability.","b":true},{"text":" You now have a page to open. This is the real gain. Open it."}]]},{"t":"h2","id":"what-search-does-not-fix","runs":[{"text":"What search does not fix"}]},{"t":"p","runs":[{"text":"The model still has to decide which pages to trust, and it decides the way it decides everything: by pattern. A confident, well-formatted page ranks with an accurate one. A content farm that copied an old syllabus looks like the exam board. If three pages disagree, the model may blend them into a fourth answer none of them said."}]},{"t":"p","runs":[{"text":"It also summarises, and summaries lose things. The exceptions, the footnote that says “for the 2024 cohort only”, the conditions: these fall out first, because they are the least predictable part of a sentence."}]},{"t":"p","runs":[{"text":"And the search itself can go wrong. A vague question produces a vague search, which fetches irrelevant pages, which produce an answer built on the wrong material. When an answer seems oddly off, look at what it searched for. Often the search query was not your question."}]},{"t":"h2","id":"the-teacher-s-rule","runs":[{"text":"The teacher’s rule"}]},{"t":"p","runs":[{"text":"For anything that must be right (a date, a figure, a rule from the board) the model’s job is to find the page, and your job is to read it. Treat a searched answer as a research assistant’s note: “I think it’s on this page, and it seems to say this.” Then open the page."}]},{"t":"p","runs":[{"text":"For anything conceptual (an explanation, an analogy, a set of likely misconceptions) search usually makes the answer worse, because it replaces the model’s broad synthesis with a summary of two websites. Turn it off for those."}]},{"t":"h2","id":"an-example-that-comes-up-every-year","runs":[{"text":"An example that comes up every year"}]},{"t":"p","runs":[{"text":"A teacher asks for the marks allocation in this year’s Grade 12 physics paper. Without search, the model produces a plausible table from previous years. With search, it may find the board’s circular, or it may find a tutoring site’s guess from last year, and it will present either with the same tone. The only safe version of this question is: “Find the official circular from the board’s own website and give me the link.” Then you read the circular."}]},{"t":"p","runs":[{"text":"That is not the tool failing. That is the tool doing exactly what it is: a very good guesser, sometimes with a web page in hand."}]}],"check":{"question":[{"text":"A teacher asks a chatbot for this year’s exam timetable and gets a confident list of dates with no links or sources shown. What is the most likely explanation?"}],"options":[{"text":[{"text":"The model looked the timetable up on the exam board’s website in the background and is reporting what it found there accurately."}],"correct":false,"tempting":[{"text":"Assumes search happened when there is no evidence of it; a search-backed answer shows the pages it used, and an answer with no sources is generated from training text."}]},{"text":[{"text":"The exam board published the timetable early enough to be in the model’s training data, so the dates are accurate up to the cutoff."}],"correct":false,"tempting":[{"text":"The current year’s timetable almost certainly post-dates the training cutoff, and even if it did not, precise dates are rare specifics the model reconstructs rather than recalls."}]},{"text":[{"text":"The model declined to search because exam timetables count as restricted information, and fell back on its own knowledge."}],"correct":false,"tempting":[{"text":"Invents a refusal mechanism; models do not decline to search on this basis, and the answer was given, not withheld."}]},{"text":[{"text":"The dates were generated from the pattern of previous years’ timetables, and none of them should be trusted."}],"correct":true,"tempting":null}],"order":[1,0,2,3],"why":null},"prev":{"slug":"where-it-is-reliable-and-where-it-is-not","title":"Where it is reliable, and where it is not","minutes":9},"next":{"slug":"the-conversation-has-a-memory-limit","title":"The conversation has a memory limit","minutes":8}}}