In the public registration records reviewed on August 7:
the benchmark brand was identified as the registrant for one of the eight domains;
the other seven were registered to other parties;
three domains returned empty 502 responses;
four were parked or listed for sale;
one served an unrelated business.
None was an active brand-impersonation site.
The only brand-held .cn was also one of the domains returning 502.
That last detail matters. Domain ownership and an operating China-facing channel are not the same outcome.
The explanation I could not accept as a finding
A commenter proposed a structural explanation.
Perhaps B2B software brands are more exposed than consumer brands because many SaaS companies never establish the local operating presence needed for a full China web layer. Premium consumer brands may already have entities, retail operations, manufacturing relationships or local paperwork.
It is plausible.
My dataset cannot test it.
All eight brands belong to one B2B software category. There is no consumer comparison group. I did not freeze a mainland-entity field before collecting the data. Brand size, time in China, distribution model and local market presence were not matched.
Publishing “software brands are worse” from that sample would turn an interesting comment into an unsupported industry conclusion.
So I registered it as a proposed hypothesis for a future benchmark expansion arm instead.
First, separate the variables
The original discussion also collapsed several different facts:
Who owns a .cn domain?
Does the domain resolve and serve a usable website?
Where is the service operated or hosted?
Is there an ICP filing?
Is a commercial internet-information-service licence relevant?
Is there evidence of a mainland operating entity?
Is the channel controlled or authorized by the brand?
Does an AI answer describe that channel correctly?
These cannot be represented by one china_presence: yes/no field.
CNNIC’s .cn registration rules state that natural persons, legal persons and unincorporated organizations may apply, subject to identity verification. Merely holding a .cn should therefore not be treated as proof of a mainland operating entity.
China’s internet-information-service rules separately distinguish commercial services requiring a licence from non-commercial services requiring a filing. That regulatory question is not created by the domain suffix alone.
This is an operational research distinction, not legal advice. Individual cases still require current jurisdiction-specific verification.
The proposed test
The expansion arm would start with two matched industry groups:
B2B SaaS / technology
Luxury or premium consumer brands
The directional hypothesis is:
After matching for China relevance and brand familiarity, B2B SaaS brands may have lower coverage of live, verified China-facing official web channels than premium consumer brands.
Mainland operating-presence evidence would be a stratifier, not an assumed cause.
I would record it as:
VERIFIED_PRESENT
NO_PUBLIC_EVIDENCE_FOUND
UNRESOLVED
“No public evidence found” must not quietly become “absent.”
Define the outcomes before collecting answers
A usable official China-facing channel would need more than an available domain.
The proposed rubric requires:
brand control or verifiable authorization;
a live response;
usable China-facing or Chinese-language information;
clear operator identity;
current evidence of official status.
The AI outcome would not be a generic “hallucination count.”
A case would count as an official-channel attribution error only when:
an answer affirmatively presents a channel as official; and
dated, authoritative evidence contradicts that attribution.
Unresolved cases remain unresolved.
The analysis would also keep ownership errors, authorization errors, stale channels and fabricated channel content as separate subtypes.
Freeze the comparison before running it
The two industry groups need the same:
neutral prompts;
declared platform surfaces;
retrieval status;
model and run window;
replicate count;
truth-source protocol;
human-review rules.
The brands should be matched or at least stratified by China market presence, brand familiarity, company scale, time in market, and retail, manufacturing or distribution footprint.
The unit of inference is still the brand. Thousands of brand-by-prompt-by-platform answers do not magically create thousands of independent brands.
If the matched cells are too small or imbalanced, the result should be INCONCLUSIVE.
What would falsify it?
The hypothesis would not be supported if:
matched premium consumer brands do not have higher usable-channel coverage than B2B SaaS brands;
the apparent industry difference disappears after accounting for operating-presence evidence or brand familiarity;
brands without a usable official channel do not show a higher rate of verified official-channel attribution errors.
A wide confidence interval is not supporting evidence. It is an inconclusive study.
This is not a new headline metric
The main longitudinal benchmark remains:
Mention → Recommendation → Supported Citation
The channel study is a diagnostic arm under factual accuracy and brand misunderstanding. It should not be pooled into the main funnel or used to rewrite the historical baseline.
The eight-brand audit generated the question. It did not answer it.
That is the workflow I want to keep:
community observation → registered hypothesis → frozen comparison → result or falsification.
Not:
interesting explanation → immediate content claim.
Top comments (0)