Social listening is not a substitute for a survey

Every cycle, a vendor promises to replace official sentiment with a stream of posts, reviews, and search queries. The demo is always impressive: a spike in anger at 2 p.m. after a fare hike. The scientific question is whether that spike maps onto households that spend.

Different populations

Social text over-represents the young, the angry, the professionalized poster, and anyone whose job is to have a take. Surveys over-represent the reachable and the patient. Search queries sit in between: they capture intent (“best cheap sofa”) with less performance, but they are polluted by media cycles and SEO. Card spend captures the basket and almost nothing of the feeling. There is no true census of mood. There are instruments with different biases.

What we take from public posts

CrowdMood does not treat a transformer scored on tweets as a Conference Board print. We use public language as a nowcast feature for sectors where people complain in public — airlines, food delivery, mobile networks — and we keep it out of sectors where they do not. We also refuse to sell “emotion percentages” without a documented lexicon or model card. If you cannot reproduce last Tuesday, you do not have data.

Search and review text can lead surveys when a price shock is local and fast: a grocery staple, a fuel spike, a cancelled festival. They lag when the story is slow and private: a quiet downshift in restaurant frequency among 55-year-olds. The desk’s job is to know which kind of week it is.

Ethics and spam

Scraping without a policy is how alternative data businesses get uninvited from the internet. We do not ingest private messages, do not deanonymize posters, and do not sell lists of people. Licensed social features are aggregates, delayed where required, and dropped when a platform’s terms change. A signal that depends on a terms-of-service violation has a short half-life.

Readers who only want the public tape can ignore this layer. It exists so that paying operators can see a faster residual — the part of mood not yet visible in monthly surveys — without confusing it for a representative poll.