SoloSwarm blueprintsWhen to use it
Your pages rank in classic search but rarely get cited in AI answers. Generative engines prefer pages that answer a question in the first lines, state facts consistently, and are easy to parse.
Swarm layout
| Role | Does |
|---|---|
| Researcher | Lists the 20–40 questions real visitors ask (from Search Console queries, support emails, on-site search), maps each to the page that should answer it |
| Editor | Rewrites the opening of each mapped page into a direct answer, adds FAQ blocks and structured data, writes llms.txt |
The researcher’s output is a table the editor works through; nothing is published until a human reviews the PR.
Editor brief
Task: GEO pass for <SITE>. Input: docs/geo-questions.md (question -> page).
For each row:
- Make the first 2–3 sentences of the page answer the question directly, with concrete facts
(prices, durations, places, limits) that already appear elsewhere on the site. Never invent facts.
- Add or update a FAQ section with that question, and FAQPage JSON-LD that matches the visible text.
- Use the same names for the business, products and places everywhere (no variants).
Also: add /llms.txt listing the key pages with one-line descriptions.
Build, open a PR, and list any question you could not answer from existing content.Approval rules
Default rules are enough: pushes and deploys wait for you. Treat any new factual claim the editor adds as a review blocker.
Done when
- Every mapped question has an answer in the first lines of its page
- FAQ JSON-LD validates and matches visible text
-
llms.txtexists and lists only live pages - A list of unanswered questions exists for a human to decide on
Lessons from real runs
- The biggest wins came from consistent entity names (one spelling of the business and its products) rather than from markup.
- Don’t let the editor “improve” facts. Answers must come from content you already stand behind.