{"course":{"slug":"ai-for-small-business","title":"AI for a Small Business","url":"https://addaly.com/learn/ai-for-small-business","level":"beginner","track":"work"},"lesson":{"slug":"a-catalogue-of-two-hundred","title":"Doing two hundred at once, and knowing how many to check","url":"https://addaly.com/learn/ai-for-small-business/a-catalogue-of-two-hundred","position":43,"of":88,"minutes":10,"answer":"Batch generation is safe when the facts come from a spreadsheet, missing fields are flagged rather than filled, and you know how many rows you checked.","keyIdea":"Batch generation is safe when the facts come from a spreadsheet, missing fields are flagged rather than filled, and you know how many rows you checked, because zero errors in twenty samples still allows a true error rate near 15%, and in fifty it is about 6%."},"body":{"format":"markdown","text":"## The workflow\n\nThe technique is from module two: write five yourself, then feed the raw facts for the rest. What changes at scale is the plumbing and the checking.\n\n**1. Get the facts into a spreadsheet.** One row per product, one column per attribute: material, dimensions, weight, origin, price, what it is for, who should not buy it. This is the work. If the spreadsheet is thin, the descriptions will be empty, and no amount of prompting fixes an empty row.\n\n**2. Write five descriptions by hand.** Properly, in your voice, using the same columns. These are the pattern.\n\n**3. Run in batches of fifteen to twenty.** Not two hundred at once. Long outputs get truncated, quality drifts across a very long response, and a batch of twenty is a checking session you will actually finish.\n\n**4. Add two columns: `checked` and `notes`.** Empty until a person has read the row.\n\n**5. Publish only checked rows.** This is the rule that makes the whole thing safe, and the one that gets abandoned under deadline pressure.\n\n## The prompt\n\n```text\nBelow are five descriptions I wrote (EXAMPLES) showing\nstyle only - do not reuse their content.\n\nThen a table of product facts. For each product write a\ndescription of 60-80 words in that style.\n\nRules:\n- Use ONLY the facts given for that product.\n- Never invent dimensions, materials, origins or care\n  instructions.\n- If a needed fact is missing, write [MISSING: field].\n- No superlatives you cannot support.\n- Output: product code, then description. Nothing else.\n```\n\nThe `[MISSING]` rule is what turns a hundred silent inventions into a hundred visible gaps you can fill from the stockroom.\n\n## How many to check\n\nThis is the part everyone guesses at, and there is a simple piece of arithmetic that answers it.\n\nSuppose you check twenty descriptions at random and every one is fine. What does that tell you about the other 180?\n\n> Figure: What a clean sample actually proves. Line chart of highest true error rate still plausible (%) against descriptions checked, every one correct: Three divided by the number checked (10, 30) (20, 15) (30, 10) (50, 6) (100, 3) (200, 1.5). Twenty clean samples out of two hundred still allows about thirty bad descriptions in the catalogue. Where you stop on this curve is set by what a wrong dimension costs you, and the discipline is simply being able to say which point you chose.\n\nThere is a rule of thumb statisticians call the **rule of three**: if you see no failures in `n` samples, the true failure rate could still plausibly be as high as `3/n`. Twenty clean samples means the real rate could be up to about 15%. That is thirty bad descriptions in your catalogue.\n\nCheck fifty and find none, and the ceiling drops to about 6%. Check a hundred, about 3%.\n\nSo the answer depends on what a bad description costs you. For decorative copy on low-value items, twenty is fine. For anything where a wrong dimension causes a return, or a wrong material causes an allergy problem, you are checking far more, and for safety-relevant fields, every single one.\n\nThe useful discipline is to know which you have chosen. \"I checked twenty and found none, so the error rate is probably under 15%\" is a defensible sentence. \"I spot-checked a few and they looked fine\" is not a statement about anything.\n\n## Check the fields, not the prose\n\nRead a batch in two passes, as in module one. First pass across all twenty: every dimension, material, price and care instruction against the spreadsheet, ignoring the writing entirely. Second pass: read three or four as a customer would.\n\nErrors cluster by field rather than by product. If one dimension came out wrong, check that column everywhere: the failure is usually in how the column was labelled or formatted, not in the individual row.\n\n## Marketplaces\n\nAmazon, Etsy, Jumia, Mercado Libre and the rest do not prohibit AI-written listings. Two things are worth knowing anyway.\n\nListings that all read alike compete badly, because shoppers skim and there is nothing to catch on. Your specifics are the defence, which is another reason the spreadsheet matters more than the prompt.\n\nAnd every platform prohibits inaccurate attributes. A generated care instruction that turns out to be wrong is not a writing problem; it is a listing violation and a return.\n\n## What to do when you find errors\n\nNot fix the rows. Find the cause.\n\nErrors in generated batches almost never scatter randomly. They come from one of four places: a column whose header was ambiguous, a product whose facts were missing so the model filled the gap, a formatting inconsistency such as dimensions written three different ways, or an instruction the prompt did not contain.\n\nFix the cause and re-run the affected batch. Fixing thirty rows by hand leaves the cause in place for the next two hundred.\n\n## Where the time actually goes\n\nAn owner budgeting for this exercise usually plans for the generation and is surprised by the split. In practice, on a catalogue of two hundred: roughly two hours assembling and cleaning the spreadsheet, forty minutes writing and testing the prompt on five products, twenty minutes of generation, and two to three hours of checking.\n\nThe model accounts for a tenth of the work. That ratio is worth knowing before you promise somebody a catalogue by Friday, and it is why the spreadsheet is the project rather than the prompt."},"check":{"question":"A shop generates 300 product descriptions, checks 25 at random, finds no errors, and publishes all 300. What can it legitimately conclude?","options":[{"text":"That the error rate is close to zero, since a random sample of 25 with no failures is representative of the whole batch.","correct":false,"tempting":"Treats an observed zero as a measured zero, when the sample is simply too small to rule out a rate that would put dozens of bad rows in the catalogue."},{"text":"That the process is sound and any remaining errors will be reported by customers, which is an acceptable trade for the time saved on a catalogue of this size.","correct":false,"tempting":"Outsources detection to customers, which works only for errors customers notice and pay for, and treats returns as a cheaper form of checking than reading."},{"text":"Nothing, because a sample drawn at random cannot represent a batch in which errors cluster by field rather than being distributed evenly across products.","correct":false,"tempting":"Correctly notes clustering and overstates it into uselessness; a random sample still bounds the overall rate, and checking by field is a refinement rather than a replacement."},{"text":"That the true error rate is probably below about 12%, which on 300 items still allows for roughly 36 bad descriptions.","correct":true}]},"prev":{"slug":"generated-images-and-copyright","title":"Generated images: what is unsettled, and the three rules that are not","url":"https://addaly.com/learn/ai-for-small-business/generated-images-and-copyright"},"next":{"slug":"being-found","title":"Being found, now that search engines answer questions themselves","url":"https://addaly.com/learn/ai-for-small-business/being-found"},"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) 'Doing two hundred at once, and knowing how many to check', in AI for a Small Business. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-small-business/a-catalogue-of-two-hundred (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":"Doing two hundred at once, and knowing how many to check","courseTitle":"AI for a Small Business","module":{"index":5,"slug":"selling-and-being-found","title":"Selling: listings, images, marketing and price"},"answer":null,"keyIdea":"Batch generation is safe when the facts come from a spreadsheet, missing fields are flagged rather than filled, and you know how many rows you checked, because zero errors in twenty samples still allows a true error rate near 15%, and in fifty it is about 6%.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"type":"line","title":"What a clean sample actually proves","series":[{"label":"Three divided by the number checked","points":[[10,30],[20,15],[30,10],[50,6],[100,3],[200,1.5]]}],"xLabel":"descriptions checked, every one correct","yLabel":"highest true error rate still plausible (%)","caption":"Twenty clean samples out of two hundred still allows about thirty bad descriptions in the catalogue. Where you stop on this curve is set by what a wrong dimension costs you, and the discipline is simply being able to say which point you chose."},"alt":"Figure: What a clean sample actually proves. Line chart of highest true error rate still plausible (%) against descriptions checked, every one correct: Three divided by the number checked (10, 30) (20, 15) (30, 10) (50, 6) (100, 3) (200, 1.5). Twenty clean samples out of two hundred still allows about thirty bad descriptions in the catalogue. Where you stop on this curve is set by what a wrong dimension costs you, and the discipline is simply being able to say which point you chose."}},"blocks":[{"t":"h2","id":"the-workflow","runs":[{"text":"The workflow"}]},{"t":"p","runs":[{"text":"The technique is from module two: write five yourself, then feed the raw facts for the rest. What changes at scale is the plumbing and the checking."}]},{"t":"p","runs":[{"text":"1. Get the facts into a spreadsheet.","b":true},{"text":" One row per product, one column per attribute: material, dimensions, weight, origin, price, what it is for, who should not buy it. This is the work. If the spreadsheet is thin, the descriptions will be empty, and no amount of prompting fixes an empty row."}]},{"t":"p","runs":[{"text":"2. Write five descriptions by hand.","b":true},{"text":" Properly, in your voice, using the same columns. These are the pattern."}]},{"t":"p","runs":[{"text":"3. Run in batches of fifteen to twenty.","b":true},{"text":" Not two hundred at once. Long outputs get truncated, quality drifts across a very long response, and a batch of twenty is a checking session you will actually finish."}]},{"t":"p","runs":[{"text":"4. Add two columns: ","b":true},{"text":"checked","b":true,"code":true},{"text":" and ","b":true},{"text":"notes","b":true,"code":true},{"text":".","b":true},{"text":" Empty until a person has read the row."}]},{"t":"p","runs":[{"text":"5. Publish only checked rows.","b":true},{"text":" This is the rule that makes the whole thing safe, and the one that gets abandoned under deadline pressure."}]},{"t":"h2","id":"the-prompt","runs":[{"text":"The prompt"}]},{"t":"code","lang":"text","label":"Text","text":"Below are five descriptions I wrote (EXAMPLES) showing\nstyle only - do not reuse their content.\n\nThen a table of product facts. For each product write a\ndescription of 60-80 words in that style.\n\nRules:\n- Use ONLY the facts given for that product.\n- Never invent dimensions, materials, origins or care\n  instructions.\n- If a needed fact is missing, write [MISSING: field].\n- No superlatives you cannot support.\n- Output: product code, then description. Nothing else."},{"t":"p","runs":[{"text":"The "},{"text":"[MISSING]","code":true},{"text":" rule is what turns a hundred silent inventions into a hundred visible gaps you can fill from the stockroom."}]},{"t":"h2","id":"how-many-to-check","runs":[{"text":"How many to check"}]},{"t":"p","runs":[{"text":"This is the part everyone guesses at, and there is a simple piece of arithmetic that answers it."}]},{"t":"p","runs":[{"text":"Suppose you check twenty descriptions at random and every one is fine. What does that tell you about the other 180?"}]},{"t":"p","runs":[{"text":"There is a rule of thumb statisticians call the "},{"text":"rule of three","b":true},{"text":": if you see no failures in "},{"text":"n","code":true},{"text":" samples, the true failure rate could still plausibly be as high as "},{"text":"3/n","code":true},{"text":". Twenty clean samples means the real rate could be up to about 15%. That is thirty bad descriptions in your catalogue."}]},{"t":"p","runs":[{"text":"Check fifty and find none, and the ceiling drops to about 6%. Check a hundred, about 3%."}]},{"t":"p","runs":[{"text":"So the answer depends on what a bad description costs you. For decorative copy on low-value items, twenty is fine. For anything where a wrong dimension causes a return, or a wrong material causes an allergy problem, you are checking far more, and for safety-relevant fields, every single one."}]},{"t":"p","runs":[{"text":"The useful discipline is to know which you have chosen. “I checked twenty and found none, so the error rate is probably under 15%” is a defensible sentence. “I spot-checked a few and they looked fine” is not a statement about anything."}]},{"t":"h2","id":"check-the-fields-not-the-prose","runs":[{"text":"Check the fields, not the prose"}]},{"t":"p","runs":[{"text":"Read a batch in two passes, as in module one. First pass across all twenty: every dimension, material, price and care instruction against the spreadsheet, ignoring the writing entirely. Second pass: read three or four as a customer would."}]},{"t":"p","runs":[{"text":"Errors cluster by field rather than by product. If one dimension came out wrong, check that column everywhere: the failure is usually in how the column was labelled or formatted, not in the individual row."}]},{"t":"h2","id":"marketplaces","runs":[{"text":"Marketplaces"}]},{"t":"p","runs":[{"text":"Amazon, Etsy, Jumia, Mercado Libre and the rest do not prohibit AI-written listings. Two things are worth knowing anyway."}]},{"t":"p","runs":[{"text":"Listings that all read alike compete badly, because shoppers skim and there is nothing to catch on. Your specifics are the defence, which is another reason the spreadsheet matters more than the prompt."}]},{"t":"p","runs":[{"text":"And every platform prohibits inaccurate attributes. A generated care instruction that turns out to be wrong is not a writing problem; it is a listing violation and a return."}]},{"t":"h2","id":"what-to-do-when-you-find-errors","runs":[{"text":"What to do when you find errors"}]},{"t":"p","runs":[{"text":"Not fix the rows. Find the cause."}]},{"t":"p","runs":[{"text":"Errors in generated batches almost never scatter randomly. They come from one of four places: a column whose header was ambiguous, a product whose facts were missing so the model filled the gap, a formatting inconsistency such as dimensions written three different ways, or an instruction the prompt did not contain."}]},{"t":"p","runs":[{"text":"Fix the cause and re-run the affected batch. Fixing thirty rows by hand leaves the cause in place for the next two hundred."}]},{"t":"h2","id":"where-the-time-actually-goes","runs":[{"text":"Where the time actually goes"}]},{"t":"p","runs":[{"text":"An owner budgeting for this exercise usually plans for the generation and is surprised by the split. In practice, on a catalogue of two hundred: roughly two hours assembling and cleaning the spreadsheet, forty minutes writing and testing the prompt on five products, twenty minutes of generation, and two to three hours of checking."}]},{"t":"p","runs":[{"text":"The model accounts for a tenth of the work. That ratio is worth knowing before you promise somebody a catalogue by Friday, and it is why the spreadsheet is the project rather than the prompt."}]}],"check":{"question":[{"text":"A shop generates 300 product descriptions, checks 25 at random, finds no errors, and publishes all 300. What can it legitimately conclude?"}],"options":[{"text":[{"text":"That the error rate is close to zero, since a random sample of 25 with no failures is representative of the whole batch."}],"correct":false,"tempting":[{"text":"Treats an observed zero as a measured zero, when the sample is simply too small to rule out a rate that would put dozens of bad rows in the catalogue."}]},{"text":[{"text":"That the process is sound and any remaining errors will be reported by customers, which is an acceptable trade for the time saved on a catalogue of this size."}],"correct":false,"tempting":[{"text":"Outsources detection to customers, which works only for errors customers notice and pay for, and treats returns as a cheaper form of checking than reading."}]},{"text":[{"text":"Nothing, because a sample drawn at random cannot represent a batch in which errors cluster by field rather than being distributed evenly across products."}],"correct":false,"tempting":[{"text":"Correctly notes clustering and overstates it into uselessness; a random sample still bounds the overall rate, and checking by field is a refinement rather than a replacement."}]},{"text":[{"text":"That the true error rate is probably below about 12%, which on 300 items still allows for roughly 36 bad descriptions."}],"correct":true,"tempting":null}],"order":[0,3,2,1],"why":null},"prev":{"slug":"generated-images-and-copyright","title":"Generated images: what is unsettled, and the three rules that are not","minutes":10},"next":{"slug":"being-found","title":"Being found, now that search engines answer questions themselves","minutes":9}}}