| name | aeo-blocks |
| description | Use when you want a content page to get cited by AI answer engines (ChatGPT, Perplexity, Google AI Overviews, Gemini, Copilot). Generates the extractable answer blocks and structured data that get pulled into AI answers: at-a-glance answer block, FAQ, HowTo, comparison table, statistic blocks, llms.txt, and the matching JSON-LD. Trigger on "AEO", "GEO", "answer engine optimization", "get cited by AI", "optimize for ChatGPT/Perplexity", "AI Overviews", "llms.txt", or "AI citations". |
| metadata | {"version":"1.0.0","pack":"CitedStack"} |
AEO Blocks — get the page cited, not just ranked
AI answer engines do not cite pages. They cite passages. Your job is to make
every key claim on a page a self-contained passage an LLM can lift verbatim into
an answer, and to describe that page in structured data the engine can parse
without rendering JavaScript.
This skill turns a topic or an existing page into the blocks that actually get
cited, plus the JSON-LD that backs them. It is the exact block system running at
full coverage on a small affiliate site whose #1 referrer is ChatGPT — ahead of
Google Search.
Traditional SEO gets you ranked. This gets you cited. A well-structured page
gets cited even when it ranks on page 2, because engines select sources on
extractability and authority, not rank alone.
The one rule that matters most
Lead every section with the answer, in 40 to 60 words, phrased so it survives
being copied out of the page with zero surrounding context.
If the first sentence under a heading only makes sense after reading the two
paragraphs above it, an LLM cannot lift it, and it will cite a competitor who
front-loaded the answer. Front-load, then elaborate.
What to do
When asked to optimize or generate a page for AI citation, produce these in order.
Do not skip the audit — generating blocks for a page that blocks GPTBot is wasted work.
1. Confirm the target query
One page, one primary query, phrased how a human actually asks it:
What is [X]? → definition blockBest [X] for [use case] → ranked list + comparison table[X] vs [Y] → comparison table (this format wins ~33% of AI citations)How to [X] → HowTo block + HowTo schema[X] price / cost → a specific number with a source and a date
Match the H1 and the first H2 to the query wording. Engines pattern-match headings
to prompts.
2. Access check (do this before anything else)
An engine cannot cite what its crawler cannot fetch. Verify robots.txt does not
Disallow these, and that the page is server-rendered (or statically rendered),
not gated or client-only:
| Engine | Bot to allow |
|---|
| ChatGPT | GPTBot, ChatGPT-User, OAI-SearchBot |
| Perplexity | PerplexityBot |
| Claude | ClaudeBot, anthropic-ai |
| Google AI Overviews / Gemini | Google-Extended |
| Copilot | Bingbot |
If any are blocked, flag it as the #1 fix. Nothing else matters until crawlers can read the page.
3. Generate the blocks
Use the templates in references/block-patterns.md.
For a typical "best X" or "X vs Y" page, generate in this order:
- At-a-glance answer block — the 40 to 60 word direct answer, right under the H1.
This is the single highest-leverage block. It is what gets quoted.
- The pick / verdict block — one-line answer to "which one", with the reason.
- Comparison table — structured, not prose. Columns are the decision criteria.
- Statistic blocks — every claim with a number gets a source and a date.
Statistics with sources are the single biggest citation lever (+37 to +40%,
Princeton GEO study, KDD 2024).
- FAQ block — 4 to 8 real questions in natural language, each answered in
one self-contained paragraph. Back it with FAQPage JSON-LD.
- HowTo block (only if the query is procedural) — numbered steps, backed by
HowTo JSON-LD.
4. Add the structured data
Emit the JSON-LD from references/schema-templates.md
that matches the blocks you generated. Content with matching schema shows 30 to 40%
higher AI visibility. Rules:
- The JSON-LD must match the visible content. Fabricated schema (FAQ entries
not on the page, ratings with no reviews) is a manipulation signal and a
compliance risk. If it is not on the page, it is not in the schema.
- Always include
Article/BlogPosting with a real author and a dateModified.
FAQPage for the FAQ block, HowTo for procedures, ItemList for ranked lists,
Product only when the page is genuinely about one product with real, sourced fields.
5. Emit the machine-readable files
See references/machine-readable-files.md.
At minimum generate or update /llms.txt. For any product or pricing page, add
/pricing.md so AI agents comparing options can parse you without rendering the page.
6. Freshness and authority pass
- Put a visible
Last updated: [date] on the page (and mirror it in dateModified).
- Every statistic gets a named source. "According to [source]" framing beats a bare number.
- Name a real author with a real, relevant one-line bio. Anonymous pages lose to attributed ones.
- Do not keyword-stuff. Unlike classic SEO where stuffing is merely useless,
it measurably reduces AI citation (about -10%, Princeton GEO).
The nine levers, ranked (Princeton GEO study, Perplexity, KDD 2024)
Apply top-down. The first three account for most of the gain.
| Lever | Visibility boost |
|---|
| Cite sources (authoritative refs with links) | +40% |
| Add statistics (specific numbers with sources) | +37% |
| Add quotations (expert quotes, named) | +30% |
| Authoritative tone | +25% |
| Improve clarity | +20% |
| Domain-specific terminology | +18% |
| Unique vocabulary / word diversity | +15% |
| Fluency / readability | +15 to +30% |
| Keyword stuffing | -10% (do not) |
Best combination: fluency + statistics. Low-authority sites benefit more — up to
a ~115% lift from citations, which is why this works on a page-2 site.
Output format
When you run this skill, return:
- The ready-to-paste content blocks (markdown or JSX, matching the project's stack).
- A single
<script type="application/ld+json"> block (or a JSON object for the
project's schema helper) containing all applicable schema.
- The
llms.txt (and pricing.md if relevant) content.
- A short checklist of what still needs a human: real author bio, real sources for
any placeholder stats, and the access/robots fix if one was found.
Never invent a statistic, a source, a price, or a review to fill a template. An
unfilled placeholder marked [NEEDS REAL SOURCE] is correct; a fabricated number
is a defect. When a page makes product, price, or factual claims, run the
factcheck-verifier agent over it before publishing — getting cited is worthless
if the cited claim is a lie, and one fabricated claim on a money page is how you
lose the citation permanently.
