Answer-engine content

Produce copy built from units an engine can extract, attribute and trust: a direct answer, headings that match the questions people ask, claims with named sources, and structure a parser can read. New page or rewrite, with a note on why each block is there.

When to use

• "Write a page that ChatGPT / Perplexity will cite for {question}."
• A page scored low on answer shape or citability in geo-audit.
• A gap from competitor-gap that needs a page of its own.

Not for: scoring a page (geo-audit), JSON-LD (schema-markup), title, meta and internal links (on-page-seo), a planned Era article (content-brief, then article-review).

Inputs

• Target query — the question this page should be the answer to. For a rewrite, derive it from the H1 and say so.
• Page — new, or a rewrite (a URL to fetch, or pasted text).
• Page type — guide, comparison, product or landing, glossary, how-to, FAQ.
• Facts and sources — what the user can stand behind: numbers with source and date, product facts, quotes, author name and credentials. This skill never supplies them itself.
• Brand (optional, Era) — with --brand, pull the brand kit and tone of voice.

Steps

1. Take inventory. List the facts the user provided, each with its source and date, plus the author and the last-reviewed date. Whatever is missing becomes [SOURCE NEEDED] or [FACT NEEDED] in the draft. Ask once for the highest-value gaps: one sourced number, the author, the date.
2. Read the current page (rewrite only). Extract the H1, the first paragraph, the headings, every claim and number, the tables and lists, the author and the dates. Keep all facts, links and names. If the body holds <IMAGE_INFOGRAPHICS_N>…</IMAGE_INFOGRAPHICS_N> markers (an Era article), keep every one in place.
3. Era context (optional, when connected and a brand is given). list_brands → brand_id. get_brand_kit for about_brand, ideal_customer_profile and brand_point_of_view (each may be null). list_tones for the brand's default tone and its master_prompt. list_keyword_clusters and list_keywords(search=<topic>) for target phrases and their ai_search_volume. list_search_queries(search=<topic>, sort_by="count", order="desc") for the exact sub-queries engines run. list_cited_domains for the domains the engines already cite in the category: candidates for outbound references, used only where the source really supports the sentence.
4. Map the questions. Turn the target query and the sub-queries into four to eight questions in reading order: the answer itself, the definition, how or why, the comparison, cost and limits, and the adjacent questions for the FAQ.
5. Draft block by block with references/page-skeleton.md, applying the patterns in references/geo-writing-patterns.md. The order of the page is the order of the questions.
6. Check the draft against the pattern checklist at the end of references/geo-writing-patterns.md. Every number has a source and a date, every name is written one way throughout, and every claim is one the user can defend.
7. Write the output below and name the hand-offs.

Output

Lead with a five-line summary:

Page · target query · mode (new / rewrite) · words before → after
Direct answer  in the first two sentences (was: paragraph 4)
Evidence  6 statistics with source and date · 3 marked [SOURCE NEEDED]
Structure  9 question headings · 1 comparison table · 1 step list · 6-question FAQ
Hand-offs  schema-markup (Article, FAQPage) · on-page-seo (title, meta, internal links)

Then:

1. The draft, in Markdown, ready to paste. For a rewrite, show each changed block as before and after, then the full rewritten page.
2. Why this block is here — a table: block · pattern it serves · what an engine can do with it. One row per block, in page order.
3. Open items — every [SOURCE NEEDED] and [FACT NEEDED], each with the claim it protects. The page is not ready to publish while any remain.
4. Hand-offs — to schema-markup (state the types: Article or Product, and FAQPage only when the FAQ is visible on the page), to on-page-seo (title and meta description built from the direct answer, internal links to the sub-question pages), and to geo-audit to re-score the live page.
5. How you will know it worked. After publishing, the target query should be a monitored prompt in Era. In four to six weeks re-read its answers (list_answers, get_answer) and look for the page URL among the citations and for the brand in the text. Without Era, ask an engine that shows sources the same question a few times. Answers vary between runs, so judge by several.

Guardrails

• Never invent a statistic, quote, study, customer, award, price or date. Use a placeholder and ask. A made-up number that gets cited does more harm than a missing one.
• Say a claim is sourced only when the source supports that exact claim.
• Write for the reader first. No keyword stuffing, no hidden text, no sentences that exist only for an engine.
• The tone of voice sets the voice, not the structure: keep the direct answer first even when the tone prefers a slow opening.
• Comparison pages state their criteria and stay accurate about competitors, including where a rival is better when that is true.
• Health, legal and financial claims need primary sources and a named expert reviewer.
• No promise of citation. Engines choose their sources per query and vary by run. A rewrite raises the odds that a page is usable, nothing more.
• This skill drafts. It does not publish, and it does not call propose_article_edit; for an Era article, hand the rewrite to article-review.

Era tools used (all optional, all read)

• list_brands, get_brand_kit, list_tones — brand, positioning, voice.
• list_keyword_clusters, list_keywords — target phrases and AI volume.
• list_search_queries — the exact questions engines run.
• list_cited_domains — sources the engines already trust.
• list_answers, get_answer — the after-publication check.

Without Era

Everything but step 3 runs on what the user provides. Questions then come from the page's own headings, support tickets, sales calls and the related questions an engine or search page shows; sources come from the user. Skip the brand-kit and tone lines and note that Era would supply the exact sub-queries and the domains the engines already cite.