# How Selfwin works: the AI employee for AI visibility > Selfwin reads your website, writes the prompts your customers ask AI assistants, runs them daily on ChatGPT, Gemini, and Perplexity, and every week drafts the content that wins the prompts you lose. You approve and publish; it never publishes on its own. - Published: 2026-09-05 - Author: Selfwin team - Canonical: https://selfwin.co/blog/how-selfwin-works - Tags: selfwin, product, ai-visibility ## Key takeaways - Onboarding is a brief, a prompt list, and a competitor list. Selfwin drafts all three from your website and you confirm them. - Every active prompt runs nightly on each engine; answers, mentions, and cited sources are kept so the trend is real. - Every Monday the visibility agent diagnoses the prompts you lose and drafts the pages, replies, and pitches to win them. - Nothing goes live without your approval. Selfwin never publishes, posts, emails, or edits your site on its own. Selfwin is built around one loop: measure where your brand shows up in AI answers, find the prompts you lose, draft the work that wins them, and wait for you to approve. Here is what each part does. ## Onboarding: a brief, prompts, competitors You give Selfwin your website. It reads the key pages, the homepage, pricing, features, docs, and about, and writes a project brief: the category in buyers' words, the platforms, the differentiators, the audiences, the use cases, pricing, and the competitors the site names. You correct it. From the confirmed brief it plans five to seven topics and writes prompts under each: keyword fragments, questions, imperative asks, and constraint-laden long-tail prompts, tagged with intent, specificity, and whether a competitor is named. Your own brand name is never in a tracked prompt. Near-duplicates are removed. You pick at least ten to track, usually around twenty-five, and confirm at least three competitors. ## Tracking: the nightly run Every active prompt runs on each engine during a nightly window in your project's time zone. ChatGPT and Gemini are asked through their web interfaces in a fresh, anonymous browser session so the answers match what a new user would see. Perplexity is asked through its API. Selfwin never bypasses a login wall or a CAPTCHA; if an engine blocks a run, the run is recorded as failed and shown in your run history. For each answer it stores the full text, which tracked brands were mentioned, the sources cited, and, where the engine exposes them, the searches it ran to ground the answer. From this it computes mention rate, share of voice, and citation counts per prompt, per engine, per day. ## The dashboard: trend, not snapshot The dashboard shows visibility over time against your competitors, by engine and by topic. Because answers vary run to run, the chart is honest about data volume: with a day of data it shows what was collected, with a week it shows a line, and with two weeks it shows a trend you can act on. ## The visibility agent: Monday's work Once a week the agent reviews every prompt you lose or are slipping on and asks why. It looks at what the winning answers cite, what those pages say, and what yours lacks. Then it drafts the fix: a page or a section for your site, a reply for the community thread the engines keep citing, or a pitch to a source that covers your category. Every draft carries its diagnosis and the prompt it targets, and it is checked against a linter so it stays factual and on brand. Drafts are queued one action per day, ordered by how much visibility each is likely to move. ## Approval: the line Selfwin does not cross Selfwin never publishes a page, posts to a community, sends an email, submits an indexing request, or edits your production site. Every draft waits for you. Approve it, publish it, and the next nightly runs show whether it worked. ## Where you work with it A daily brief lands in your Slack channel. The whole job is also available to Claude over MCP, so from Claude Code or the Claude app you can ask how visibility moved, review the queued action, or delegate the next one. Everything is in the dashboard as well. ## What it costs to run Selfwin is self-hosted and free. You bring your own model and browser-session keys, and you can read every line of what it does. ## FAQ ### What does Selfwin actually do every day? It runs every active prompt on ChatGPT, Gemini, and Perplexity during a nightly window, records whether your brand and your competitors were mentioned and which sources were cited, and updates your dashboard and daily brief. ### Does Selfwin publish content automatically? No. It drafts pages, community replies, and outreach, queues them with a diagnosis of the prompt they target, and waits for your approval. Publishing is always your action. ### Which AI engines does Selfwin track? ChatGPT and Gemini through their web interfaces, and Perplexity through its API. Claude is available for delegating work over MCP, not as a tracked engine. ### How do I work with Selfwin day to day? Through a daily brief in Slack, the dashboard, or Claude over MCP, where you can ask about visibility, review drafts, and delegate the next action.