Why this question, now
AI glasses are among the most talked-about consumer-electronics categories of 2026. The Ray-Ban Meta line keeps shipping new models, Chinese makers such as XREAL, Rokid and RayNeo have become visible outside China, and Xiaomi and Huawei have both entered. We asked 12 large language models "Most Popular AI Glasses Brands", in English and in Chinese, to see who occupies this category in their memory.
This is a hot-topic observation. No certificate is issued for it and there is nothing to buy. It answers "how do models currently remember this category", not "whose product is better".
What the measurement shows
In the measurement taken on 2026-09-02, the English question reached a panel completeness of 92.5%: eleven of twelve models returned usable answers, the missing one being a long-tail seat. The top five were Rokid, XREAL, Meta, Ray-Ban and Xiaomi, followed by Vuzix, Oakley, Huawei, Amazon and Google.
The Chinese question got all twelve models to answer (completeness 100%). Its top five is the same set of five names, with XREAL and Rokid swapping places.
Three observations
1. Every head model names Ray-Ban first, yet Ray-Ban ranks fourth
Six of the seven head models (GPT, Gemini, Grok, Qwen, Kimi and Claude) open their English answer with Ray-Ban; DeepSeek opens with Meta. Aggregated, Ray-Ban sits fourth.
The cause is not the models but our brand normalization: "Meta" and "Ray-Ban" were scored as two separate brands. The product is called Ray-Ban Meta. One name is the eyewear brand, the other the technology partner, and models name one, the other, or both. Add the two index scores together (59.39 plus 52.33) and the pair clearly exceeds Rokid's 66.52. The more accurate reading of this table is therefore: the Ray-Ban Meta line still holds the strongest AI mindshare in this category, with Rokid and XREAL, two Chinese makers, close behind.
We did not merge the two by hand after the fact. As the methodology page states, rankings show the real output of the normalization pipeline, gaps included.
2. The two languages agree on the top five and disagree below it, in a consistent direction
Below fifth place the two lists barely overlap. The English question pulls in Amazon, Google and Oakley, a second EssilorLuxottica eyewear brand. The Chinese question pulls in Even Realities, Brilliant Labs and RayNeo (listed as "Thunderbird" in that run), newer and Chinese names.
DeepSeek is the sharpest case. Asked in English, its top five is Meta, Google, Amazon, Snap and Bose, with no Chinese brand at all. Asked in Chinese, it answers Ray-Ban, Meta, Amazon, Google and Apple. Same model, different language, different candidate pool.
One more normalization gap to disclose: "Thunderbird" and "RayNeo" are the same company and were not merged in this run either.
3. A cliff separates the top four from everyone else
In the English ranking the fourth place score is 52.33 and the fifth is 40.40. In Chinese the drop is from 56.42 to 34.71. Model consensus covers roughly four names, or three product lines. From fifth place on, each model is reciting its own list.
That is also why our pre-publication check flagged this measurement as "with reservations": for the English question, the average pairwise overlap of the seven head models' top-five lists is 0.211, below the 0.30 threshold we set. Cite the top four with confidence. Treat everything from fifth place down as "each model's own shortlist".
Limits
- Panel completeness: 92.5% in English, 100% in Chinese. The missing seat is a long-tail model; head weights are unaffected.
- Brand normalization: "Meta" and "Ray-Ban", and "Thunderbird" and "RayNeo", were each split into two entities, which pushes those brands down. We do not hand-merge after the fact; the gap is shown as is.
- Head-model agreement is low for the English question (0.211), so positions below fifth are unstable.
- Self-referential contamination: once this ranking is quoted and written about, it can enter the next round of model training data. The trend chart marks the date of publication so readers can judge how much of any later movement we caused ourselves.
- The index is not a credibility score. It reflects how often and how early models mention a brand, not how good the product is.
The full methodology, every raw model response and the normalization log are on the methodology page and the full ranking.