When the story fights its only public record, GEO can only amplify the doubt
A B2B-marketplace storefront we turned down: five findings checkable on its own profile page, three sets of public data on how answer engines verify, and why we would rather lose the deal — GEO on that foundation hurts overseas buyers first, and the answer ecosystem in the end.
Miaowa GEO earns its revenue from engagements, and publishing a refusal is not good business. We write it anyway, because this industry is in a dangerous window: "guaranteed AI recommendation" pitches are already circulating, and once the answer ecosystem is polluted by reputation-packaging, the bill goes first to overseas buyers, then to every Chinese factory that does the patient work of getting its facts straight. In the third quarter of 2026 a merchant approached us about GEO. Its entire online presence was one storefront on a major B2B marketplace — no website of its own. We completed the assessment and declined, not because AI "hadn't seen it yet" but for the opposite reason: on its current foundation, the more clearly AI sees it, the more contradictions come into view. This is the second declined assessment we have published, for the same reason as the first: "which businesses should not be doing GEO yet" deserves a public answer. We can afford to lose a deal; we cannot afford to lose the standard. Every finding below is one you can check against your own storefront profile.
What we observed
The storefront describes itself as a specialist manufacturer, while the "business type" field on the same profile page reads trading company. The years of experience it claims predate its own stated founding year by more than a decade, and the industry in its registered name does not match the categories it actually sells. All three contradictions sit on a single page — one scroll and a buyer has seen them all.
How to check this yourself：Open your own storefront profile and check every claim in the company introduction against the structured fields on the same page — business type, founding year, category scope.
B2B marketplaces ship a whole trust layer for buyers — verification badges, certificate sections, patent and trademark records, transaction history. On this storefront every one of those fields is empty and no manufacturer-verification badge is present. With the marketplace's "verified manufacturers only" filter on, a name search does not return the shop; it appears only when the filter is switched off.
How to check this yourself：Count how many verifiable fields on your profile — certificates, patents, trademarks, transaction history — are non-empty, then search for yourself with the marketplace's "verified" filter on. That filtered view is what buyers see.
The introduction claims a range of over a hundred models, major group clients and exports to multiple countries — yet every verifiable section of the profile is empty and the platform tenure badge reads one year. Not one heavyweight claim lands on a checkable field. To an answer engine this is not "incomplete data"; it is a claims-to-evidence ratio out of balance.
How to check this yourself：Take the three strongest claims in your introduction and ask: in which checkable field or third-party record could a buyer — or an AI — confirm each one?
Throughout the assessment we were unable to obtain verifiable operating-entity information — registered name, registration record or address, none of it available to corroborate the storefront's own story. For a merchant whose introduction claims many years in the trade, this is the easiest gap to close, and the most telling one left open.
How to check this yourself：Ask yourself: if a buyer or a service provider asked right now for entity documents that match your storefront's story, could you produce them within ten minutes?
Asked about the company by its English name, a mainstream AI assistant answered that no reliable public information about it exists — noting that this absence is itself a signal worth attention — and then offered two similarly named companies that do have public records. A diligent overseas buyer checking it today either finds nothing, or gets handed someone else.
How to check this yourself：Ask a mainstream AI assistant "what do you know about this company" using your English name, and see whether it anchors to you — or steers the buyer to a similar-sounding someone else.
1 other site assessed in the same period showed the same pattern. This describes a category, not one company.
The five findings point at one thing: generative engines answer buying questions in a way ranking-era search never did — they verify the entity first, cross-checking a company's story against its profile fields and third-party records, and only what survives verification can appear in an answer. Buyer behaviour is shifting in step: the due diligence that used to happen after an inquiry is now done for the buyer in the seconds it takes an AI to answer. We tested where this merchant stands. Asking as a buyer — even reusing the category wording from its own page — neither of two mainstream AI assistants put it on the candidate list, and every company they did list has a public record. Asked about it by name, the AI answered that no reliable public information exists, then offered two similarly named companies that do. Its real problem is not that AI doesn't know it — it is that the moment AI does, the first thing AI reads is the merchant's story fighting the merchant's own profile.
GEO amplifies verifiable evidence. When the evidence fights itself, what gets amplified is doubt.
Why this does not work in GEO terms
- Before recommending a supplier, a generative engine verifies the entity: it cross-checks the story against profile fields and third-party records, and cites only what agrees across sources. A story that contradicts its own profile on the same page is merely "poor content" to ranking-era search, but a net liability to an answer engine — which will either fail to anchor the company, or faithfully relay the contradictions to the buyer, or mistake a similarly named business for it. None of the three outcomes helps, and the deeper the engine reads, the more thoroughly the contradictions surface.
- There is no rate card for the answer itself — the fundamental difference between this channel and booths or bid rankings. In SE Ranking's public study of 50,006 US commercial prompts, 25.94% of answers carried ads, yet in 96.37% of those ad slots the advertiser never appeared in the answer body; the firm's own two-week ad test bought some 97,000 impressions for 1,263 clicks and next to no signups. A position that cannot be bought cannot be budgeted past verification — which is why "paying for GEO" can never substitute for making the facts solid.
- Brands enter AI answers mainly through two doors: entity questions and comparison questions. In Moz's public dataset — 50,000 prompts, a model-generated research sample — 97% of brand mentions came from those two types. A merchant whose entity information contradicts itself bricks up both doors at once: entity questions cannot anchor it, comparison questions dare not cite it. No volume of added content routes around a broken foundation.
- Export lead generation's three classic channels — trade fairs, marketplaces, search — all act after a buyer starts looking for suppliers; AI Q&A acts before the shortlist even forms. The buyer's path through AI breaks into six steps — discovery, shortlisting, verification, matching, comparison, due diligence — and the verification step is now performed by the AI on the buyer's behalf by default, just as the marketplace's "verified manufacturers" filter performs it on the platform side. On this channel credibility is not a bonus point; it is the ticket in. Without it, the other five steps never reach you.
- Our method starts from fact verification: name, legal entity, tenure and capabilities aligned field by field before anything is put in front of AI. With no verifiable operating entity provided, verification cannot begin — and any "optimisation" from that point could only work on the story itself, which is precisely what answer-engine verification, and our method, exist to defend against.
- This road is open to export merchants — in the right order. In an authorised case we have published, an electronic-components supplier first tightened its positioning, unified its public facts and built out its evidence list; it then began entering candidate answers to unbranded sourcing questions on some AI platforms, and the client confirmed a corresponding rise in organic traffic, a buyer making contact by email, and a closed deal. Declining this engagement is not us warning export merchants off GEO; it is us refusing the sequence of "package first, backfill facts later." We do not take projects whose results could not be honestly accepted.
What we did not verify
- We did not search company registries and make no finding about the merchant's actual registration date or trading history; every contradiction cited rests solely on the self-declared fields of its own profile page.
- We did not assess its product quality, fulfilment capability or actual export record. This page is a statement of our engagement criteria, not a finding on the merchant's commercial standing.
- The behaviour of the "verified manufacturers" filter reflects the marketplace as tested on the assessment day; platform filter and verification rules may change.
- This assessment reflects the profile as it stood in Q3 2026 and has not been rechecked.
Fix these and we are glad to look again
- A consistent public operating identity — registered name, founding year and business type that agree with the storefront's own story, with entity documents the merchant is willing to show.
- Every heavyweight claim in the introduction lands on at least one checkable anchor: a certificate number, a platform verification badge, a queryable third-party record or an accessible website of its own.
- Complete at least the minimum of the marketplace's own trust layer — obtain the verification badge so buyers using the "verified" filter can see you at all; that filtered view is the default one, on the platform as well as in AI.
- Acceptance of the order of work — facts made consistent first, visibility second — and of acceptance criteria that come with denominators and can be re-measured.
We publish this call together with its criteria to make one thing plain: the only long-term asset in the GEO business is credibility — the practitioner's, the client's and the answer ecosystem's, all tied to the same rope. When a merchant's story contradicts its own profile and its entity cannot be verified, the only "optimisation" that would move anything is helping an unverifiable story get past verification — which is not GEO but reputation-laundering for a claim, paid for ultimately by overseas buyers. The same standard binds us: whoever shows you results — ourselves included — keep asking the three questions: what is the denominator, what was the environment, can it be re-measured. If your storefront matches any one of the five findings above, make the facts agree everywhere and put your legal identity on the record first — then talk GEO. Once that is done, you are welcome back for a fresh assessment; how the work is done and how results are accepted are already public on this site, methodology and authorised cases alike.