> ## Documentation Index > Fetch the complete documentation index at: https://docs.promptingcompany.com/llms.txt > Use this file to discover all available pages before exploring further. # Help me improve the AEO of my product > Ask your agent to find where your product falls short in AI answers and make the changes you approve in TPC. Ask your agent this, with `` replaced by your product's name: ```text theme={null} Help me improve the AEO of my product ``` The agent reads your analytics, prompts, content, and knowledge files, suggests changes, and makes the ones you approve. ## Before you ask * Connect the server and sign in. See [Connect the MCP server](/mcp/connect). * Have your organization slug ready and put it in the same message. See [Find your organization slug](/mcp/connect#find-your-organization-slug). * Use the product name as it appears in TPC. * Decide how far the agent may go, for example drafts only, or drafts and publishing. * Generating content drafts needs an active or trial Pro or Enterprise subscription. ## What the agent does The agent chooses its own calls. A typical run follows these steps. `list-products` returns the product's slug and ID. `explore-analytics` with `metric: "sov"` and `dimension` set to `prompt`, `topic`, or `competitor` shows where your share of voice is low and who ranks above you. With `metric: "citations"` and `dimension: "source"`, it shows which sources answer engines cite. `list-prompts` shows the prompts TPC tracks. `listContent` lists your live pages, or drafts in review when the agent sets `status`. `getContent` reads one page or draft in full. `listKnowledgeFiles` and `getKnowledgeFile` read the product's knowledge files. `list-prompt-suggestions` lists candidate prompts that are not tracked yet. The agent proposes changes. After you approve them, it runs the write tools below. To have TPC write drafts for the prompts where you are weakest, say so in your message, for example: "Generate drafts for my five lowest share-of-voice prompts." The agent then calls `writeContent` with `action: "generate"` and those prompt IDs, and each draft stays linked to the prompt it was written from. Otherwise the agent may write pages itself with `writeContent` and `action: "create"`, which are not linked to a prompt. ## Write tools | Tool | What it changes | | - | - | | `writeContent` | Drafts and live pages on your site. See [Content actions](#content-actions). | | `reviewContent` | Whether a draft is approved, rejected, published, or back in review. | | `removeContent` | Deletes a page or draft, or restores an archived live page. | | `add-prompt` | Adds a prompt for TPC to track across answer engines. | | `modify-prompt` | A tracked prompt's text, topic, category, keywords, sources, or metadata. | | `delete-prompt` | Archives a prompt. With `hard: true`, deletes it permanently. | | `generate-prompt-suggestions` | Replaces the current prompt suggestions for the product, or for one topic, with a new batch. | | `modify-prompt-suggestion` | A suggestion's text, persona, or answer engines. | | `review-prompt-suggestions` | Approves or rejects suggestions. Approving creates tracked prompts and conversation queries. | | `manageKnowledgeFiles` | With `action: "put"`, creates or replaces a file in the product's [Knowledge Base](/guides/knowledge-base) on the live default branch. It cannot write `schema.md`. | | `modify-product` | The product's name, domain, description, keywords, differentiators, or competitors. | ### Content actions Each content tool takes an `action`. `writeContent` writes drafts and edits live pages: * `generate` writes new drafts from up to 25 tracked prompts. `create` starts a draft from text you provide. Both land in review, not live. * `revise` replaces a draft's text or sends feedback for a rewrite. * `update` edits a live page. * `cancel` stops a generation that is still running. `reviewContent` moves drafts toward your site: * `approve` accepts a draft. For blog and FAQ drafts, which is what `generate` writes, approving also queues the draft for publishing. * `publish` publishes a draft that is already approved. * `reject` sends a draft back with a reason. * `revert` takes a published page back to review. `removeContent` deletes and restores: * `delete` removes a page or draft. * `restore` brings back an archived live page. ## Approval Your MCP client asks for your approval before it runs a write tool, unless you have allowed that tool to run without asking. The server marks each write tool as one that changes data. Its instructions also tell the agent to confirm with you first. That covers adding, modifying, or deleting anything, writing a knowledge file, and approving prompt suggestions. Read these before you approve: * Deleting a draft in review with `removeContent` is permanent. You cannot restore it. * `delete-prompt` with `hard: true` deletes the prompt permanently. * `generate-prompt-suggestions` replaces the current suggestions. * Approving prompt suggestions adds tracked prompts and may change future share of voice results. If a write tool times out, the change may still have gone through. Ask the agent to check before it retries. ## What you get back * The gaps the agent found and the changes it proposes * For each change you approve, the tool's result, such as new draft IDs and titles, the new or changed prompt, the number of suggestions approved or rejected, or the commit for a knowledge file * New drafts, which stay in review until you approve them. Generated drafts reach `pending_review` when generation finishes. You can also review drafts in TPC. See [Create content](/guides/create-content). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.