Small-Batch Custom Work: Why AI Skips Your Factory

For an export business, AI search optimization (GEO — Generative Engine Optimization) is about getting cited when AI answers your customers' questions. When a buyer asks "which factory does small-batch custom work", AI picks only from the information it can read.

One small detail first

The owner of a hardware parts factory in Shenzhen ran a test a while back.

He searched Doubao and DeepSeek for "which factory can do small-batch custom hardware parts". Both engines named competitors — two of them his long-standing rivals.

He assumed he simply had not advertised enough. The check said otherwise — the phrase "small batch" never appeared once on his own website. The product pages said "supports all kinds of customisation", and said nothing at all about minimum order quantity, sampling lead time or the range of materials he works in.

Yet all of it was right there in his workshop: minimum order quantity 500 pieces, three-day sampling, any common material you like. The owner knew it. His regular customers knew it. Only AI did not.

Step 1: Understand how buyers find factories now

Buyers used to find suppliers by going to a B2B platform and clicking down through category after category.

Now more and more of them hand a whole sentence to AI: which factory in China can make small-batch custom hardware parts? For parts with a 1,000-piece minimum, should I go to a factory or a trading company? How many days for sampling, and how is the sample fee calculated? Can this material be done in small batches?

Those questions share one thing: they carry specifications, quantities and process requirements — and not a single brand name.

The buyer does not know your name, and has no intention of learning it. What he wants is an answer he can act on directly.

That is the first thing AI search optimization has to deal with: what a customer throws over is a sentence, not your name.

Factories in the Pearl River Delta run small order volumes and heavy customisation, so the questions their buyers put to AI are unusually specific — exactly this shape.

Step 2: AI does not read your platform back office

AI can only read one kind of thing: information that has been written down as text and placed somewhere AI can reach it.

However thoroughly you fill in your product detail pages on a B2B platform, the platform's data is not necessarily open to AI and AI may not be able to retrieve it. If you cannot find yourself through these AI engines, it is not because you never wrote it down — it is because you wrote it somewhere AI cannot read.

The first step in GEO is to move your information from "somewhere AI cannot read" to "somewhere AI can".

Step 3: Write down what AI needs to verify

When a customer asks AI, AI is in fact checking a few things on the customer's behalf. You have to write those things down.

None of this has to be invented. You already do it all — you have simply never written it down.

One more thing, and it matters most: replace adjectives with numbers. Change "flexible MOQ" to "minimum order quantity 500 pieces"; change "fast lead time" to "three-day sampling, 15-day mass production". AI grabs numbers, not adjectives.

Step 4: Adapt per platform — don't blast one article everywhere

Post the same text verbatim to five platforms and AI flags it as duplicate content, which pushes your GEO citation rating down. ERNIE Bot reads Baijiahao, Doubao reads Toutiao, DeepSeek reads Zhihu and long documents, Tencent Yuanbao reads WeChat Official Accounts — you have to adapt each one separately.

Two and a half months later, that owner searched the same question in the same AI engine. His factory's name was in the answer.

One honest note to finish: nobody can promise you will appear in AI answers. What you can do is put accurate capacity figures, specifications and boundaries where AI can find, understand and verify them.

Frequently asked questions

We mainly do small batches, not bulk. Do we still need this?

More than ever. A small-batch customisation need cannot be expressed as a category, so the buyer can only ask it as a sentence — which is exactly the scenario AI handles best.

Won't publishing our MOQ and sample fees scare customers away?

The cost of hiding them is higher. If you keep them hidden, AI can only quote the numbers your competitors have published — which means handing the customer over.

Will AI pick up the store information on our B2B platform?

No guarantee. Platform data is not necessarily open to AI. What you control is your own website and public content platforms.

How long until it works?

No promises. The general pattern: on time-sensitive platforms such as Toutiao, weight concentrates in the first week or two; Baijiahao takes roughly three months; Zhihu and your own website keep working long term. It is ongoing work.

For a more detailed method, see www.chyihang.com/en/geo/.