The levers and their playbooks
Every playbook below drafts or proposes. Nothing is published, sent or posted without the step your autonomy level allows, and every claim it writes has to trace to a confirmed Brand Book fact or a page it read in the same run.
Pages that answer
write.comparison-page: a new “X vs Y” page. Your column comes from confirmed Brand Book facts; the rival’s column comes only from their own pages, read in the same run and dated. A rival fact we could not read is written “not stated on their site”, never guessed. The rival wins at least one row, and the page says when to pick them.write.alternatives-page: a new “alternatives to X” page. The ranking criteria come before the list, every entry carries one sourced fact, your product ranks where the criteria put it, and the page says you make one of the products listed. With fewer than three alternatives it can source, it writes no list and tells you.write.stats-page: a statistics page for your category. Every figure is read from the organisation that measured it, written exactly as published, with the year it describes and the date it was read. Fewer than eight such figures, and no page is written.write.answer-block: the one passage near the top of a page that says what you are and who it is for, in the words the answer engines already use for the winners. It replaces or moves one passage and nothing else.write.rewrite-for-extractability: rewrites one passage that leans on “it” and “this” so it names its subject and states its fact in one sentence. It adds no fact, number or customer that is not already there.
Internal links
link.answer-page-inbound: picks up to three of your pages that answer engines already cite, or that their bots already read, and the sentence on each where a link to the page that should answer belongs. It never claims how many links a page already has: we do not hold a full link map of your site yet.link.insert-links: turns words already in that sentence into a link. It never adds words, never touches navigation or footers, and checks the target page loads first.
Bots and URLs
technical.bot-access: reads your robots.txt and each page’s robots tag per answer-engine bot. Blocking a training bot (GPTBot, ClaudeBot, Google-Extended) costs no citations, and it says so; blocking a search bot (OAI-SearchBot, Claude-SearchBot, PerplexityBot) is the finding. It proposes removing a search-bot block only when your Brand Book’s policies do not say you chose it, and changes one group at a time.technical.canonical-and-redirects: when one page answers at two addresses, it picks the one answer engines cite most and proposes a canonical on the other; it fixes internal links to pages that return 404. Redirect chains are not visible to our page reads, so it says “not measured” rather than guessing, and a server redirect is a task for you.
Other people’s pages
These draft only. They never send, never post on a forum, community or Q&A site, never ask a reviewer to change a review, and never offer payment, a link swap or a free account in exchange.offpage.listicle-inclusion: a ranked list that leaves you out. It writes your entry in the list’s own format from confirmed facts, with an honest limitation when the list’s entries carry one, and drafts one short note to the editor the page names. With no published contact it writes the entry and no note. A list that sells placement is not this playbook: a paid placement is a separate, disclosed decision for you.offpage.mention-fix: a page that names you with a wrong fact, a dead link or no link. It quotes their sentence, sets it against your confirmed fact and the public page that proves it, and drafts one correction. A figure that could be right for an older plan or period is not corrected.
Your company facts
offpage.entity-sync: reads each profile of your company that shows a fact (Wikidata, Crunchbase, LinkedIn, your directories), quotes what it says, and drafts the copy kit you file from your own account. It uses confirmed values only, never asks for a password and never signs in. Entity sync’s Changes carry this playbook.technical.structured-data: schema.org markup (Organization, Product or SoftwareApplication with Offer, FAQPage) built only from confirmed Brand Book facts and text visible on the page. It removes markup that claims what the page does not, such as a rating with no reviews on the page. Its first line says what the evidence says: markup is hygiene, not a citation lever, because live-fetching assistants read the visible text.technical.llms-txt: an llms.txt from confirmed facts and pages read in the same run, for the agents and developer tools that read one. Its first line says that most llms.txt files are never requested and answer-engine bots rarely fetch them. It never writes an instruction to AI models into the file. If your product feed is on, it matches it.
Run one yourself
GET /v1/playbooks lists every playbook, and POST /v1/playbooks/{name}/runs starts one, for example technical.bot-access. Over MCP, list with type: "playbook" lists them and the playbook tool runs one. See the endpoints.
A playbook’s work is held to its own quality checks. These playbooks are new: none has passed its checks on the 20 examples it needs, so each runs no further than L3, “Ship after 48 hours unless stopped”, whatever your lane is set to.