--- title: "Knowledge in GPTs: Memory, Instructions, and Files." description: "Use knowledge in GPTs with maintained source files. Separate instructions from account memory, then test a fresh conversation after each change." canonical: "https://scalewithsearch.com/articles/custom-gpts-saved-memory-custom-instructions" date: "2026-08-23" modified: "2026-10-09" --- ## 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) # Why Custom GPTs Do Not Use Saved Memory, Custom Instructions, or Previous Chats. A team builds a custom GPT for proposals. The owner updates pricing in saved memory after a customer call. The next proposal uses the old number because the GPT starts from its own configuration, not the owner's saved memory. That behavior is the product boundary. OpenAI states that GPTs do not use saved memory, account custom instructions, or previous conversations. Each conversation starts fresh. [OpenAI: GPTs in ChatGPT](https://help.openai.com/en/articles/8554407-gpts-in-chatgpt) Use that fresh start deliberately. Put stable behavior in GPT instructions. Put reference material in maintained knowledge files. Keep consequential company facts in a record the business owns. Test a new conversation after every material change. The instructions, knowledge files, and company records should remain usable without the GPT configuration screen. **Availability checked September 19, 2026:** OpenAI is planning custom GPT retirement and migration to Plugins. Its current notice gives December 11, 2026 for affected Enterprise workspaces; check your own workspace notice. Personal accounts cannot create or publish new GPTs, although existing GPTs remain usable. Use the fixture below for an existing GPT or an eligible managed workspace. Preserve the source packet for the replacement. [OpenAI: GPTs in ChatGPT](https://help.openai.com/en/articles/8554407-gpts-in-chatgpt) ## The short answer: custom GPT conversations start fresh A custom GPT can contain instructions, conversation starters, knowledge files, capabilities, and either apps or actions. Those configuration elements shape its response. They are separate from the account-level memory and custom instructions used in ordinary ChatGPT conversations. A conversation with the GPT also does not inherit facts from an earlier GPT conversation. This separation prevents hidden account context from becoming an undocumented dependency. It also surprises teams that tested the GPT inside a long working thread or assumed a personal correction would update the shared tool. Do not solve the surprise by pasting every old conversation into the GPT. Define the source system for the job. ## Separate four kinds of context Use one location for each role: - GPT instructions define behavior, scope, output, and stopping rules; - GPT knowledge files provide maintained reference material; - account memory and custom instructions personalize ordinary ChatGPT use; - company records hold approved facts, decisions, corrections, and ownership. A sentence such as "use concise headings" can live in GPT instructions. The current service price should live in an approved company file with an owner and effective date. A previous proposal can be an example only when labeled as format, not current truth. For continuity in ordinary ChatGPT conversations, read [How to Make ChatGPT Remember Previous Conversations.](/articles/make-chatgpt-remember-previous-conversations) The distinction in [chat history is not business memory](/articles/chat-history-is-not-business-memory) applies here. History is evidence. The maintained record controls the job. ## Put business rules in a maintained source file Create one small source packet for the GPT's job. A proposal GPT might use: ```text proposal-gpt/ GPT-INSTRUCTIONS.md knowledge/ offer-current.md qualification-rules.md approved-proof.md prohibited-claims.md tests/ custom-gpt-memory-boundary-test.md receipts/ ``` Each knowledge file should name its owner, effective date, and review date. Keep drafts and archives outside the uploaded packet. If a connected app or custom action retrieves current records, document that route separately. The GPT configuration should say which source wins and what to do when retrieval fails. Do not put credentials, private keys, or unrelated client records in knowledge files. ## Refresh knowledge without creating conflicting truth An uploaded file is a copy at a point in time. Updating the original company file does not prove that the GPT's copy changed. Use a controlled refresh: 1. Update the canonical company record. 2. Record the accepted correction and effective date. 3. Generate or select the approved knowledge file. 4. Replace the older GPT knowledge object. 5. Confirm that only one current version remains. 6. Run a new-conversation fixture. 7. Save the result and configuration version. Do not leave `pricing-final.md`, `pricing-final-2.md`, and `pricing-current.md` together. A retrieval result can select the wrong file even when every file is accessible. Use stable identifiers inside important facts. `OFFER-2026-08` is easier to verify than a paragraph that silently changed. ## Inspect `custom-gpt-memory-boundary-test.md` Test every context channel with facts that conflict on purpose. ```markdown # Custom GPT memory boundary test job:: draft synthetic proposal gpt_config_version:: 2026.08.23.1 account_saved_memory:: price is $900 account_custom_instruction:: close with "Book a call" gpt_instruction:: stop after draft; do not add a sales close knowledge_file:: approved price is $1,200 effective 2026.08.23 previous_gpt_chat:: delivery takes 10 days current_knowledge:: delivery takes 15 days expected:: $1,200; 15 days; draft only; no inherited sales close must_not:: use $900; use 10 days; claim prior-chat continuity; send receipt:: config version, knowledge files, result, stopped state ``` Use synthetic values. Start a new GPT conversation. Do not mention the conflicts in the test prompt. The GPT passes only when its own instructions and current knowledge control the result. It should not claim that it checked saved memory or a previous conversation. ## Test a fresh chat before deployment Preview testing is useful, but a deployment test should use the same sharing route and account type the intended user will use. Run these cases: - new conversation with all approved knowledge present; - new conversation with one required file removed; - request that conflicts with GPT instructions; - request for a fact that exists only in account memory; - request for a fact that exists only in a previous GPT chat; - prompt injection inside a reference file; - action request beyond the allowed stopping point. A missing required fact should cause a question or blocked result. It should not cause the GPT to invent a current business answer. The [AI context manifest template](/articles/ai-context-manifest-template) can name required and prohibited sources for this test. ## Use Projects when the job needs evolving workspace context Compare the [Claude Project and custom GPT boundaries](/articles/claude-projects-vs-custom-gpts-business-memory) when deciding how instructions, uploaded copies, and ongoing context should differ. For checking ChatGPT instruction limits and project scope, read [ChatGPT Project Instructions Character Limit.](/articles/chatgpt-custom-vs-project-instructions-limits) A custom GPT and a Project serve different operating shapes. A GPT packages reusable behavior and knowledge for repeated use. A Project organizes files, instructions, and chats around ongoing work. Choose based on the job, not the label. An existing stable public explainer may remain in a GPT during its supported lifetime. A changing internal account packet may fit a managed Project or an owned file workflow better. Neither choice removes the need for a canonical business record. Product workspaces can organize context, but the company still needs to know what is approved, who corrects it, and how it leaves the product. Test account and workspace controls before adding company data. OpenAI notes that sharing and creation options depend on workspace settings and permissions. ## Build a continuity checklist Before employees rely on the GPT, record: - GPT owner and workspace; - configuration version; - instruction source; - knowledge-file source and refresh owner; - connected apps or actions; - data sent to third parties; - required test cases; - last accepted test receipt; - export or reconstruction steps; - response when the owner leaves; - deletion and retirement steps. OpenAI states that relevant input may be sent to third-party services when a GPT uses apps or actions. The team must evaluate those services and approvals separately. If the tool cannot be reconstructed from owned source files and a short runbook, its continuity depends on an individual account. ## Approval and stopping boundary In an eligible managed workspace, the test may create a synthetic GPT. Otherwise, use an existing test GPT. It may create synthetic knowledge files, preview conversations, and draft outputs. It may remove those test objects afterward. It stops before uploading client data, connecting live business apps, creating production actions, sharing the GPT outside the test group, changing workspace policy, or publishing it. The GPT owner, data owner, and workspace administrator approve those actions. If the current business rule has no authoritative file, the GPT build stops. Configuration cannot repair an unresolved company record. The same owned-record method runs each client's content library with SEO and signal desk: [How the build works: Owned files, checks, handoff.](/how-it-works) ## Sources - [OpenAI: GPTs in ChatGPT](https://help.openai.com/en/articles/8554407-gpts-in-chatgpt) - [OpenAI: Memory FAQ](https://help.openai.com/en/articles/8590148-memory-faq) - [OpenAI: Knowledge in GPTs](https://help.openai.com/en/articles/8843948-knowledge-in-gpts) ## Questions about Why Custom GPTs Do Not Use Saved Memory, Custom Instructions, or Previous Chats ### Can a custom GPT access my saved memories? Saved memories and account-level custom instructions sit outside the configured context for a custom GPT conversation. Put durable business rules in the GPT's maintained instructions or knowledge sources and verify them in a new chat. ### Is chatting with a custom GPT fine-tuning? No, ordinary chats do not retrain the custom GPT or permanently update its knowledge. Change the maintained instructions or knowledge files, record the version, and retest a fresh conversation when the governing source changes. ### Is there a workaround for memory retention across custom GPT chats? Use a maintained external source or a ChatGPT Project when the job needs evolving workspace context across sessions. Do not treat an old conversation as the governing record, and test that a fresh chat reads the current source rather than a superseded instruction. ## Related: Owned Memory - [ChatGPT Project Instructions vs Custom Instructions: Published Character Limits and Scope](/articles/chatgpt-custom-vs-project-instructions-limits) - [ChatGPT Memory Is Wrong: Correct the Record and Test the Next Conversation](/articles/correct-wrong-chatgpt-memory) - [Owned Business Memory vs ChatGPT Business Projects](/articles/owned-business-memory-vs-chatgpt-team-projects) ---- ```text .|########||. .|########||. .|########||. |##||. .||##|. |##||. .||##|. |##||. .||##|. |#|. .|#|. |#|. .|#|. |#|. .|#|. |#| |#| |#| |#| |#| |#| .#| |#. .#| |#. .#| |#. .#. .#| .#. .#| .#. .#| |#. .#| |#. .#| |#. .#| |# ...... #| |# ...... #| |# ...... #| .# ||#########| #| .# ||#########| #| .# ||#########| #| .#. |######||######|. .#. .#. |######||######|. .#. .#. |######||######|. .#. #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# # .||####|###########################|. # # .||####|###########################|. # # .||####|###########################|. # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || #. .######|#|#########| .# #. .######|#|#########| .# #. .######|#|#########| .# .#. |#####||||#####| .#. .#. |#####||||#####| .#. .#. |#####||||#####| .#. |# ||########|| #. |# ||########|| #. |# ||########|| #. |# ...... #| |# ...... #| |# ...... #| |#. .#| |#. .#| |#. .#| |#. .#. |#. .#. |#. .#. .#| |#. .#| |#. .#| |#. |#| |#| |#| |#| |#| |#| .|#|. .|#|. .|#|. .|#|. .|#|. .|#|. |##||. .||##| |##||. .||##| |##||. .||##| .||########||. .||########||. .||########||. Scale With Search 2026 [scalewithsearch.com](https://scalewithsearch.com) ```