| name | geo-content-operator |
| description | Generate GEO-optimized (Generative Engine Optimization) B2B export articles that get cited by AI search (ChatGPT, Google AI Overviews, Gemini). Produces an inverted-pyramid fact-dense lead, a comparison table, buyer FAQs, and Article + FAQPage JSON-LD. Use when writing SEO/AI-search content for a manufacturer, exporter, or B2B supplier. Two modes — local (bring-your-own model, this skill is the method) or hosted (set PAIBAO_LICENSE_KEY to ground generation in a dynamic knowledge base). |
GEO Content Operator
Write B2B export articles that AI search engines cite, not just rank. The difference
between content AI Overviews ignore and content they quote is fact density + structure +
entity disambiguation. This skill is that method.
When to use
A manufacturer/exporter/B2B supplier needs an article that wins AI-search visibility for a
buyer-intent topic (e.g. "how to choose axle configuration for a fuel tanker trailer").
Two modes
- Local (open, default): you (the model) follow the method below. No account needed.
- Hosted (pro): if
PAIBAO_LICENSE_KEY is set, run scripts/generate.sh "<topic>" <lang>
to call the hosted engine, which grounds the article in the client's dynamic knowledge
base (tacit, selection-driving facts elicited via expert interview) — the differentiated,
competitor-proof layer this local method cannot reproduce. Returns html + markdown + jsonld.
The method (local mode)
Before writing, get the business dossier: real product lines + specs, target markets,
and — most important — the client's tacit selection knowledge (the things they treat as
common sense but buyers get wrong). If you don't have it, interview the user for it first;
that hidden knowledge is what makes content citable.
Then produce, in order:
-
Inverted-pyramid definitional lead (~60–90 chars, one sentence). Lead with the single
most specific figure. This is the sentence AI quotes. No adjectives, no marketing.
e.g. "A 3-axle 50CBM side tipper exceeds Kazakhstan's first-checkpoint axle-load limit above 24t payload."
-
Body, 900–1300 words, ##/### sections shaped as query fan-out (answer the
sub-questions a buyer would ask). Quantitative claims only — every claim carries a number,
spec, mm, tonnage, %, or standard. Include at least one comparison/spec TABLE.
-
"## What most buyers get wrong" — 160–220 words of the client's tacit knowledge, exact
figures verbatim. This is the differentiated, AI-citable insight competitors lack.
-
Buyer FAQs — 3–5 real RFQ-stage questions with fact-dense answers. These map 1:1 to
FAQPage schema.
-
A proprietary data note — your single most specific spec/number, framed as a curiosity
gap (in hosted mode this is rendered with data-nosnippet).
-
JSON-LD entity disambiguation — emit Article + FAQPage + publisher Organization
(see references/jsonld.md). This tells AI engines exactly what entity the content is about.
Rules (what makes it GEO, not generic SEO)
- Fact density over prose. If a sentence has no number/spec/standard, cut or sharpen it.
- No "everyone knows" filler. Surface the non-obvious selection rule.
- Quote exact figures verbatim in the lead and ≥2 body sections.
- One canonical entity — consistent product/spec naming throughout + in JSON-LD.
- Buyer intent, not keyword stuffing. Sections answer real purchase questions.
Output
Emit: the article as Markdown (or HTML), plus the JSON-LD block. In hosted mode the
script returns all three (html/markdown/jsonld) ready to paste into any CMS; the JSON-LD goes
in <head> as <script type="application/ld+json">.
See references/method.md for the full dimension checklist and references/jsonld.md for the
schema templates.
