Service structure

Five building blocks, and the order they actually pay off

Most of the work in this category is not mysterious; it is sequenced badly. Content gets written before the facts agree, profiles get bought before there is a page worth pointing at, and translations go live before anyone has read them. The five blocks below are the ones that change what a model can say about you — each with the failure mode of running it on its own.

Find your weakest block Reference pricing in USD

Overseas reference price starts at US$1,800–4,000 per month, minimum term 3 months. Scope and fees are set per market, language and content volume in a written proposal.

Why five, and why this order

Each block assumes the one before it

Marking up facts that disagree with your own pages hands a model two candidate values and no way to choose. Translating a fact sheet nobody has agreed multiplies the error into every language you publish.

Usually one block is missing, not five

In most reviews one block is already adequate and another is absent entirely. Adding a fifth piece of content to a site whose specification table is wrong changes nothing in any answer.

A fixed list keeps quotes comparable

Because the blocks are the same five every time, a proposal can be read as a statement about which blocks you are buying — and a second quote can be compared with the first.

The five blocks, one at a time

Each block is described by what it does to an answer, not by what it is called. Nothing below is a client result or a measured uplift — these are mechanisms, deliverables and the check you can run on each.

01Machine-readable facts

The same values, in named fields

What a model does with it

An assistant answering “who supplies this component with that certification” is looking for a value attached to a named field, not a paragraph that implies it. Structured data does not make a claim true — it makes the claim addressable: this number is the minimum order quantity, this string is the certificate number. Without it a model that has read your page can still answer. It simply fills the gaps itself, and a filled gap is delivered in exactly the same tone as a fact.

How you check it

Open the page source, search for application/ld+json, and compare three fields against your specification table: one dimension, one certification number, one price. If any of the three differs, the markup is wrong no matter what a validator reports.

What it produces

JSON-LD for the organisation, product lines and offers, with every value copied from your own source documents rather than from the marketing page
A field-by-field diff between the markup, the visible page and the source document, listing what disagreed
Validator output and a rendered screenshot, filed with the round record so the state is comparable later

Failure mode when run alone

Markup that contradicts the visible page is worse than no markup. The model now holds two candidate values with no basis for choosing between them, and which one it repeats is neither predictable nor correctable after the fact.

02Decision-stage content

Pages that answer a question a buyer has actually typed

What a model does with it

An assistant assembles its answer from sentences that already resemble an answer. A page headed “our advantages” contributes nothing, because no buyer asked that. A page stating that one model suits workshops under a given floor area while the other needs three-phase power can be quoted almost intact. The unit to write is the decision, not the product.

How you check it

Run your fixed question set, then read the answer for a sentence that could only have come from your page. A generic answer means the page was generic. A competitor being cited means they wrote the decision you skipped.

What it produces

Comparison pages, specification explainers and scenario pages, each carrying the question it answers in its first line
One page per real fork in the buying decision, not one page per keyword
A list of the questions the pages still do not answer, carried into the next round instead of being quietly dropped

Failure mode when run alone

Volume with no decision in it. Forty posts about industry trends, seasonal greetings and award announcements answer none of the questions buyers ask, and a model has no reason to prefer them to silence.

03Surfaces you publish on

Your own words, on domains you do not control

What a model does with it

A model's picture of a company is assembled from more than the company site. Trade directories, association member lists, marketplace pages, review platforms and professional profiles are all readable, and each one that states the same fact independently raises the chance that the fact survives into an answer. This block is about publishing on those surfaces deliberately, rather than waiting to be scraped.

How you check it

Open each profile in a private window without logging in. If you cannot read it, neither can a crawler. Then search one exact phrase from the profile to see whether the platform page is indexed at all.

What it produces

A shortlist of surfaces where your buyers actually look, each checked for whether it is publicly readable at all — a login-walled profile contributes nothing to a crawler
Profile text and answer-form content kept to the same facts as the site, written in the same voice rather than a second marketing dialect
A register of live links with the date each was last confirmed, because profiles lapse silently

Failure mode when run alone

Profiles that repeat marketing copy. A second copy of an unverified claim is not corroboration — it is the same claim twice, and a model that deduplicates by wording counts it as one source.

04Language and format coverage

The same facts in the language the question is asked in

What a model does with it

A buyer asking in German receives an answer built mainly from German-language material. If your specifications exist only in English, you are not a weaker candidate for that answer — you are absent from the material it was built from. Format works the same way: a specification that lives only inside a PDF attachment, or only in a narrated video with no transcript, is effectively unpublished.

How you check it

Ask your own question set in the target language and count how many answers mention you at all. Then spot-check one translated page: do the units, model numbers and certification codes still match the source? If the numbers moved, nobody read it.

What it produces

Reviewed translations of the fact sheet and of the pages that carry the decision, read by someone fluent in the target language rather than accepted from a tool
Transcripts and subtitles for product video; alt text and captions for specification images
Language declarations that match which pages actually exist — a mismatch here fails silently, and is one of the most common faults we find

Failure mode when run alone

Machine translation published without review. Units convert wrongly, model suffixes disappear, and a certificate code that shifts by one character becomes a false regulatory statement in a language you cannot audit.

05Third-party corroboration

Someone else stating the same fact

What a model does with it

This is the block that changes the confidence of an answer rather than its content. Your own page is a claim; your page plus a trade-body record plus an independent publication saying the same thing is a fact a model will repeat without hedging. It is also the slowest block, because it depends on other people's editorial decisions and other people's timelines.

How you check it

Search an exact claim in quotation marks. If only your own domain returns it, the claim is uncorroborated however prominently you display it. Then look at the sources that do return it: independent, or the same press release on five sites?

What it produces

An inventory of third-party sources that already mention you and what each one says — including the ones that say something wrong
Corrections and additions pursued through each body's own process: registry updates, member listings, directory edits, each with the outcome recorded
New coverage only where an outlet's rules permit it, with the submission and its result logged either way

Failure mode when run alone

Paid placements that read as advertising, and syndicated text across outlets. Five copies of one press release is one source wearing five hats; models that cluster by content treat it as one, and buyers do too.

The order is a diagnosis, not a preference

Which block goes first follows from what is currently wrong. Find your row: the two columns on the right are the first two months of work, and the remaining blocks can wait without loss.

Starting states, how each one shows up, and which two building blocks to run first
Your starting state	How you notice it	Run first	Then
Your own pages disagree with each other	Two pages give different order quantities, or the PDF and the product page differ	01 Machine-readable facts — agree the source, then mark it up	02 Decision-stage content
The site is consistent but thin	An assistant describes your category correctly and your company vaguely	02 Decision-stage content	03 Surfaces you publish on
The site is fine and nothing else confirms it	Answers hedge, or attribute your claims to no one in particular	05 Third-party corroboration	03 Surfaces you publish on
You exist in one language only	Questions asked in the buyer's language never mention you at all	04 Language and format coverage	02 Decision-stage content

If you cannot tell which row you are in, that is what a first round is for: agree a question set, record the answers as they stand today, and read them. The gaps are usually visible within one sitting — and a record you ran yourself is more reliable than an audit report you were handed.

What none of the five will fix

The blocks are deliberately narrow. Saying what they do not cover is the only honest way to describe them.

A product buyers do not want. This work changes whether you are describable, not whether you are preferable.
A category nobody asks about. If no one types the question, there is no answer to be part of — worth checking before commissioning anything.
A reputation problem. Models repeat what sources say; if the sources say something bad and true, clearer markup makes that clearer, not less clear.
A slow or uncrawlable site. If pages take ten seconds to render, the content is not read, and nothing downstream matters.
A guarantee. No block buys a citation, a position inside an answer, or a date by which either happens.

Where this leads

The artifacts and their checks

Each block produces specific deliverables, and every deliverable has a check you can run without us. That list is on the deliverables page, item by item.

Deliverables and acceptance

The same five, by industry

If you would rather see the blocks expressed as the questions buyers in your sector ask, four sectors are written out with their question sets and first build.

Industry solutions

What each block costs

Unit prices for markets, languages, platforms, content volume and question count, with three worked examples you can recompute yourself.

Reference pricing

If you would rather read the same five blocks arranged by where the buyer is in their decision — problem, comparison, purchase, usage, expansion — the buying-journey page walks them stage by stage, with the asset each stage needs and the record it leaves behind.

Which one is weakest for you?

Send your website and one sentence about what you sell and where. We will say which of the five we would start with and which we would leave alone — including when the honest answer is none of them, and the product page needs fixing first.

Email us Reference pricing

We answer in English. Invoicing currency and payment terms are specified in the proposal.

Operated by 南宁超级月亮跨境电商有限责任公司 (registered company name in Chinese).

Five building blocks, and the order they actually pay off

Why five, and why this order

Each block assumes the one before it

Usually one block is missing, not five

A fixed list keeps quotes comparable

The five blocks, one at a time

01Machine-readable facts

What a model does with it

What it produces

02Decision-stage content

What a model does with it

What it produces

03Surfaces you publish on

What a model does with it

What it produces

04Language and format coverage

What a model does with it

What it produces

05Third-party corroboration

What a model does with it

What it produces

The order is a diagnosis, not a preference

What none of the five will fix

Where this leads

The artifacts and their checks

The same five, by industry

What each block costs

Which one is weakest for you?

Ask AI about moonai-geo.com