# Build your "how I like work done" file: prompt chain Answer a few questions about how you work and what you like, turn the answers into a short list of rules for an AI tool, test each rule on a sample task, and keep a file you can paste at the start of any chat. **Made for:** Anyone who uses an AI tool at work and wants its answers to fit how they like things done, without technical setup. ## Before you start Have these ready: - **What you do and who you write for** (required): Describe your job and the people you usually write or prepare things for. - **How you like writing to sound** (required): Tone, length, and level of detail. Words you like and words you never use. - **How you like answers laid out** (required): For example short paragraphs, bullets, tables, or the answer first and the reasons after. - **What annoys you in AI answers** (required): List the habits you want it to stop. Be specific. - **What it should always do** (optional): Things you want every time, such as asking when something is unclear or showing its sources. - **A real task you do often** (required): One typical job you give an AI tool. It is used to test the rules. **How to use this chain** 1. Open a new chat in any AI assistant. 2. Copy the Step 1 prompt, replace every [PASTE HERE: ...] slot with your material, and send it. 3. Keep each answer: later steps ask you to paste it in. Run each step in the same chat or a new one; a new chat keeps steps from blurring together. 4. After a step that has a check, run the check before you move on. ## Step 1: Turn your answers into rules Write a short list of clear rules for an AI tool, each with the reason you gave and a way to test it. Copy this prompt: ````text You are an expert at writing clear working instructions for an AI assistant. Your task is to turn the person's answers below into a list of plain rules that an AI assistant can follow. What they do and who they write for: [PASTE HERE: What you do and who you write for] How they like writing to sound: [PASTE HERE: How you like writing to sound] How they like answers laid out: [PASTE HERE: How you like answers laid out] What annoys them: [PASTE HERE: What annoys you in AI answers] What it should always do: [PASTE HERE: What it should always do] Rules for you: - Write 6 to 12 rules. Each rule is one instruction, written so it can be followed or broken, not a wish. "Put the answer in the first sentence" is a rule. "Be clear" is not. - Use the person's own words where you can. Do not add preferences they did not give. - Give each rule a short reason taken from their answers. - For each rule, write a "Test" that says what you would see in an answer that follows it. - Do not write the final file yet. Return one markdown table and nothing else: | # | Rule | Reason | Test | ```` **You get back:** A table of 6 to 12 rules, each with a reason and a test. **Check after Step 1: Every rule has a reason and a test** Check by eye: Every row of the table has something written in "Rule", "Reason" and "Test" (a blank or a dash does not count). If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. ## Step 2: Try each rule on your sample task Do the sample task while following the rules, then check each rule against the result. Copy this prompt: ````text You are an assistant who follows written working rules exactly, and then checks your own work. Your task is to do the sample task below while following every rule, and then check whether each rule was followed. The sample task: [PASTE HERE: A real task you do often] Background about the person: [PASTE HERE: What you do and who you write for] The rules, with their tests: [PASTE HERE: the answer from Step 1 (Turn your answers into rules)] Rules for you: - Do the task first, following every rule. Then check each rule against what you wrote, using its test. - Be honest. If a rule was not followed or could not be tested on this task, say so. - Do not invent facts for the task. Use a placeholder such as [name] where a detail is missing. Return in this shape: The result of the task: (the text you wrote) Then one markdown table: | # | Rule | Followed | What I saw | The Followed column must say exactly yes or no. ```` **You get back:** The sample task done with the rules, and a table that says whether each rule was followed. **Check after Step 2: Each rule is marked yes or no** Check by eye: Every row of the table has one of "yes" or "no" in the Followed column. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. ## Step 3: Fix the rules that failed Reword any rule that was not followed so that it is easier to follow, and keep the ones that worked. Copy this prompt: ````text You are an expert at writing working instructions that an AI assistant follows reliably. Your task is to improve the rules that were not followed in the test. The rules: [PASTE HERE: the answer from Step 1 (Turn your answers into rules)] The test result: [PASTE HERE: the answer from Step 2 (Try each rule on your sample task)] Rules for you: - Keep every rule that was followed, word for word. - For a rule that was not followed, reword it so it is more specific, add an example of what to do, or split it in two. Explain in a few words what you changed. - Do not add new preferences and do not remove a rule the person asked for. - If every rule was followed, say so and change nothing. Return one markdown table and nothing else: | # | Rule | Changed | The Changed column must say exactly yes or no, and the Rule column holds the final wording. ```` **You get back:** The final list of rules, with the reworded ones marked. **Check after Step 3: Every row says if it changed** Check by eye: Every row of the table has one of "yes" or "no" in the Changed column. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. ## Step 4: Write the file you paste into your AI tool Put the final rules into one short file, written to the AI, that you can paste at the start of any chat. Copy this prompt: ````text You are an expert at writing the short instructions people paste at the start of an AI chat. Your task is to write the final "how I like work done" file from the rules below. What they do and who they write for: [PASTE HERE: What you do and who you write for] The final rules: [PASTE HERE: the answer from Step 3 (Fix the rules that failed)] Rules for you: - Write it to the AI, in the second person ("You are helping...", "Put the answer first"). - Open with two sentences on who the person is and who they write for. - Group the rules under short headings, such as Tone, Layout and Always do, and Never do. - Keep every rule from the final list. Do not add rules and do not add facts about the person. - Keep it to 350 words or fewer. Do not use jargon. Return the file as plain text and nothing else. ```` **You get back:** A short file of instructions, ready to paste at the start of any AI chat. **Check after Step 4: The file is 350 words or fewer** Check by eye: The answer is no longer than 350 words. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. ## Final checks Run these on the finished result. **Final check 1: Every rule is in the file** Open a new chat (so the check is not swayed by the work) and paste this check prompt: ````text You are a careful reviewer. You do not fix or rewrite the work; you only decide whether it passes one check. The check: Every rule is in the file Compare the file with the final rules below. Is every rule in the table present in the file, with the same meaning, and does the file add no rule that is not in the table? The final rules: [PASTE HERE: the answer from Step 3 (Fix the rules that failed)] The work to check (the answer from Step 4 (Write the file you paste into your AI tool)): [PASTE HERE: the answer from Step 4 (Write the file you paste into your AI tool)] Reply with exactly PASS or FAIL on the first line. On the second line, say in one sentence why. If it fails, name the exact item or line that fails. Do not add anything else. ```` If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 2: Each rule was tested on the sample** Check by eye: Every row of the table has something written in "Followed" and "What I saw" (a blank or a dash does not count). If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 3: The rules sound like you** Ask yourself: Read the file. Does each rule say something you really want? Cross out any you would not miss, then try the file in a new chat on a real task and see if the answer feels right. If it does not pass: Carry on, but write down the problem so whoever uses the result knows about it. ## What you end up with - A short file of rules to paste at the start of any AI chat, or to save in your tool's settings. - The table of rules with the reason for each, in case you want to change them later. ## Changing this chain - Keep a separate file for each kind of work (for example one for client emails and one for internal notes) by running the template again with different answers. - After a week of use, add whatever you keep correcting to "What annoys you" and run the template again. - Some AI tools have a settings place for standing instructions. Paste the file there so you do not have to paste it each time. - To change the word limit, change 350 in the last step's rules and in its check, both. --- From Logic Lab (logiclabhq.com). Free to use and adapt, with credit.