Atlas Insight is a free site. Connect your own AI assistant (Claude, ChatGPT or similar) to 13.5 million analysed posts from Vietnamese car forums and check any belief about your customers in minutes, before you build.
Explore the map yourself ↗ For desktop Chrome; it draws about 1% of the posts.
You do not learn a new chatbot. Your own assistant reads the posts.
The map draws 155,886 posts, about 1% of the total.
Why check first
Most product ideas do not work, so teams that look at customers before building waste less. Airbnb's founders went and looked at their listings. Microsoft counted how many tested ideas worked; Pendo counted the features people rarely use.
Airbnb's weekly revenue after the founders replaced hosts' phone photos with professional ones
Joe Gebbia, co-founder, in First Round Review (his own account)
tested ideas at Microsoft did what they were built to do
Kohavi and colleagues, Microsoft, 2009 and 2013
of features in the average software product are rarely or never used
Pendo feature report, 2019 (615 customer accounts, vendor data)
Now you can look too: 13.5 million analysed posts by car owners, read by your AI assistant.
What car owners told each other about insurance
For 23 years, car owners in Vietnam have swapped advice on public forums. We counted 13.5 million of their posts by topic. The worry that comes up most is not the one the 2026 product brief we tested (next section) set out to fix.
We kept only posts by car owners about insurance: 76,236 of them, with sellers and ads removed.
More than a third of them are about a claim, and about one in seven doubts the insurer will pay at all. Sales calls come up in fewer than one post in a thousand.
Insurance buys you peace of mind. Buying it is easy, claiming is hard, and only the tough and the well informed get paid.
The first two rows come from topic tags an AI model gave each post; the others count keyword matches. Base: 76,236 insurance posts written by car owners (sellers and ads removed).
Counted on 2026-10-15 by an AI assistant connected to the posts. Full answers: question 12; questions 33, 38 and 42.
A brief, tested
A Vietnamese car insurer briefed us on what it believes about car owners. We checked six of those beliefs against the posts: 4 of 6 did not hold.
| Row | The belief | What car owners wrote | Result |
|---|---|---|---|
| 1 | Car owners want a paper certificate | 17 posts in 23 years link it to police stops | Says something else |
| 2 | Car owners ignore reminder calls, fearing scams | 0 of 12 scam-word posts were about calls | Wrong |
| 3 | Car owners do not want to renew | 1,194 posts about renewing; 21 about forgetting | Wrong |
| 4 | Buying the policy means heavy paperwork | 764 posts complain about paperwork at the claim, 248 at purchase | Wrong |
| 5 | Car owners value points, bundles, roadside help | 3 roadside posts in 2022-2026; 0 of 10 value points | |
| 6 | Car owners buy insurance in the company's toll-payment app | 31 posts by car owners mention it, all years |
Six things we remember verbatim from Tasco Insurance's October 2026 briefing on Luma, not all of its beliefs, used as a baseline to check.
Wrong = the posts say the opposite. Says something else = few posts, and they say something else. Barely mentioned = too little talk to judge. Holds = the posts agree. The six, with the posts behind each →
Forum silence is not proof. Car owners who buy inside an app may never post on a forum, and much car talk after 2020 moved to social apps we did not collect. The two larger squares come from AI topic tags; the two smallest from keyword counts.
Counted on 2026-10-15 by an AI assistant connected to the posts. Full answers: questions 5 to 10 and 12.
Will the insurer actually pay?
Forums grew and shrank, and apps arrived, but the share of car owners asking whether their insurer will actually pay stayed close to 15 percent.
The mood is lopsided once a claim starts. Of 27,111 posts written while making a claim, 38 percent read as negative and 6 percent as positive. We found 116 posts that weigh price against trust. In 94 of them, a pricier insurer that pays is worth it. Only 22 say they are all alike, so buy the cheapest, and most of those mean they are all equally bad at paying.
Many drivers learn the rules of their policy only when the repair bill arrives. 2,836 posts discuss terms or exclusions, and 130 describe learning a rule only at claim time. In the most-liked refusal stories, the reasons are a flooded engine, exclusions for natural hazards, and being told to claim from the other driver.
Every insurer is the same: friendly when you pay them, different when you claim.
Only at the repair did I learn that on a 1.5 million dong scratch, I pay the 1 million deductible.
Read it as about one post in seven, every period. The forums shrank: 8,487 insurance posts by car owners in 2012, 619 in 2025, so the last bar rests on far fewer posts. Shares come from a topic tag an AI model gave each post; we did not check this one tag by hand.
Counted on 2026-10-15 by an AI assistant connected to the posts. Full answers: questions 12 and 22; question 42.
Who car owners ask
Many car owners pick an insurer by starting at the repair shop and working backwards.
738 posts describe picking the insurer the repair shop works with, and 413 credit the shop with handling the paperwork.
The car dealer matters more each decade. Insurance sold with a new car grew from 4.2 percent of car owners' insurance talk in 2003-2012 to 11.6 percent in 2022-2026.
Every repair shop has an insurer it knows. Buy that one and the claim goes more smoothly.
Mentions in insurance posts by car owners: repair shop 5,826 / car dealer 2,638 / a friend 873 / apps 275 / sales calls 50, of 76,236 insurance posts written by car owners (sellers and ads removed). Keyword counts; one post can name several. "Car dealer" includes insurance sold with a new car.
Counted on 2026-10-15 by an AI assistant connected to the posts. Full answers: questions 13 and 17; questions 33 and 40.
Ask it anything
We asked an AI assistant connected to the posts 50 questions. Seven answers on what car owners raise instead are below. The rest, including the assistant's detour into beer, road trips and clubs, are on Questions.
One answer, start to finish
This is how row four of the table above was made. The belief went in as plain words. The assistant counted posts complaining about paperwork when making a claim and when buying the policy, read a sample and linked each one.
The question, as typed
“Using the atlas, test this belief: issuing car insurance paperwork is heavy. Compare complaints at purchase and at the claim.”
Five posts it quoted
In practice
How this was made
We started from 14.5 million public posts on three Vietnamese car forums, September 2003 to October 2026. 13.5 million of them are analysed, which here means placed on a topic map, and each one links back to its page.
We kept car owners only, removing insurance agents, repair shops and car dealers advertising, news reposts and copies. That left 76,236 insurance posts written by car owners (sellers and ads removed).
Counts come from keyword searches and from topic tags an AI model gave each post. On a hand check the tags were right about 8 times in 10. A count is posts, not people, and a post shows what someone wrote, not whether it is true. Quotes are English paraphrases.
We host it. It only reads, answers about 30 questions a minute, and you send us no customer data. Data, consent and how to connect →
| Measure | Number | Since |
|---|---|---|
| Posts collected, each linked to its original | 14,531,108 | 2003-09 to 2026-10 |
| Posts analysed by topic | 13,518,238 | 2026-10 |
| Insurance posts by car owners (sellers and ads removed) | 76,236 | 2003 to 2026 |
| Topic tags correct, on posts read by hand | 8 in 10 | hand check, 2026-10 |
Ask it yourself
Every number on this page came from the posts, and your own assistant can check any of them.
Atlas Insight is a free pilot product. Anyone can use it now. Send your AI assistant to atlas.vad37.com. Let it read millions of raw posts by car owners and form its own opinion.
Free, no login. Questions: atlas@vad37.com.