OVERSEAS SENTIMENT
Methodology
This dashboard tracks public sentiment about Chinese car brands overseas. Below we explain where the data comes from, how each item is assessed, how to read the figures, and which conclusions cannot be drawn.
Where the data comes from
Each period covers 30 days. Search terms are set by brand and market to collect public content: YouTube videos and comments, X, Reddit, TikTok, news (Google News) and local car forums (e.g. Pantip in Thailand, Drive2 in Russia). Facebook and Instagram have only small samples.
Map reviews of dealer outlets are counted separately, appear only on the "Dealers" page and are not included in reputation scores.
Collected content is cleaned first: empty posts, verbatim reposts, near-duplicate copies and items that do not mention a monitored brand are removed. A model then filters out content unrelated to the brand, as well as paid posting and advertising.
Reviews by channel this period
BYD (incl. Denza) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
| BYD (incl. Denza) | Count | Share |
|---|---|---|
| YouTube | 7,017 | 49% |
| News | 2,167 | 15% |
| X | 1,920 | 13% |
| TikTok | 1,584 | 11% |
| 1,189 | 8% | |
| 258 | 2% | |
| Pantip | 179 | 1% |
| 106 | 1% | |
| Telegram | 4 | 0% |
Reviews by channel this period
Chery (incl. Omoda, Jaecoo, Exeed, Jetour) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
| Chery (incl. Omoda, Jaecoo, Exeed, Jetour) | Count | Share |
|---|---|---|
| YouTube | 6,983 | 55% |
| News | 1,840 | 15% |
| TikTok | 1,710 | 14% |
| X | 1,064 | 8% |
| 766 | 6% | |
| 172 | 1% | |
| Telegram | 60 | 0% |
| Pantip | 57 | 0% |
| 11 | 0% | |
| Drom | 1 | 0% |
Reviews by channel this period
Geely Auto (incl. Galaxy) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
| Geely Auto (incl. Galaxy) | Count | Share |
|---|---|---|
| YouTube | 3,769 | 54% |
| TikTok | 1,327 | 19% |
| News | 807 | 12% |
| 456 | 7% | |
| X | 363 | 5% |
| 185 | 3% | |
| Telegram | 43 | 1% |
| 39 | 1% | |
| Drom | 6 | 0% |
Reviews by channel this period
MG (SAIC MG) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
| MG (SAIC Motor MG) | Count | Share |
|---|---|---|
| YouTube | 2,771 | 45% |
| TikTok | 1,236 | 20% |
| 910 | 15% | |
| News | 850 | 14% |
| X | 187 | 3% |
| 123 | 2% | |
| 86 | 1% |
Reviews by channel this period
GAC (incl. Aion, Hyptec) · 2026-10-16 to 2026-11-14 · Generated 2026-11-19 14:38
| GAC (incl. Aion, Hyptec) | Count | Share |
|---|---|---|
| YouTube | 2,846 | 61% |
| TikTok | 605 | 13% |
| News | 457 | 10% |
| 311 | 7% | |
| X | 263 | 6% |
| 93 | 2% | |
| 62 | 1% | |
| Pantip | 40 | 1% |
| Telegram | 16 | 0% |
Reviews by channel this period
Maxus (MAXUS / LDV overseas) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
| Maxus (MAXUS / LDV overseas) | Count | Share |
|---|---|---|
| TikTok | 270 | 43% |
| YouTube | 162 | 26% |
| News | 102 | 16% |
| 55 | 9% | |
| 37 | 6% | |
| 5 | 1% | |
| X | 1 | 0% |
Reviews by channel this period
Foton · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
| Foton | Count | Share |
|---|---|---|
| TikTok | 729 | 57% |
| YouTube | 284 | 22% |
| News | 172 | 13% |
| 39 | 3% | |
| 30 | 2% | |
| X | 16 | 1% |
| 13 | 1% | |
| Telegram | 2 | 0% |
Reviews by channel this period
Sinotruk (HOWO / SITRAK) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
| Sinotruk (HOWO / SITRAK) | Count | Share |
|---|---|---|
| TikTok | 753 | 64% |
| YouTube | 267 | 23% |
| News | 92 | 8% |
| X | 50 | 4% |
| 8 | 1% | |
| 5 | 0% |
Reviews by channel this period
Yutong · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
| Yutong | Count | Share |
|---|---|---|
| TikTok | 390 | 45% |
| YouTube | 217 | 25% |
| X | 119 | 14% |
| News | 77 | 9% |
| 58 | 7% | |
| 7 | 1% | |
| 2 | 0% |
Who said it
Each post is classified into five categories by poster: users, influencers, media, official brand accounts and dealers.
"Word of mouth" counts only users, bloggers and media. Content posted by brands and dealers is listed separately and excluded, so that brands' own promotion does not inflate ratings.
News items are always classified as media; comments are classified as netizens by default, with names that look like brands or dealerships sent to the model for judgment; authors of main posts such as YouTube channels and TikTok accounts are classified by the model based on account name, follower count and sample content.
How to assess each item
The Claude model reads each item one by one and translates non-Chinese content into Chinese. It then judges whether the item is really about the monitored brand, the overall sentiment (positive, neutral, negative or mixed), which models are mentioned and the sentiment toward each, and which topics come up, such as quality, price, range and after-sales service.
When a post mentions several models, each is assessed separately rather than applying the post's overall sentiment to every model. Comments are assessed together with the title of their parent post, so that comments like "this one is better" can be attributed to a specific model.
How markets are defined
Use the strongest available evidence: the location in the account profile, the media site's country domain, place names mentioned in the text, the market searched at collection time, or a language that maps to only one market (e.g. Thai). Comments follow the market of their parent post.
BYD (incl. Denza): 11% of user reviews this period could not be assigned to a market; these are counted in the overall and channel figures but not in market-level figures. BYD (incl. Denza) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
Chery (incl. Omoda, Jaecoo, Exeed, Jetour): 7% of word-of-mouth content this period could not be assigned to a market; it is included in overall and channel figures but not in per-market figures. Chery (incl. Omoda, Jaecoo, Exeed, Jetour) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
Geely Auto (incl. Galaxy): 9% of word-of-mouth content this period could not be assigned to a market; it is counted in overall and channel figures but not in per-market figures. Geely Auto (incl. Galaxy) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
MG (SAIC MG): 9% of word-of-mouth content this period could not be attributed to a market; it is included in the overall and channel figures but not in the market-level figures. MG (SAIC MG) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
GAC (incl. Aion and Hyptec): 6% of word-of-mouth content this period could not be assigned to a market; it is counted in overall and channel figures but not in per-market figures. GAC (incl. Aion and Hyptec) · 2026-10-16 to 2026-11-14 · Generated 2026-11-19 14:38
Maxus (MAXUS / LDV overseas): 8% of user reviews this period could not be assigned to a market; these are counted in the overall and channel figures but not in market-level figures. Maxus (MAXUS / LDV overseas) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
Foton Motor: 4% of word-of-mouth content this period could not be attributed to a market; it is included in the overall and channel figures but not in the market-level figures. Foton Motor · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
Sinotruk (HOWO / SITRAK): 5% of word-of-mouth content this period could not be attributed to a market; it is included in the overall and channel figures but not in the market-level figures. Sinotruk (HOWO / SITRAK) · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
Yutong: 15% of word-of-mouth content this period could not be attributed to a market; it is included in the overall and channel figures but not in the market-level figures. Yutong · 2026-10-17 to 2026-11-14 · Generated 2026-11-19 14:38
How to read the numbers
Net sentiment = positive share − negative share, ranging from −100% to +100%.
Each net sentiment score comes with a 95% interval. Where two intervals overlap, we do not say which is higher.
Groups with fewer than 30 entries are listed only; no conclusions are drawn.
If more than half of the negative comments on a market or model come from a single thread, this is flagged separately and treated as a one-off event, not the market's overall sentiment.
The "Engagement-weighted" page weights each post by 1 + likes, replies and reposts to show reach and influence. Read it alongside the post-count view.
Alerts and event briefs
An internal alert is triggered when any of the following apply: the negative share over the last 7 days is more than 15 percentage points above the previous level and statistically significant; negative posts on the same model and topic appear in more than 3 markets within 7 days; or a single original post gets 15 negative replies within 7 days.
Alerts are ranked by consequence: those involving safety, quality or spread across markets come first. Alerts where more than half of the negative sentiment comes from a single original post are downgraded one level. Major events get a one-page event brief, available on the "Report downloads" page.
Accuracy
Manual verification is under way; accuracy rates and the verification method will be published here once complete. Until then, the sentiment figures on this dashboard are model judgements that have not yet been validated by manual spot checks.
The agreement rate between the two models only shows that results are stable; this dashboard does not treat it as accuracy.
Limitations
Counts depend on search terms and collection scope; they do not represent total online volume and are not suitable for comparing discussion levels between brands.
Only a small amount of public content is accessible from Facebook and Instagram, so markets where discussion takes place mainly on these two platforms may be underrepresented.
Most author profiles on X have no identifiable country; these are counted in the overall total only.
Intervals are calculated item by item and do not account for correlation between comments under the same post, so actual uncertainty is somewhat greater.
Sentiment is classified by a model and may misread sarcasm, dialect and industry jargon.
Disclaimer
This dashboard is based on automated monitoring of public online content. The data and views do not represent the brands concerned and should not be cited as fact without verification. Quoted excerpts are anonymised, and the public version does not link to original posts.