--- title: "Give AI a context file so it knows your business without model training" description: "Training changes a model's weights and memory changes what it reads, so learn when fine-tuning pays and when a context file is enough." canonical: "https://scalewithsearch.com/articles/ai-memory-vs-ai-training" date: "2026-01-28" modified: "2026-10-02" --- ## 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) # Give AI a context file so it knows your business without model training. People ask: "How do I train AI to remember my business?" In most cases the question is wrong. You do not train the model. You inform it. Training and memory solve different problems. Training changes the model. It is expensive, technical, and slow. Memory gives the model context. It is cheap, simple, and takes effect at the next session. This article explains what each one does, why people mix them up, and when training is actually the right tool. ## Know what training means Training a model changes its weights: the internal parameters that decide how it responds. You feed it thousands or millions of examples, and it adjusts those parameters to fit the patterns. There are three kinds. **Pre-training** is what the model makers do. OpenAI, Anthropic, and Google train base models on very large datasets of books, websites, and code. Pre-training produces the GPT, Claude, and Gemini model families. It costs millions and takes months. A small business does not do this. **Fine-tuning** takes a pre-trained model and trains it further on your data. You supply labeled examples: inputs and the outputs you want. The model shifts its behavior toward your examples. It needs technical skill and a real budget. Most businesses do not need it. **Reinforcement learning from human feedback (RLHF)** uses human ratings of model outputs. The model learns which responses people prefer. RLHF is part of how chat assistants became conversational. It is expensive and complex, and a small business does not do this either. All three change the model's internal behavior. That is training. ## Know what memory means Memory gives the model context. You supply information, the model reads it, and it uses the information to respond. Memory does not change the model. It changes what the model sees. There are three ways to give a model memory: - **In the conversation.** You put context in the prompt: "I'm a real estate agent. I manage 200 leads. My CRM is [name]." The model reads it and answers with it. It lasts for that conversation. - **Product memory features.** ChatGPT Memory, Claude Projects, and Perplexity's memory store facts or files and use them in later conversations. Each works inside its own product. - **Context files.** You write a Markdown file with your business details: who you are, what you do, your frameworks, your processes. The file stays on your disk. An assistant such as Claude Code reads it every session, and the text it reads goes to the model provider for processing. This memory persists and works with more than one tool. None of these involve training. They supply information; the model does not learn anything permanent. The [comparison of six AI memory approaches](/articles/ai-memory-tools-comparison) sets product memory and context files beside custom GPTs, fine-tuning, and RAG. ## Compare the two side by side | Aspect | Training (fine-tuning) | Memory (context files) | |---|---|---| | What it does | Changes model behavior permanently | Supplies information the model reads each session | | Relative cost | High: labeling, training, and evaluation work | Low: your time to write the file | | Time to set up | Weeks to months | Same day | | Skill required | High: data labeling, model evaluation, API integration | Low: write a Markdown file | | New information | Needs retraining | Edit the file | | Portability | Tied to one model or API | Works with any assistant that reads files | | Typical use | Specialized behavior | Business context: clients, SOPs, brand voice | | Makes sense when | The model must behave differently from the base model | The model must know specific facts about your business | ## See why people confuse them Part of the confusion is language. People say "train AI on my content" when they mean "make AI aware of my content." An assistant that does not know your business looks like it needs to learn. It needs context. Vendors add to the confusion. Many advertise "custom AI trained on your data" when the product is retrieval-augmented generation (RAG). RAG puts your documents into the prompt next to the question. That is context, not training. History adds the rest. In 2020 to 2022, fine-tuning was the usual way to make a model "know" a business, because context windows were small. Context windows are now much larger, so for most business facts you load the context instead. ## Know when training is the right tool Training makes sense in four cases: - **Specialized behavior.** A model that reads radiology images, analyzes case law in a fixed way, or follows a company's coding conventions without instructions. The base model lacks the behavior, so fine-tuning adds it. - **Data that cannot leave your infrastructure.** When data cannot leave your infrastructure, choose an approved local model. Fine-tune only if evaluation shows a behavior gap. - **Strict output format.** Every output must follow an exact structure, such as a fixed JSON or XML schema or a structured report. Fine-tuning can hold a format more reliably than prompts. - **Cost at high volume.** At millions of API calls a month, a tuned model can use fewer tokens, because the instructions live in the weights instead of every prompt. Most businesses fit none of these. They need the assistant to know what they do. ## Follow a January-to-April update cycle Suppose you fine-tune on your business in January. In February, you sign three clients. In March, you change pricing. In April, you update the service list. The training run still reflects January unless you supply newer context or update the model. This is an illustrative case, not a measured customer result. To update the weights, prepare new examples, run and pay for another training job, evaluate it, and deploy the replacement. To update supplied facts, edit the source file and load it in the next session. Confirm the new answer before relying on it. Editing a file does not guarantee the assistant opened it. ## Know when a context file is enough For most business uses, context is the answer: - **Client work.** You want the assistant to know your active clients and their preferences. Keep the records in your CRM. Keep a client context file that says what matters and where the records live. - **Brand voice.** You want emails in your style. Put voice rules and labeled examples of your writing in a context file. - **Procedures.** You want the assistant to follow how you onboard clients, structure content, or handle support tickets. Write the SOPs as instructions. - **Frameworks.** You use a specific sales method, pricing model, or planning framework. Document it in Markdown. A context file handles all of these. It is faster, cheaper, and easier to update than training. [Why plain text is the durable layer for AI memory](/articles/plain-text-ai-memory) explains what belongs in the file and what belongs in a database. ## Write the context file Name the file `CLAUDE.md`, `CONTEXT.md`, or anything else; the name matters only to the tool that loads it automatically. Write six sections: 1. Who you are: name, role, business. 2. What you do: services, clients, projects. 3. Your frameworks: how you think and work. 4. Your SOPs: step-by-step processes. 5. Your voice: examples of your writing. 6. Your clients: key details and current projects, with links to the system of record. Save it in an Obsidian vault or any folder. Claude Code reads `CLAUDE.md` each session in that folder. The assistant now knows your business because you informed it. When a client detail changes, edit the file. The next session reads the change, with no retraining and no extra cost. When you move from ChatGPT to Claude, the file moves with you, because it is yours. [Replace the model without rebuilding the business memory](/articles/replace-the-model-keep-business-memory) covers that move. For an illustrative agency setup, separate three records: | File | Contents | |---|---| | `CLIENTS.md` | Active clients, contacts, industry, current projects | | `BRAND.md` | Voice, style rules, content frameworks | | `CAMPAIGNS.md` | Current goals, messages, deliverables | A client email loads the client record and voice rules. Ad copy also loads the live campaign. Update each record when its facts change. The file split lets one campaign change without replacing the agency's standing voice. ## Compare the effort The figures below are illustrative labor assumptions for one planning scenario, not a sourced January 2026 comparison. Scope, data quality, rates, and vendor fees vary. Compare the work your own evaluation requires. **Fine-tuning a model:** - data labeling: 100 hours or more; - a fine-tuning service or platform fee; - testing and evaluation: 20 to 40 hours; - timeline: 4 to 12 weeks. **Building a context file:** - write `CLAUDE.md`: 2 to 4 hours; - set up Obsidian and Claude Code: about 30 minutes; - ongoing cost: your time, plus an assistant plan that includes Claude Code; - timeline: the same day. Under these assumptions, labeling and evaluation take 120 hours or more, while the context setup takes fewer than five. Those are scenario totals, not typical measured results. Updating facts in supplied context requires an edit; updating weights requires another evaluated training run. ## Test ten facts before you train 1. Write ten questions about current contacts, project status, and services. 2. Put their answers in one context file. 3. Load it in a fresh session and ask all ten questions. 4. Record each correct, missing, and wrong answer. Check file-read evidence. 5. Edit three answers and repeat in another fresh session. 6. If you already have a fine-tuned model, ask it the same questions. Check whether all three edits appear in the answers. A failure can mean stale retrieval, a file that did not load, or model error. Keep the results rather than assuming files guarantee recall. If current facts pass but the style still fails a separate evaluation, fine-tuning may earn its cost. ## Decode "AI trained on your content" When a vendor offers "AI trained on your content," the product is usually one of three things: - **RAG.** Your documents sit in a database. For each question, the system searches it and puts the relevant passages into the prompt. That is context, not training. - **Embeddings and vector search.** Your content becomes numeric representations. A query finds similar content and adds it to the prompt. Still context. - **Fine-tuning.** The vendor really does train a model on your data. This is rarer, slower, and more expensive. Ask which one you are buying. RAG is a legitimate choice for a large document set. It is not training, and it does not decide which of two conflicting documents is current. [RAG is retrieval; business memory also needs authority](/articles/rag-vs-business-memory) explains that gap. For a map of the kinds of memory in AI systems, see [types of memory used in AI systems](/articles/types-of-memory-ai-systems-business). If your goal is "make AI remember my business," write the context file. Update it when the business changes. The assistant reads it every session. ## Related: AI memory - [Split one AI context file into a knowledge base your assistant can navigate](/articles/how-to-build-ai-knowledge-base) - [Build a local-first AI setup that sends only the context each request needs](/articles/local-first-ai-architecture) - [Set up an Obsidian vault that routes Claude Code to the right context](/articles/obsidian-ai-vault-setup-guide) ---- ```text .|########||. .|########||. .|########||. |##||. .||##|. |##||. .||##|. |##||. .||##|. |#|. .|#|. |#|. .|#|. |#|. .|#|. |#| |#| |#| |#| |#| |#| .#| |#. .#| |#. .#| |#. .#. .#| .#. .#| .#. .#| |#. .#| |#. .#| |#. .#| |# ...... #| |# ...... #| |# ...... #| .# ||#########| #| .# ||#########| #| .# ||#########| #| .#. |######||######|. .#. .#. |######||######|. .#. .#. |######||######|. .#. #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# # .||####|###########################|. # # .||####|###########################|. # # .||####|###########################|. # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || #. .######|#|#########| .# #. .######|#|#########| .# #. .######|#|#########| .# .#. |#####||||#####| .#. .#. |#####||||#####| .#. .#. |#####||||#####| .#. |# ||########|| #. |# ||########|| #. |# ||########|| #. |# ...... #| |# ...... #| |# ...... #| |#. .#| |#. .#| |#. .#| |#. .#. |#. .#. |#. .#. .#| |#. .#| |#. .#| |#. |#| |#| |#| |#| |#| |#| .|#|. .|#|. .|#|. .|#|. .|#|. .|#|. |##||. .||##| |##||. .||##| |##||. .||##| .||########||. .||########||. .||########||. Scale With Search 2026 [scalewithsearch.com](https://scalewithsearch.com) ```