01
The numbers
| 2026-08-01 | 2026-10-01 | |
|---|---|---|
| Mentioned — all 9 assistants in both runs | 7.3% | 21% |
| Mentioned — western assistants | 11% | 33.6% |
| Mentioned — Chinese assistants | 2.6% | 5.1% |
| Own site cited — western assistants | 10.5% | 33.6% |
22 buyer questions without the brand name, each asked twice per assistant. Baseline: first scheduled run; 6 of 9 assistants partly timed out.
Per assistant, and what to discount
The jump is one run: Gemini went from 2.3% to 68.2% between 09-16 and 10-01, and the three runs before read 14.1, 9.3 and 10.6. Wait for November before quoting it.
| Assistant | 2026-08-01 | 2026-10-01 |
|---|---|---|
| ChatGPT | 9.1% | 45.5% |
| Perplexity | 25% | 50% |
| Claude | 0% | 4.5% |
| Gemini | 2.6% | 68.2% |
| Grok | 18.2% | 0% |
| 豆包 Doubao | 0% | 0% |
| 通义千问 Qwen | 0% | 0% |
| DeepSeek | 0% | 0% |
| Kimi | 10.5% | 20.5% |
Rates are the plain average of per-assistant rates over assistants measured in both runs; assistants added later are left out on both sides. Citation counts western assistants only — the Chinese APIs return no source list.
02
What we did
- Wikidata entries with third-party references; IMDb credit linked through sameAs.
- A /proof page where every claim links to its evidence; press cards with bylines on /about.
- Public prices, plus a Chinese cost page for the Chinese questions.
- Profiles on Clutch and models.com; www → apex 301.
03
Still weak
Chinese assistants: only Kimi names the site (20.5%); Doubao, Qwen and DeepSeek name 0 times. Grok went from 34.1% in August to 0.