# How the autopilot works > What DispatchSEO does on its own for a cloud site - how it picks topics, how often it publishes, when it pauses itself, and the few things you control. Source: https://dispatchseo.com/docs/autopilot All docs in one file: https://dispatchseo.com/llms-full.txt On DispatchSEO Cloud, a new site runs on autopilot. You don't connect an AI of your own. Our server picks the keywords, writes the articles and publishes them on a fixed weekly schedule. This page explains what it does, when it does it, and what you can change. If you run DispatchSEO yourself (self-hosted), your sites still use your own coding agent, and nothing on this page applies to them. ## Before the first article Once you finish setup, three things happen without you: 1. We read your site. The server visits up to 150 of your pages and notes your sections, page titles and the pages you link to most. 2. We describe your site. From that reading and your homepage, a model writes a short profile: a tagline, a one-line description, a longer description and the categories you cover. You can rewrite any of it on **Settings > Brand**. Once you save your own version, we never overwrite it. 3. We research keywords. The server looks for searches people make around your categories (and, once Google Search Console has data, around the searches you already show up for). It keeps up to six ideas queued. The first article usually starts the next morning. If we can't read your homepage (some sites block servers, or sit behind a login), the dashboard asks you to describe your site on **Settings > Site** instead, and the first article starts from your description. Every article publishes the moment it passes the quality gate. There's no waiting for a publishing hour, and nothing for you to approve first. ## How topics are picked You never approve ideas. Each keyword has to pass every check below, in this order: - It gets at least 50 searches a month. - Its difficulty is under a ceiling set by your site's authority. A new site gets 10 out of 100, and the ceiling rises as your site earns links. - It's in the language your site is set to. - It isn't on your list of [topics to avoid](#what-you-control). - You don't already have it. That covers anything you've published, queued or written about before, near-duplicates included: "how to choose a crm" and "how to choose crm software" count as one article. - It fits your site. A judging model asks whether your readers would search it, whether your site can answer it credibly, and whether the searcher wants to learn or compare rather than buy right now. Topics that won't still be searched in a year are skipped too. The ones left are ranked by how much traffic they can realistically bring (searches a month, discounted by difficulty), and the best go to the front of the queue. Research runs again once a week. If nothing passes, research widens once: five more points of difficulty and at least 20 searches a month instead of 50. If that still finds nothing, the dashboard says "No winnable keywords found yet" and research tries again the next week. Suggesting a topic yourself is the fastest way past that. ## How often it publishes | Plan | Articles a week | |---|---| | Starter (and the free trial) | 1 | | Growth | 2 | | Scale | 3 | The week resets on Monday (UTC). Two articles for the same site never start less than two days apart, so three a week lands roughly Monday, Wednesday and Friday. The dashboard shows the date of the next one. There is no daily option, on purpose. A young site that suddenly publishes every day looks like a content farm to Google, and a steady pace does better. An article that stops because of a problem on our side (a supplier outage, our keyword data budget for the day) doesn't count against your week, and the next try isn't held back by the two-day spacing. ## How each article is written For every article the server reads the pages that rank on page one for the keyword and pulls out facts with the exact quote behind each one. It plans an article that covers what searchers expect and adds one thing those pages don't have. Then it writes from that research only, checks every figure and claim against its source, and checks the result isn't too close to any page-one article or to your own earlier posts. An article that can't beat page one isn't written at all. Then it publishes the same way as before. You get a pull request on your repo (merged for you on automatic publishing once its checks pass), a post on your WordPress site, or a page on the blog we host at yoursite.com/blog. An article only shows as **Live** once its page loads on your site and shows the article itself. A pull request that is still open shows as **Waiting for a merge**, with a link to it, and the dashboard warns you if any article hasn't gone live within two days. We start tracking an article's keyword on Google once the article is live, and check rankings weekly. ### The hosted blog If your repo has no blog folder, we host the blog for you and open one pull request in your repo that forwards yoursite.com/blog to it. Until that pull request is merged, the dashboard shows a **Merge the forwarding pull request** card linking straight to it. Your first article goes out either way; after that, new articles wait until /blog forwards, because a page nobody can reach from your site does you no good. We re-check the forwarding every week. If your site's host isn't one we can open a pull request for, **Settings > Publishing** shows the one rule to add yourself. If articles have nowhere to go yet, the dashboard asks you to **Choose where articles go**, and nothing is written until you do. ## When it pauses itself The autopilot stops writing if the numbers say more articles would not help: - **Search Console impressions collapse.** Your last 7 days get fewer than 40% of the impressions the previous four weeks predict. This only counts on a site that had at least 500 impressions in those four weeks. A drop like that usually means a Google update or a manual action. - **Google shows your new articles to nobody.** Your last three articles have been indexed for three weeks or more, and none of them has a single impression. When it pauses, the dashboard tells you which one happened, and we email you. Articles that were already written still publish. It doesn't start again by itself. Look at Search Console first, then press **Resume autopilot** on your dashboard's Home when you're ready. ## When something goes wrong Home shows a short note when something needs a look: articles that keep stopping, one that hasn't gone live in two days, keyword research that can't run. Problems on our side say so and fix themselves; for the rest, the note says what to do. A single blip doesn't show up at all. You can also pause it yourself at any time from the dashboard. ## What you control These settings are all you need to change. Everything else runs by itself. - **Pause and resume.** - **Brand.** Your tagline, short and long description, and categories. The writer uses the short description to decide what is on-topic, and research starts from the categories. - **Tone.** House rules the writer follows on every article, for example "Write in British English" or "Never mention competitors by name". Up to 2000 characters. - **Topics to avoid.** Up to 50. A topic blocks every keyword that contains all of its words, so "crypto" blocks "best crypto wallets". Ideas already queued under a new blocked topic leave the queue straight away. - **Suggest a topic.** Type a search phrase. It goes through the same checks as research (on-topic, not blocked, not a duplicate) and is either queued or turned down with the reason. - **Author.** A name, a short bio and a page on your own site. With a name, articles list that person as the author. Without one, the author is your site. - **Your blog.** Where articles go and how the hosted blog looks. - **Google Search Console** and **billing.** Every one of these is also an MCP tool (`get_autopilot`, `pause_writer`, `set_tone` and so on, listed under [MCP tools](/docs/mcp-tools#autopilot)), for anyone who would rather ask an AI to change them.