Prompt patterns.

The first twelve are about what you put in the prompt. The second twelve are about how you work with the model once it answers. Start anywhere: each page stands on its own and takes a few minutes.

Writing the ask

1. 01State the deliverableName the thing you want back, such as a reply, a ranked list or three sentences for a report, instead of asking the model to look at something.5 min
2. 02Name the readerSay who will read the result, what they already know and what they will do with it, so the model picks the right depth, terms and order.5 min
3. 03Give it the factsPaste the facts only you have, such as numbers, names, dates and decisions, so the model works from them instead of inventing plausible ones.5 min
4. 04Set a lengthGive the answer a size to fit, in words, items, minutes or pages. A size is also a priority, because it makes the model choose what matters.5 min
5. 05Ask for a formatName the shape of the answer, such as a table with set columns, bullets, sections or an email with a subject line, so it fits where it is going.5 min
6. 06Say what to avoidName what to leave out, such as topics, claims, words or sources, and give the reason, so the model can apply the limit to cases you did not list.5 min
7. 07Describe a good resultSay what a good answer must do, in terms someone could check, so the model aims at it and you can test the result against it.5 min
8. 08Show one examplePaste one sample of what you want, such as last week's update or an approved reply, and ask the model to match its shape with new content.5 min
9. 09Examples and counterexamplesShow one sample that hits the mark and one that misses, and say what separates them, so the model learns the line you care about.5 min
10. 10What to do when unsureTell the model to ask, flag or leave a blank when a fact is missing or unclear, so gaps show up instead of being filled with confident guesses.5 min
11. 11Fence off pasted textPut pasted material between clear markers, such as triple quotes or named tags, so the model reads it as material to work on, not as instructions.5 min
12. 12Put long documents firstPaste long material at the top and put the question at the end, opened with a phrase that points back to the material.5 min

Working with the model

1. 13Ask it to ask questions firstHave the model interview you before it writes, so the first draft is built on your facts instead of its best guesses.7 min
2. 14Plan, then writeAsk for an outline first, fix the structure while it is cheap to change, then have the model draft from the version you agreed.7 min
3. 15Draft, then critiqueHave the model check a draft against criteria you name, line by line, and fix only what fails, instead of asking it to make the draft better.7 min
4. 16Split a big jobBreak one huge request into a chain of smaller prompts, and check each result before it feeds the next step.6 min
5. 17Compare options in a tableLay a decision out as options against criteria you name, so every option answers the same questions and the gaps show.6 min
6. 18Ask for reasoning firstAsk for the calculation or the evidence before the conclusion when you need to check it, and skip it where it only adds length.7 min
7. 19Start the answer for itGive the model the first line or the structure its answer must begin with, so the opening says what you need and the rest follows.6 min
8. 20Extract to JSONAsk for JSON with exact keys, types and null for anything missing, so a spreadsheet import, automation or script can read the output as it is.6 min
9. 21Ask for three versionsAsk for three versions that each take a different approach you name, labeled, then pick one instead of editing a single draft toward an unknown target.6 min
10. 22Turn a chat into a promptAfter a good back-and-forth, ask the model to write the prompt that would get the final answer in one go, then test it and save it.6 min
11. 23Have AI write the promptAsk a model to draft or improve a prompt, give it the facts only you have, and check the result for invented details and padding before you use it.6 min
12. 24When a role helpsA role line sets a voice and a lens but adds no knowledge. Say who the reader is and which checks an expert would make instead.7 min

Want them in order, with exercises? Take the free prompt course. Something already went wrong? Find the fix.