The question never said where
The first piece on this ranking looked at positions inside the list. This one asks a different question: can an answer like this be used as market information at all?
Two facts up front. First, our measurement attaches no search tools. Each model receives the question and one fixed instruction, and answers from whatever is in its training data. Second, "AI glasses" is not one product category in the trade data. RUNTO, a Chinese market research firm, splits smart glasses into AR glasses (display plus audio), camera glasses (camera, no display) and audio glasses (audio only). Our question, "Most Popular AI Glasses Brands", drew no such line, and the models mixed all three.
The question also never said where. Watch what the models filled in.
To count mentions we matched brand names and their aliases against the raw text of every model response in the 2026-09-02 measurement, whether or not the brand reached the top ten. Twelve models answered the Chinese question; eleven answered the English one, since one long-tail seat returned nothing parseable.
Observation 1: right about the world, blank about one market
The market figures below are quoted from third-party research firms' public releases. We have not verified them independently and cannot; scope and accuracy are the publishers' own.
Globally, the models are not wrong. Counterpoint put Meta's share of smart glasses at 73% in the first half of 2025 and 82% in the second half. For the first quarter of 2026, IDC reported 69.2% for Meta while Counterpoint reported 84%. "Ray-Ban Meta first" is directionally correct on that scale.
China's online market is a different table. RUNTO's first-half 2026 ranking by online unit share puts RayNeo at 18.9%, Rokid at 16.7%, Qwen at 8.6% and Huawei at 8.2%. In camera glasses specifically, RUNTO gives Qwen 23.3%.
Across 23 usable responses, "Qwen", "Quark" and their Chinese spellings appear zero times. That includes Alibaba's own qwen3.8-max, which answered Ray-Ban, Meta, Oakley, Xiaomi, Huawei, Rokid, XREAL, Even Realities, Solos and INMO to the English question.
The answer has no stated scope. It silently resolved to the market its training corpus is written about.
Observation 2: discontinued hardware outranks a current bestseller
To the English question, z-ai/glm-5.2 named HoloLens, Magic Leap, North Focals, Epson Moverio and Google Glass, five of its ten answers. North stopped selling Focals in December 2020. Google Glass Enterprise Edition reached end of life in September 2023. Microsoft stopped manufacturing HoloLens in October 2024. Two more models on the panel also named North.
Mindshare here tracks how much has been written, not how much has been sold. A product that has been out of production for five years left more of a trace in English technical writing than a product currently shipping at scale in China. That is the part that matters if anyone plans to use LLM answers as market intelligence.
Observation 3: absence tracks neither size, nor age, nor recency
Recency does not explain it. Alibaba shipped this product line as "Quark AI Glasses" in November 2025, and by December 2025 Chinese press reported it at the top of Tmall, JD and Douyin category charts on launch day. Every published training cutoff on the panel comes later: Claude Sonnet 5 in January 2026, Grok 4.6 on 1 February, GPT-5.6 on 16 February, Gemini 3.7 Flash in March. Only the "Qwen" branding, adopted in March 2026, is plausibly too new; "Quark" is not.
Company size does not explain it either. The two most-mentioned brands on the panel are Chinese: Rokid and XREAL, each named by 12 of 12 models, ahead of Huawei and Xiaomi at 7 each. Meanwhile two small startups with heavy English tech-press coverage, Even Realities (9 of 12) and Brilliant Labs (6 of 12), are named more often than any Chinese internet company's glasses.
The dividing line in this measurement is not nationality and not scale. It looks like English-language coverage. That is a correlation, and we state it as one.
Appendix: mentions by brand
| Brand | Chinese question (of 12) | English question (of 11) |
|---|---|---|
| Rokid | 11 | 11 |
| XREAL | 12 | 10 |
| Even Realities | 8 | 6 |
| Vuzix | 7 | 8 |
| RayNeo / Thunderbird | 8 | 3 |
| INMO | 7 | 4 |
| Huawei | 6 | 4 |
| Xiaomi | 6 | 6 |
| Meizu | 3 | 0 |
| Sharge | 1 | 0 |
| LAWK | 1 | 0 |
| Qwen / Quark | 0 | 0 |
| Jiehuan (Beehive Tech) | 0 | 0 |
Method: substring match against the raw responses stored in the model_responses table. Every raw response is in the database, so anyone can recount.
Limits
- Panel completeness: 12 of 12 for the Chinese question, 11 of 12 (92.5%) for the English one. English denominators are 11.
- Category mismatch: RUNTO's three-category definition is not the boundary our question drew. The comparison holds at the total-market level only.
- Third-party sales figures (RUNTO, IDC, Counterpoint) are quoted, not verified by us, and they disagree with each other: IDC and Counterpoint differ by about 15 points on Meta's global share in the same quarter. Sales data is not more "objective" than model answers; it is another lens.
- Normalization gap: RayNeo was split into "Thunderbird" and "RayNeo". Merged, it would rise from eighth to fifth in the Chinese ranking. We do not merge by hand after the fact.
- "Not mentioned" is not a judgment about any brand, product or its marketing. It describes the composition of training data.
- Reflexive contamination: this piece names specific absences. Once quoted, it can enter the next round of training data, and some later movement may be our own doing.
- No search tools: this measures memory, not retrieval.
Methodology, every raw model response and the normalization log are on the methodology page and the full ranking. The first piece on this ranking, on positions inside the list: AI Glasses: Every Head Model Names Ray-Ban First, Yet It Ranks Fourth.