# Review Product Context Before an Agent Generates a Campaign [Read the original article]() By Garry ยท Founder Published: 2026-09-29T20:11:56.951Z Updated: 2026-09-29T20:13:22Z 6 min read Categories: Publishing workflows Reviewed product context gives an AI agent a reliable starting point for a Caroush campaign. ![A rounded mint sculpture rests on a stone pedestal inside an ivory arch, surrounded by translucent blue panels.]() ## Key takeaways - Inspect the product name, description, current website reference, reviewed audiences, editorial themes, and associated assets. - Treat analysis output as a proposal to verify against the real product and editorial objectives. - Update the record when the approved product facts or editorial preferences have materially changed, using the documented workflow and required fields. Reviewed product context gives an AI agent a reliable starting point for a Caroush campaign. Before generation, inspect the saved product, audience preferences, content themes, and relevant reference assets so the assistant does not build a polished campaign on an outdated interpretation. A product analyzer can suggest useful themes, but a suggestion is not the same as a business-approved claim or audience decision. Treat the saved record as an operational input to review, especially when the product, positioning, or offer has recently changed. ## What should you inspect in the saved product? Inspect the product name, description, current website reference, reviewed audiences, editorial themes, and associated assets. Compare them with the approved product source and identify discrepancies before requesting generation. For a fictional booking service, an older record may describe individual practitioners while the current campaign targets small teams. That change affects examples, terminology, and the next step. The agent should not simply combine both audiences into a generic message for everyone. The [Caroush catalog]() documents list\_products, get\_product, and list\_product\_assets. Use the actual owned product identifier and inspect the returned context. A similar name is not enough when several products or versions exist in the workspace. Your [audience segmentation guide]() can help define the reader situation. The saved audience preference should serve the current task, not become an unquestioned permanent label for every campaign. ## How should you treat product-analysis suggestions? Treat analysis output as a proposal to verify against the real product and editorial objectives. Confirm audience, themes, and factual interpretation before those suggestions guide a generation run. Caroush documents fetch\_product\_details as fetching a public product page and returning suggestions without saving a product. Product creation and updates can queue analysis. These are specific capabilities, not a general browsing or research service that verifies every fact across the web. [Google's helpful-content guidance]() emphasizes useful, accurate, original information. A product page summary may be a starting point, but it does not necessarily contain the practical explanation a reader needs. Add the verified use case and limitations that make the campaign useful. If the analyzer infers a theme that the business does not want to pursue, record the decision. For the booking service, a theme about medical outcomes would be inappropriate if the product simply manages appointments. The agent should distinguish the customer's industry from claims the software itself can support. ## When should the product record be updated? Update the record when the approved product facts or editorial preferences have materially changed, using the documented workflow and required fields. Do not rewrite a shared product merely to satisfy one narrow campaign request. The current update\_product schema requires the product identifier and name and preserves omitted fields as documented. Read it before constructing a request. If the intended change is only a campaign-specific angle, keep that in the campaign brief instead of altering the shared product's lasting context. [Anthropic's context-engineering guidance]() supports curating relevant context and preserving decisions across tasks. Apply that principle by separating durable product truth from temporary campaign instructions. An agent needs both, but they should not be confused. The [brand voice guide]() belongs alongside this record as writing guidance. It cannot authorize stronger feature claims. If the product source says that a task is assisted, a preference for confident copy should not turn that into fully automatic behavior. ## How do you confirm generation used the right context? Record the selected product identifier, approved context version, campaign brief, and resulting content identifiers. Review the generated output against those inputs rather than assuming the tool used them correctly in every sentence and image. Caroush's documented generate\_carousel workflow can use reviewed product audiences by default. That makes the pre-generation check valuable: an incorrect saved audience can influence the whole sequence. Read the current schema for the options relevant to your request and avoid unsupported arguments. Use the [AI carousel generator]() after the product and reader task are clear. Inspect slide copy, images, call to action, and product fidelity. A successful generation job proves that output was produced; it does not prove the campaign is factually or editorially ready. The [UGC creative brief guide]() offers a related source-of-truth discipline. Whether the next artifact is a hook, carousel, or text post, preserve the actual product behavior and mark proposed creative interpretation separately. ## Keep reference assets aligned with the product Review attached screenshots, logos, and photographs for version accuracy. A correct description paired with an old interface image can mislead readers. Tell the assistant which reference is current and which material should be treated as historical or illustrative. For the booking service, an image of the individual-practitioner setup may not explain a team workflow. The agent can identify the mismatch if the brief names it, but a person should inspect the actual asset before publication. Do not assume the media title establishes what the picture shows. If the needed asset is absent, route that gap to the normal production process. The current MCP media tools work with existing owned assets and do not establish a general direct-upload capability. The campaign should wait for the appropriate reference rather than invent a screen that appears plausible. ## Use a product-context review prompt A practical request is: > Read the selected owned product and compare it with this approved campaign source. Identify outdated facts, conflicting audiences, unsupported themes, and missing reference assets. Recommend whether to update the shared product or keep the difference in the campaign brief. Do not generate content until the relevant context is resolved. This prompt produces a decision rather than another large summary. The assistant should explain why each discrepancy matters to the intended output. A changed audience may affect examples; a changed feature may invalidate the central claim; a changed logo may only affect the visual package. Keep the decision owner visible. The agent can propose a correction, but the business's product and editorial owners establish what is current and intended. Save their decision with the resulting record so future campaigns do not reopen the same ambiguity. ## Test with an intentionally stale preference Use a harmless review example in which a saved audience differs from the campaign's approved reader. Ask the assistant to identify the conflict and recommend the smallest correction. It should not silently overwrite the shared context or blend both audiences into vague copy. Then inspect a generated draft against the resolved brief. Does the example fit the reader's situation? Does the call to action match the actual offer? Are limitations preserved? These checks reveal whether the context review improved the work in a practical way. A sound product-context process reduces downstream revision because the assistant begins with a coherent description of the product and task. It does not remove the need to review final content. It makes that review more focused by giving people a clear basis for deciding what the campaign should and should not say. ## Distinguish product truth from campaign preference A campaign may intentionally focus on one use case without changing the product's general audience. Record that choice in the brief. Updating the shared product to match every temporary campaign can make future generation inconsistent and erase useful context for other teams. For the booking service, a team-oriented campaign can emphasize coordination while the product still serves individual practitioners. The agent should know that the campaign selects one reader situation, not that the business has permanently abandoned the other audience. When deciding where a change belongs, ask whether another campaign should inherit it automatically. A corrected feature limitation probably should. A seasonal tone or one narrow example probably should not. This simple distinction keeps shared product records stable while allowing individual content briefs to remain specific and useful. ## Sources - [Effective context engineering for AI agents \\ Anthropic]() - [Creating Helpful, Reliable, People-First Content \| Google Search Central \| Documentation \| Google for Developers]() - [Caroush public tool catalog]() ## Frequently asked questions ### What should you inspect in the saved product? Inspect the product name, description, current website reference, reviewed audiences, editorial themes, and associated assets. Compare them with the approved product source and identify discrepancies before requesting generation. ### How should you treat product-analysis suggestions? Treat analysis output as a proposal to verify against the real product and editorial objectives. Confirm audience, themes, and factual interpretation before those suggestions guide a generation run. ### When should the product record be updated? Update the record when the approved product facts or editorial preferences have materially changed, using the documented workflow and required fields. Do not rewrite a shared product merely to satisfy one narrow campaign request. ### How do you confirm generation used the right context? Record the selected product identifier, approved context version, campaign brief, and resulting content identifiers. Review the generated output against those inputs rather than assuming the tool used them correctly in every sentence and image. ## About the author Garry Gaurav Sapkota builds Caroush, a workspace for creating, scheduling, and publishing social content. - [https://x.com/gauravsapkotanp]()