--- title: "SEO for AI agents" description: "AI agent SEO from ShoutRank. Make the page an agent can fetch, quote, and act on, without treating one agent run as a Google ranking." canonical: https://shoutrank.com/ai-agents markdown: https://shoutrank.com/ai-agents.md --- # SEO for AI agents AI agent SEO from ShoutRank. Make the page an agent can fetch, quote, and act on, without treating one agent run as a Google ranking. AI agent SEO is the work of being the page an agent fetches when it is asked to compare, recommend, or take the next step. An agent does more than answer. It opens pages, follows a link, and sometimes tries to complete a task. If the offer lives behind a script the agent never runs, or behind a login, the agent cites whoever published the fact in plain HTML. ShoutRank keeps this log apart from [Claude](https://shoutrank.com/claude-seo), [ChatGPT](https://shoutrank.com/chatgpt-seo), and the [Google AI Overview](https://shoutrank.com/google-ai-overview-seo). A chat reply is not an agent run. An agent run is not a ranking. ShoutRank does not guarantee a ranking or a citation. ## The problem ### The agent cites a directory, not you. A tool list states the price, the limit, and the alternative in one table. Your product page assumes the reader already knows the category. ### The useful page is not in the HTML. The first response is a shell. Pricing, docs, and the comparison render only after a script. The agent stops at the shell. ### One transcript becomes the strategy. A single signed-in run named you. The next run, with a clean session, named someone else. The team ships against the flattering one. ## How ShoutRank approaches it We name the agents in the brief and give each a fixed task set: compare, recommend, and the step a buyer would ask an agent to take. We save the pages fetched and the action the agent attempted. If the page cannot be fetched, that is [technical SEO](https://shoutrank.com/technical-seo). If the page can be fetched and still does not answer, that is [LLM optimization](https://shoutrank.com/llm-optimization). We do not add hidden instructions telling an agent what to do. ## What the work includes ### Agent list Which products are in the program, and which were not run. ### Task set The comparisons and next steps a buyer would hand an agent. ### Fetch log URLs the agent opened, including shells, logins, and third-party pages. ### Page fixes Facts moved into HTML an agent can read without executing the whole app. ## Process ### Discovery Confirm which agents buyers already mention, and which tasks they ask those agents to finish. ### Audit Run the task set and store the pages fetched, the names, and the step the agent took or refused. ### Strategy Separate a fetch failure from a content failure. They are not the same fix. ### Implementation Put the answer, the limit, and the next step in HTML. Leave login walls labeled as login walls. ### Monitoring Re-run the same tasks. A blank stays blank until that agent is actually checked. ### Optimization Drop tasks no buyer asks, and keep agent runs off the chat and overview sheets. ### Reporting The fetch log, the action, and whether the assigned URL was the one used. ## Why ShoutRank ### Agents are their own log A chat mention is not evidence an agent fetched you. We do not merge the two. ### HTML before choreography An agent that cannot read the page cannot take the step. Rendering is part of this service. ### No invented coverage We report the agents and tasks we ran. We do not publish a share of every agent on the internet. ## How success is measured Tasks where the agent opened the URL you assigned. Tasks where it named you and fetched a different page. Steps it could not finish because the fact was not in the HTML. A named client result is published when that client clears the figure. Until then the report is the measurement. ## Questions ### Is this the same as Claude SEO or ChatGPT SEO? No. Those pages cover a reply in a chat. This page covers an agent that fetches pages and tries to act. The logs stay separate. ### Can you make an agent book, buy, or sign up? We can make the public page say the offer, the limit, and the next step. We do not control the agent, and we do not guarantee it will choose you. ### Do you run every agent product? No. The brief names the agents. Anything not run is left blank. A blank is not a zero. ## Related - [Claude SEO](https://shoutrank.com/claude-seo): The chat reply, when the product is Claude and not an agent. - [Technical SEO](https://shoutrank.com/technical-seo): When the agent only receives a shell. - [LLM optimization](https://shoutrank.com/llm-optimization): The passage an agent can lift without inventing the offer.