From sign‑in to a measured change, in eight steps.

Under Your AI keys, add an Anthropic (Claude) key, an OpenAI key, or both. The provider bills you directly. We show only the last four characters, and there is no shared key: without one, drafting stays off.

Briefs come from the search terms people already use to find you. A new-post brief names the topic, the terms it targets and the sections it should cover. Dismiss one you do not want, with a reason.

Pick Claude or OpenAI and name the author. The draft is written in the background from the brief, your real search terms and your existing pages, in a minute or two. Each draft shows its cost; a workspace spends at most $25 a month.

The draft lists what a person must check: each fact the AI relied on, any number it did not declare, links to pages that do not exist and overlap with a page you already have. Tick each one and note how you checked it. Edit freely; a diff shows what you changed from the AI's version.

Once every flag is resolved, the post becomes a proposal like any other. A new page is always high risk, so it needs two approvers, neither of them the author. Deploy opens a pull request adding the post to your repository; it goes live when a person merges it.

Every crawl checks your robots.txt against the crawlers ChatGPT, Claude, Perplexity, Google and Bing use to fetch pages for their answers. AI search readiness lists any you shut out and what share of your pages each one is refused. Refusing the training-only crawlers (GPTBot, ClaudeBot) is your choice and is shown but never counted against you.

If your site has no /llms.txt, one click builds it from the pages the last crawl found indexable, with the titles and descriptions you already publish. It is a proposal like any other change: two approvals, then a pull request that goes live when a person merges it. Some AI tools read the file; no engine has said it uses it for ranking, so it is not scored.

Add up to eight questions, typed in or picked from the suggestions. Questions people actually searched in Google come first, with the impressions your site earned for each; question-shaped topics follow. Keep them the way a customer would ask; the question is sent exactly as written and never names your site.

Add a Claude, OpenAI or Perplexity key; each engine you have a key for is asked. Your provider bills you directly, and a workspace spends at most $10 a month on these checks. Perplexity is the cheapest per answer.

Enable the routine (weekly; daily is refused) or run it once. Each question gets a verdict per engine: cited, with your place among the sources; named but not linked; or not cited, with the sites cited instead. Under the grid, Who gets cited counts every source across the run's answers, with your site and your tracked competitors marked, and each source labelled as a forum, video, review site, reference or other, with what to do about it. Compare runs over weeks rather than trusting any single one.

Answer engines quote the passage directly under a question. Each crawl flags question headings with no 15–100 word answer beneath them, answered questions with no FAQ markup, and FAQ markup that lists questions the page does not show. For answered questions without markup, Draft proposal builds FAQ markup from the page's own headings and paragraphs, with no AI, as a pull request. For questions without an answer, Draft direct answers has your own Claude or OpenAI key write one answer per question; you check every claim and figure, then send it for approval. It becomes a high-risk edit to that page's source file, with each answer placed under its heading and FAQ markup built from them, and it goes live only when a person merges the pull request.

Most searches end without a click, so the zero-click scorecard tracks what clicks miss, comparing the latest four weeks with the four before: searches that name your site, the share of impressions answered on the results page, direct visits from GA4, and how often AI answers name you. Brand terms come from the site's name and domain; replace them if people search for you another way. It describes a trend and never claims a change caused it.