--- title: "Turn repeated AI formatting instructions into rules the model reads every session" description: "Move the format and voice rules you retype in every AI chat into a file the model loads each session, then check each output against them." canonical: "https://scalewithsearch.com/articles/ai-doesnt-follow-my-instructions" date: "2026-01-28" modified: "2026-09-25" --- ## Site navigation - [Scale With Search](https://scalewithsearch.com/) - Real estate - Real estate - [Real estate](https://scalewithsearch.com/for/real-estate) - Work - Start here - [Send your brief](https://scalewithsearch.com/work#send-your-brief) - [Prepare your six-question brief](https://scalewithsearch.com/work#prepare-your-six-question-brief) - Build - [Site build, content library with SEO, signal desk](https://scalewithsearch.com/work) - For your business - Trades and home services - [Auto body and collision shops](https://scalewithsearch.com/for/auto-body-and-collision-shops) - [Foundation and home repair contractors](https://scalewithsearch.com/for/foundation-and-home-repair) - [Garage door and fencing contractors](https://scalewithsearch.com/for/garage-door-and-fencing-contractors) - [HVAC contractors](https://scalewithsearch.com/for/hvac-contractors) - [Janitorial and commercial cleaning companies](https://scalewithsearch.com/for/janitorial-and-commercial-cleaning) - [Locksmiths](https://scalewithsearch.com/for/locksmiths) - [Moving companies](https://scalewithsearch.com/for/moving-companies) - [Pest control companies](https://scalewithsearch.com/for/pest-control-companies) - [Plumbing and electrical contractors](https://scalewithsearch.com/for/plumbing-and-electrical-contractors) - [Restoration and water or fire damage companies](https://scalewithsearch.com/for/restoration-and-water-fire-damage) - [Roofing companies](https://scalewithsearch.com/for/roofing-companies) - [Towing companies](https://scalewithsearch.com/for/towing-companies) - [Tree services and landscaping companies](https://scalewithsearch.com/for/tree-services-and-landscaping) - [Solar installers](https://scalewithsearch.com/for/solar-installation) - [General contractors](https://scalewithsearch.com/for/general-contractors-and-construction) - [Paving, concrete, and flooring contractors](https://scalewithsearch.com/for/paving) - Practices and professional services - [Bookkeeping and tax practices](https://scalewithsearch.com/for/bookkeeping-and-tax-practices) - [Dental practices](https://scalewithsearch.com/for/dental-practices) - [Family and criminal defense law firms](https://scalewithsearch.com/for/family-and-criminal-defense-law-firms) - [Med spas and aesthetics practices](https://scalewithsearch.com/for/med-spas-and-aesthetics) - [Personal injury law firms](https://scalewithsearch.com/for/personal-injury-law-firms) - [Veterinary clinics](https://scalewithsearch.com/for/veterinary-clinics) - [Gyms and fitness studios](https://scalewithsearch.com/for/fitness) - [Therapy and outpatient health practices](https://scalewithsearch.com/for/therapy-and-outpatient-health) - [Medical billing companies](https://scalewithsearch.com/for/medical-billing) - [Insurance agencies](https://scalewithsearch.com/for/insurance-agencies) - [Financial advisors](https://scalewithsearch.com/for/financial-advisors) - [Property management companies](https://scalewithsearch.com/for/property-management) - [Recruiting and staffing agencies](https://scalewithsearch.com/for/recruiting-and-staffing) - [Architects and interior designers](https://scalewithsearch.com/for/architects-and-interior-designers) - [Logistics and supply chain companies](https://scalewithsearch.com/for/logistics-and-supply-chain) - Agencies, MSPs, and manufacturing - [IT and managed service providers](https://scalewithsearch.com/for/it-and-managed-service-providers) - [Machine shops and precision manufacturers](https://scalewithsearch.com/for/machine-shops-and-precision-manufacturing) - [Marketing agencies and freelancers](https://scalewithsearch.com/for/marketing-agencies-and-freelancers) - [SEO agencies and consultants](https://scalewithsearch.com/for/seo-agencies-and-consultants) - [Small manufacturers and fabricators](https://scalewithsearch.com/for/small-manufacturers-and-fabricators) - Restaurants, shops, studios, and nonprofits - [Restaurants and hospitality businesses](https://scalewithsearch.com/for/restaurants-and-hospitality) - [Retail stores and ecommerce sellers](https://scalewithsearch.com/for/retail-and-ecommerce) - [Photographers, event planners, and travel agents](https://scalewithsearch.com/for/photographers) - [Churches and nonprofits](https://scalewithsearch.com/for/churches-and-nonprofits) - [All industries](https://scalewithsearch.com/for/) - Learn - For your office - [Office job guides](https://scalewithsearch.com/guides/) - [Browser calculators](https://scalewithsearch.com/tools/) - Start here - [How it works](https://scalewithsearch.com/how-it-works) - [Free Starter Kit](https://scalewithsearch.com/kit/business-memory-starter-kit.zip) - [Synthetic specimen](https://scalewithsearch.com/specimen/working-session-specimen.zip) - Guides - [The Complete Guide to Business Memory for AI Agents](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [The Complete Small-Business Guide to AI Agent Governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [The Complete Guide to Leaving Vendor AI Memory](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - Articles by cluster - [Business memory](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [Agent governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [Migration and ownership](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - For machines - [llms.txt](https://scalewithsearch.com/llms.txt) - [llms-full.txt](https://scalewithsearch.com/llms-full.txt) - [Machine view](https://scalewithsearch.com/?view=machine) - Company - Evidence - [Proof](https://scalewithsearch.com/proof) - Company - [About](https://scalewithsearch.com/about) # Turn repeated AI formatting instructions into rules the model reads every session. You have told the assistant many times: sentence case for headings, not title case. Hyphens for bullets, not asterisks. Paragraphs of four sentences or fewer. A summary section at the top of every document. It agrees. Then it produces another document with title-case headings and asterisk bullets. No summary. Six-sentence paragraphs. The model is not stubborn. Instructions typed into a chat message apply to that chat. When the session ends, they end with it. The next conversation starts without them. ## Why chat instructions do not stick A chat product gives the model instructions from two kinds of place. Standing instructions load before your conversation. They include the provider's system prompt and, where you set them, your custom instructions or project instructions. They apply to every chat in their scope. Conversation messages exist only in the current session. When you type your format rules into a message, they live here. They shape the answers in that chat. They do not carry into tomorrow's chat. This creates an illusion of teaching. It feels like you train the assistant each time you correct it. In fact, you type the same rules into a new temporary context that will disappear again. ## What repeated instructions cost Count the time on one document in a worked example. You type the format rules: two minutes. The output ignores half of them, so you correct it and explain again: three minutes. The revision fixes some problems and adds new ones: two more minutes. That is seven minutes for a task that should take thirty seconds. At five documents a day, it is 35 minutes a day, close to three hours a week, and about 150 hours over a working year. The time is not the only cost. Each correction round also interrupts the task you meant to finish, and the interruption repeats on every AI-assisted task. ## Why product memory does not solve it ChatGPT has a memory feature. Claude has Projects and its own memory. Perplexity has Projects. Each one can carry some preferences across sessions. In practice, each captures fragments and misses structure. The [ChatGPT versus Perplexity memory test](/articles/chatgpt-vs-perplexity-memory-business) shows why research evidence and business memory need different records. A memory feature might save "prefers hyphen bullets" and miss the paragraph rule. A project might keep the summary rule and lose the heading rule. The model follows some rules and skips others, and you still correct the output. These features also stay inside one product. ChatGPT memory does not move to Claude. Claude project settings do not work in another tool. You maintain separate instruction sets in each tool, or you accept inconsistent output. The limits differ by product too; the guides to [ChatGPT project and custom instruction limits](/articles/chatgpt-custom-vs-project-instructions-limits) and [Claude custom and project instruction limits](/articles/claude-custom-project-instructions-limits) list them. ## Move the rules into a file A context file is a Markdown document that holds your actual instructions: format rules, output structure, tone, and required terms. You write the rules once, in a file outside the conversation. At the start of each session, the tool loads the file before your first prompt. Claude Code loads `CLAUDE.md` files at launch, as Anthropic's page [How Claude remembers your project](https://code.claude.com/docs/en/memory) describes. In a chat product, put the same text in project instructions or custom instructions. A format section can be this short: ```markdown ## Document format - Headings: sentence case. Capitalize only the first word and names. - Bullets: start each item with "- ". Never use "*". - Paragraphs: four sentences or fewer. - Every document starts with a "Summary" section of three lines. - Word count goes on the first line: "Words: 1,240". ``` On Monday, you paste raw content and write "Format this as a blog post." The file has loaded, and the output uses sentence-case headings, hyphen bullets, short paragraphs, and a summary at the top. On Tuesday, a new session loads the same file. You paste different content with the same request, and the output follows the same rules. You did not repeat them. The same pattern covers more than format. Voice rules, fact-check requirements, citation rules, and review steps all belong in the file when you repeat them across sessions. ## Build the instruction set from your own repeats Start with what you already repeat. 1. Open your recent AI conversations. 2. List every instruction you typed more than once. 3. Group the list into sections: format, voice, output structure, and domain terms. 4. Rewrite each item as a rule someone could check. 5. Put the sections into the context file. Step 4 matters most. "Use a professional tone" is vague, and the model can read it a different way each time. "No exclamation marks. Active voice. Sentences under 25 words." is specific. A person, or a script, can check whether an output meets it. Then keep one habit. When you type the same instruction for the third time, it goes into the file. If you ask three times in one week for word counts at the top of documents, that request is a rule. ## Check the output against the rules A file puts your rules in every session. It does not force the model to obey them. Anthropic's documentation states that Claude Code treats `CLAUDE.md` as context, not as enforced configuration. The same documentation notes that specific, concise instructions are followed more consistently, and that contradictory rules can lead the model to pick one arbitrarily. So add a check. Many format rules are mechanical, and a short script can test them: ```bash # Flag asterisk bullets and headings that are not sentence case grep -n '^\* ' draft.md grep -nE '^#+ .* [A-Z][a-z]+ [A-Z][a-z]+' draft.md ``` The second pattern is rough: it flags two capitalized words in a row, which also catches proper names. Use it as a prompt for review, not as a verdict. Give every rule a check method, even when the method is a person: | Rule | Check method | |---|---| | Hyphen bullets only | Search for lines that start with `* ` | | Summary section first | Confirm the first heading is "Summary" | | Paragraphs of four sentences or fewer | Count sentences in the longest paragraph | | Word count on the first line | Compare the stated count with a word counter | | Active voice | Reviewer reads the first two paragraphs | A rule without a check drifts, because nobody notices when the model stops following it. When a rule fails, look at the rule before you blame the model. Is it specific? Does it conflict with another rule? Is it buried in the middle of a long file? Fix the file, then run the same request in a fresh session. For rules with real consequences, such as "never send an email without approval," a file is not enough. Those rules need a control at the point where the action happens. The article [ChatGPT custom instructions are not hard constraints](/articles/chatgpt-custom-instructions-not-hard-constraints) shows how to build and test that control. ## Test the file in three fresh sessions Run this test before you rely on the file. 1. Close every open session. 2. Start three new sessions, one at a time, with the file loaded. 3. Give each session the same request and different raw content, such as "Format this as a blog post." 4. Run the check script, or the check table, on each output. 5. Record which rules failed in which session. The file passes when all three outputs follow every rule in the format section. A rule that fails in one session but not the others is written too loosely. A rule that fails in all three is missing, contradicted by another rule, or buried. Fix the file and run the three sessions again. Keep the test content, so you can rerun the same test after a product update. ## Keep one source for every tool A file also makes your rules portable. The same text works in Claude Code, in Claude or ChatGPT project instructions, and in any tool that accepts context at session start. You keep one source of truth, and each assistant reads from it. Date the file and note each change. When a rule stops working after a product update, the history shows what changed and when. The workflow in [how corrections become part of the system](/articles/corrections-become-system-memory) turns each fix into a rule the next run applies. ## Related: AI memory - [Move business context out of custom instructions and into files](/articles/why-custom-instructions-fail) - [Find where Claude Code /init stops and plan the memory it leaves out](/articles/claude-code-init-limitations) - [Turn each business document into a data feed your AI memory can trust](/articles/documents-as-data-feeds-ai-memory) ---- ```text .|########||. .|########||. .|########||. |##||. .||##|. |##||. .||##|. |##||. .||##|. |#|. .|#|. |#|. .|#|. |#|. .|#|. |#| |#| |#| |#| |#| |#| .#| |#. .#| |#. .#| |#. .#. .#| .#. .#| .#. .#| |#. .#| |#. .#| |#. .#| |# ...... #| |# ...... #| |# ...... #| .# ||#########| #| .# ||#########| #| .# ||#########| #| .#. |######||######|. .#. .#. |######||######|. .#. .#. |######||######|. .#. #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# # .||####|###########################|. # # .||####|###########################|. # # .||####|###########################|. # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || #. .######|#|#########| .# #. .######|#|#########| .# #. .######|#|#########| .# .#. |#####||||#####| .#. .#. |#####||||#####| .#. .#. |#####||||#####| .#. |# ||########|| #. |# ||########|| #. |# ||########|| #. |# ...... #| |# ...... #| |# ...... #| |#. .#| |#. .#| |#. .#| |#. .#. |#. .#. |#. .#. .#| |#. .#| |#. .#| |#. |#| |#| |#| |#| |#| |#| .|#|. .|#|. .|#|. .|#|. .|#|. .|#|. |##||. .||##| |##||. .||##| |##||. .||##| .||########||. .||########||. .||########||. Scale With Search 2026 [scalewithsearch.com](https://scalewithsearch.com) ```