What AI Thinks Your Factory Makes

01The case

The company in this case is anonymized, but the facts are from our own measurement work. It is an established manufacturer of home appliances — the kind of factory Cixi is known for — selling to overseas buyers under its own brand. It has a website, certifications, and a product line it has made for years. By traditional measures it is perfectly visible: search its name and you find it. Then we asked AI systems, in buyer language, what this brand is and what it makes.

02Three systems, three wrong companies

Error one: the wrong industry. One system described the brand as a maker of industrial laundry equipment — commercial machines for laundromats and hotels. The factory makes household appliances for homes. The confusion is understandable in hindsight: the product category sits near laundry, the brand's English footprint is thin, and the system reached for the most documented neighbor category and stopped there.

Error two: the wrong name. A second system rendered the brand's name as a different, similar-sounding name — effectively attributing the company to another business entirely. Every fact it then produced belonged to the wrong entity. For a buyer, this is worse than invisibility: the inquiry that should have reached this factory is now describing someone else.

Error three: the wrong business model. A third system, asked in Chinese, guessed the brand was an online shop selling small appliances — a retailer, not a manufacturer. The factory's single most important fact, the thing its buyers care about most, was not merely missing. It was replaced.

Now read the demo scores again with those errors in mind. Mention rate 100% — the name appears. Share of voice 54.5% — the name appears often, in more than half the category conversation. Citation rate 0% — not one answer used the company's own pages as a source, which is exactly why the descriptions drifted: with no citable statement of its own facts, each system improvised from fragments. High visibility, zero accuracy. If we had reported only the first two numbers, the client would have celebrated.

03Why this happens

AI systems do not look brands up in a registry. They reconstruct an identity from whatever text about the brand exists where they can read it. When the third-party footprint is thin — few directory entries, little trade press, no consistent descriptions — the reconstruction fills gaps with priors: the nearest category, the nearest similar name, the statistically most common business model. The brand's own site, if its facts are not written in a form that can be lifted and quoted, does not correct the record, because the systems were never reading it as evidence in the first place.

04The cognition audit method

The fix starts with a different audit, and it runs before any visibility work. Our brand cognition audit has three steps. Ask directly: put buyer-style questions about the brand itself to each platform — what does this company make, where is it, who does it sell to — in the languages its buyers use. Verify claim by claim: break each answer into individual factual claims and grade every one against ground truth — right, wrong, or invented. Trace the source: for each wrong claim, work out where the system likely got it — a namesake company, a stale directory entry, a category assumption — because the fix differs by origin. A namesake collision needs disambiguation signals; a stale listing needs correction at the listing; a pure invention needs the fact published, quotably, where systems actually read.

Only after cognition is repaired does visibility work mean anything. There is no point earning more mentions for a brand the answer describes wrongly — each new mention just repeats the error to another buyer. Identity first, then exposure, then recommendation. That order is not a preference; this case is what the reverse order produces.

© 2026 Jun Liu · JUN & GEO. This article is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0) — https://creativecommons.org/licenses/by-sa/4.0/. You may share and adapt it with attribution and under the same license.