> ## Documentation Index > Fetch the complete documentation index at: https://docs.stefanbrain.com/llms.txt > Use this file to discover all available pages before exploring further. > ## Agent Instructions > Follow the content contract in AGENTS.md. > Treat text as the canonical explanation; video and screenshots enhance it. > Do not publish unverified product behavior or duplicate an existing canonical article. # Give Chat the right context > Give StefanBrain a model to analyze, then add the correct Product or offer context and adapt the structure to your job. ## The result you want When an advertorial, ad or landing page gives you a useful model, analyze why it works before adapting it to your Product. This keeps the source's reusable structure separate from its offer-specific wording and details. ## Practical use cases Break down its section order and persuasion flow, then reuse the framework for your Product. Attach an ad image or screenshot and analyze its hook, visual hierarchy, proof, product visibility and CTA before creating variations. Use the Meta connector to compare current ads and investigate the hooks and formats behind the results. Attach an export, name the metrics and date range, and ask which patterns deserve the next test. Carry the selected direction, source facts, and next task into a fresh conversation without rebuilding the brief from scattered messages. You normally do not need to select a tool. Describe the result you want and let StefanBrain choose the useful capabilities. Select a tool only when you need its rigid controls or output. Use the smallest relevant context set. Do not attach 300 pages because they might help. Start with the source, Product and files that can change this result, then add more only when something is missing. ## Recommended approach: analyze, then adapt The model provides the structure. Your Product provides the facts and direction for the new work. Paste the public URL, attach the file, or add a screenshot. Name the item StefanBrain should treat as the model. Analyze the model before asking for a new draft. ```text theme={"system"} Here's an advertorial I want to learn from. Analyze why it works and break its structure into a reusable framework and outline we can use for @Product. Separate the reusable structure from the source offer's wording and product details. ``` Chat showing a native advertorial model request and a reusable structure map before Product context is added Review the framework before adapting it. If StefanBrain misunderstood a section, explain what it does and ask for a revised framework. ```text theme={"system"} The testimonial is proof after the mechanism section, not the mechanism itself. Update the framework around that distinction. ``` Type `@` and choose the Product for the new work. Start inside the correct Project when the job also needs its files or campaign direction. Keep the same Chat so StefanBrain can use the framework you approved. ```text theme={"system"} Now turn that framework into an advertorial outline for @Product. Keep the structure and persuasion flow. Use this Product's audience, mechanism, proof, and objections. For each section, include its purpose, key message, and evidence needed. ``` Chat showing a Product mention and an adapted outline with each section's purpose, key message and evidence needed Review the outline before asking for a complete draft. ## Apply the same workflow to ads and screenshots For an ad image or screenshot, analyze the visible structure before asking for variations. For a complete worked example, follow [Create native ad concepts from a winning ad](/knowledge-base/images/create-native-ad-concepts-from-winning-ad). ```text theme={"system"} Analyze this ad as a model. Break down the hook, visual hierarchy, product visibility, proof, CTA, and the order a viewer is likely to notice each element. Return the reusable creative structure and the product-specific details we should replace for @Product. ``` Then add or confirm the correct Product context and ask StefanBrain to adapt the structure to a defined audience, message, and format. In that focused workflow: 1. Attach the reference ad, its copy, and the image you want to keep. 2. Ask StefanBrain to explain the ad's reusable structure before requesting new ideas. 3. Request several distinct concept families that preserve the useful image but change the angle, setting, or story. 4. Remove generic ideas, explain what the response is overusing, and run a narrower follow-up. 5. Save the strongest concepts, then [generate the chosen direction in the Image workspace](/knowledge-base/images/use-the-image-workspace). The Chat result is a set of creative directions. Generate and review the actual images before treating any direction as ready to use. ## Use connector data when direct access improves the analysis A connector is useful when the job depends on current data in another service. For Facebook and Instagram ads, enable the intended Meta account under **Connectors**. [Browse connector guidance](/knowledge-base/connectors-automations) or [learn how connector access can be shared with your team](/knowledge-base/connectors-automations/share-connectors-with-your-team). ```text theme={"system"} Use the enabled Meta account to compare ads from [start date] through [end date]. Show which hooks and formats deserve deeper investigation using [the metrics that matter to this decision]. Return the observed patterns, supporting data, and three next tests. ``` If the connector is unavailable, attach a CSV and use the same request. Name the file, date range and metrics. ## Choose the right context source | Context source | Use it when | How to add it | Boundary to remember | | - | - | - | - | | **Product** | The task depends on reusable offer, product, or brand context. | Type `@` and choose the Product. | Correct stale Product facts at the source instead of overriding them in every Chat. | | **Current Project** | Related Chats need the same instructions, files, and linked Products. | Start the Chat inside that Project. | Keep unrelated clients and workstreams in separate Projects. | | **Project file** | The example or evidence should support several Chats in one Project. | Upload it to the Project, wait until it is ready, then reference it by name. | A newly uploaded file may not be searchable until processing finishes. | | **Chat attachment** | The model, CSV, image, or screenshot belongs to this conversation. | Select **Add files or photos**, paste it, or drag it into the composer. | Tell StefanBrain what role the attachment plays in the job. | | **Saved Output** | A prior result in the Chat is the exact source for the next step. | Type `@` and choose it under **Saved Outputs** when available. | Name what should stay and what you want to transform. | | **Public URL** | The task depends on a specific public page. | Paste the URL and name the section or question. | If the page cannot be read, attach an export or screenshot. | | **Connector** | The task needs current data from a connected service. | Open **Connectors** and enable the intended account. | An `@Provider` mention names the provider but does not select a specific account. | | **Existing Chat** | You need the exact prior analysis or decision trail. | Continue that Chat. | Put reusable facts in a Product or Project. | When several sources are present, say which one supplies the structure and which one supplies the offer facts. ## Carry useful context into a new Chat Stay in the current Chat when you are still refining the same result. Start a new one when the next job needs its own conversation, such as turning research into a script or preparing a separate client deliverable. Bring a short handoff with you. Name the selected version and include the decisions that affect the next task, rather than pasting the entire conversation. ### Create the handoff in the original Chat Ask for a brief you can inspect before the next Chat starts drafting: ```text wrap theme={"system"} Create a handoff I can paste into a new Chat to continue this work. Include the goal and audience, the Product facts we used, the direction we chose, the exact latest copy or outline, and the constraints that still apply. Separate final decisions from open questions. Include the source names and anything I need to attach again. Leave out rejected drafts. End with the next task: [what I want to do next]. Return the handoff here in the Chat. ``` Read the handoff against the version you chose. Replace stale details and add any missing source before copying it. ### Example: research to an email draft For a fictional desk lamp, a useful handoff might look like this: ```text wrap theme={"system"} Next task: Write one email introducing the lamp to people setting up a small home desk. Product facts: USB-C powered; three brightness settings. Use the attached Product brief for the remaining details. Chosen direction: A focused workspace in a small room. Start with the reader clearing a spot to work, then introduce the lamp. Keep: Plain, conversational wording and one CTA to view the lamp. Do not add a discount or a deadline. Sources to attach: Product brief and the two selected brand-voice examples. Open question: Which product-page link should the CTA use? Ask before inserting it. ``` The next Chat has a job to do, a chosen direction, and a clear distinction between known facts and missing information. ### Start the next Chat with the handoff Open a new Chat in the intended Project if the work belongs there. Paste the brief, select the correct Product, and attach the source files the next task needs. Use the first response to catch missing inputs before generating another draft: ```text wrap theme={"system"} Use this handoff to continue the work. First confirm the selected direction and which attached sources support it. Ask for anything essential that is missing, then complete the next task. Don't restart the research or bring back rejected directions. ``` Pasting a handoff does not attach the original files or give another person access to them. Add the relevant files to the new Chat instead of assuming they travel with the summary. For instructions you want to reuse across several Chats, see [Manage multiple clients with Projects](/knowledge-base/products-projects/manage-multiple-clients-with-projects). ## Common mistakes * **Drafting before analyzing.** Map and approve the structure first. * **Mixing the source offer with your Product.** Give each source one role. * **Adding everything available.** Use only sources that can change the result. * **Giving a vague correction.** Explain what was misunderstood and what should change. * **Pasting a URL without a goal.** Say what StefanBrain should inspect and why. * **Using `@Provider` as account selection.** Confirm the intended account under **Connectors**. ## Related articles * [Start your first useful Chat](/knowledge-base/get-started/start-your-first-chat) * [Products, Projects, and Chats](/knowledge-base/get-started/products-projects-and-chats) * [Use the Image workspace](/knowledge-base/images/use-the-image-workspace) * [Browse connectors and schedules](/knowledge-base/connectors-automations) This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.