--- title: "Keep your AI context across chatbot sessions." description: "Use your AI tool to recall past conversations from a context file. Test context carryover in a new session and learn where built-in memory features stop." canonical: "https://scalewithsearch.com/articles/why-do-ai-chatbots-forget" date: "2026-01-28" 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) # Keep your context when an AI chatbot resets between sessions. To recall past conversations and carry context into your AI tool, keep a maintained context file and test it in a new session. You spend an hour on a project with ChatGPT. The responses are excellent. You close the browser. The next day you open a new chat, and it does not know who you are. The same thing happens with Claude, Gemini, and Copilot. Different companies build these products, and they share the same behavior. It is not a bug. It follows from how the models work, and you can plan around it once you know the mechanism. ## Know that chatbots are stateless A chatbot does not store what you said last week and recall it when you return. Each conversation runs in a temporary space called the context window. The context window is everything the model can see for the current answer. It holds the conversation from the first message to the latest one, plus any files you added. When you start a new chat, the window starts empty. The previous conversation is not loaded. Engineers call this stateless design. The model keeps no state between sessions, and each conversation stands alone. Vendors build it this way for speed and cost. To load every past conversation into every new session would be slow and expensive. ## Read context windows as dated capacity A context window is measured in tokens. A token is a piece of a word, and one token averages about three quarters of an English word. A short greeting such as "Hello, how are you?" is about six tokens. Window sizes change with each model release. Treat these figures as observations at a date, not as current limits: | Product or model | Window | About | Observed | |---|---|---|---| | ChatGPT | Differs by plan and by model | Varies | OpenAI plan page | | Claude Sonnet 4.5 | 200,000 tokens | 150,000 words | Anthropic release, September 2025 | | Newer Claude models, such as Claude Sonnet 4.6 | 1 million tokens | 750,000 words | Anthropic documentation, September 2026 | | Gemini 1.5 Pro | Up to 2 million tokens through the API | 1.5 million words | Google, June 2024 | Check the vendor's current page before you plan around a figure. Large numbers can still run out fast. Suppose you paste three items at the start of a chat: - a project brief of 2,000 words - a style guide of 1,500 words - three examples of 1,000 words each That is 6,500 words, or about 8,700 tokens, before the conversation starts. In a window of 8,000 tokens, the setup alone would not fit. In a larger window it fits, but it uses context on setup, not on the work. The [comparison of context windows and persistent memory](/articles/context-window-vs-persistent-ai-memory) separates the size of a window from memory across sessions. ## Know why saved history does not help Most chatbots save your chat history. You can scroll back and read last month's conversations. History is not memory. A saved chat is not loaded into a new session. When you start a fresh conversation, the window is empty. Your past conversations exist, but they are not active. To make the model use them, you copy old messages into the new chat. That is you doing the remembering. Memory features narrow the gap. ChatGPT Memory and Copilot Memory save discrete facts, and since April 2025 ChatGPT can also draw on selected past chats. They keep "prefers Python." They do not keep the long discussion about why you chose an asynchronous design over threads. ## See what happens as a chat fills In a long conversation, the model does not stop at the limit. Products handle an overflow in different ways. Some drop the oldest messages. Some summarize them. Well before the limit, a second effect starts. Models do not use every part of a long context equally. A 2023 study found that models use information at the start and end of a long input more reliably than information in the middle (Liu et al., "[Lost in the Middle](https://arxiv.org/abs/2307.03172)"). In practice, context you give at message 1 can lose weight by message 30. New information in the middle of a long chat may never shape later answers. The longer the conversation, the less predictable recall becomes. The guide to [why AI gets worse in long conversations](/articles/why-does-ai-get-worse-in-long-conversations) covers the symptoms and a restart procedure. ## Know where memory features stop ChatGPT, Copilot, and Gemini each have a memory feature. You can say "remember this," and the product saves a note. These features store facts: your name, preferences, and project names. They do not store reasoning, methods, or working context. Grok also extracts details from past chats, and a [Grok memory recall test](/articles/grok-memory-problems) shows what reaches a new chat. Three gaps remain: - **They keep conclusions, not conversations.** You discuss three pricing models and reject two. Memory may keep "uses tiered pricing" and drop the reasons. - **They do not compose context.** Memory can hold three facts: you use Python, you work on Project Alpha, and you prefer asynchronous code. The model does not reliably combine them into "write asynchronous Python for Project Alpha" unless you ask for it. - **They stay inside one product.** A memory persists across sessions in the product that saved it. It does not move from ChatGPT to Claude, or from one vendor account to another. ## Know why every vendor has the limit This is not specific to one company. OpenAI, Anthropic, Google, and Microsoft build chatbots on the same kind of model. The reason is compute. To write a response, the model processes every token in the context window. In the standard transformer design, that work grows quadratically with length. Twice the context means about four times the work. Storage is not the constraint. These companies can store your chats. The cost is the work to process a larger context for every answer. So every chatbot resets between sessions and adds memory as a selective layer on top. ## Know why the common workarounds break People try four fixes: - **Paste a context document into each session.** It works until the document is long enough to crowd the window before the conversation starts. - **Use memory features.** They help with small details. They do not carry working context. - **Summarize past chats and paste the summary.** The summary keeps what you decided and loses why. - **Stay in one conversation for weeks.** It works until the window fills, and then the start of the conversation fades. All four share one flaw: they work against the design. Each patch also needs you to curate what the AI should know, which costs the time the AI was supposed to save. Curated context still decays inside a long chat. And none of the patches follows you to another product or device. ## Give the model a file to read Stop trying to make the chatbot remember. Give it a file to read. Write one Markdown file with who you are, what you do, how you work, what you tried, and what you decided. Keep it in your project folder. Each time a session starts, the tool reads the file, so the session starts with your context and without a manual paste. Claude Code works this way. It is a terminal program that reads a `CLAUDE.md` file from the project folder at every session start. The [guide to Claude Code](/articles/what-is-claude-code) covers setup. The file approach works for four reasons: - **It loads in full each session.** A new session reads the current file from the start. It is not immune to long-chat effects inside one session, so keep it short and restart when a chat grows long. - **It is version-controlled.** It is a text file, so you can track it in Git, share it with a team, and update it as the project changes. Everyone reads the same context. - **It composes.** One file can point to others, so you structure context in the way the work needs. - **You decide what it holds.** A memory feature chooses what to keep. With a file, you write down what matters and leave out the rest. ## Run a carry-over test Test your setup with a planted fact. 1. Write a harmless, specific fact in your context file, such as "Project codename: BLUE HERON." 2. Start a new session and ask, "What is the project codename?" 3. Close the session and start another one the next day. Ask again. 4. Open the same file in a second AI tool and ask again. 5. Change the codename in the file and ask once more in a new session. A pass means the correct codename in every session and in both tools, and the new value after the change. A product memory feature usually passes step 3 and fails step 4. A file passes both, because the context lives outside any one product. The [guide to switching AI models without losing context](/articles/switch-ai-models-without-losing-context) extends the test to a full working packet. ## Related: AI memory - [Set up an Obsidian vault that routes Claude Code to the right context](/articles/obsidian-ai-vault-setup-guide) - [Turn your second brain into context an AI reads every session](/articles/second-brain-for-ai) - [Build your AI context outside the chat app so your setup stops failing](/articles/why-your-ai-setup-isnt-working) ---- ```text .|########||. .|########||. .|########||. |##||. .||##|. |##||. .||##|. |##||. .||##|. |#|. .|#|. |#|. .|#|. |#|. .|#|. |#| |#| |#| |#| |#| |#| .#| |#. .#| |#. .#| |#. .#. .#| .#. .#| .#. .#| |#. .#| |#. .#| |#. .#| |# ...... #| |# ...... #| |# ...... #| .# ||#########| #| .# ||#########| #| .# ||#########| #| .#. |######||######|. .#. .#. |######||######|. .#. .#. |######||######|. .#. #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# # .||####|###########################|. # # .||####|###########################|. # # .||####|###########################|. # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || #. .######|#|#########| .# #. .######|#|#########| .# #. .######|#|#########| .# .#. |#####||||#####| .#. .#. |#####||||#####| .#. .#. |#####||||#####| .#. |# ||########|| #. |# ||########|| #. |# ||########|| #. |# ...... #| |# ...... #| |# ...... #| |#. .#| |#. .#| |#. .#| |#. .#. |#. .#. |#. .#. .#| |#. .#| |#. .#| |#. |#| |#| |#| |#| |#| |#| .|#|. .|#|. .|#|. .|#|. .|#|. .|#|. |##||. .||##| |##||. .||##| |##||. .||##| .||########||. .||########||. .||########||. Scale With Search 2026 [scalewithsearch.com](https://scalewithsearch.com) ```