New AI Reads Your Angry Complaint — and Flags When Companies Ignore You

When you file a complaint with a bank, an airline, or a retailer, somebody — or increasingly some software — decides how serious it is. Most of that software sorts complaints into simple buckets: negative or not negative. A new research paper from Peiheng Gao and two co-authors argues that's too crude. Their system treats how upset you sound as a sliding scale, like a temperature reading rather than an on/off switch.

Here's how it works. An AI reads each complaint and gives it a negativity score. The researchers then plot all those scores as a curve, which lets them compare complaints that got money back with complaints that were rejected. Finally, they link the scores to the actual dollar amounts and company responses. If a complaint reads as extremely angry but the company paid out almost nothing, the system raises a flag. Think of it as an audit tool that spots mismatches between how badly someone was wronged and how little they were offered.

The honest catch: anger alone is a poor predictor. The researchers found that complaints which won money and complaints that lost look almost identical in tone — the curves overlap heavily. So this is not a lie detector for your frustration, and it can't tell you whether you deserve a refund. It only becomes useful when combined with the payout data, which companies already hold privately.

So what does this mean for you? If regulators and consumer agencies adopt tools like this, they could automatically spot patterns — say, a company that consistently offers tiny settlements to its most furious customers. That kind of monitoring is slow and manual today. Over time, it could nudge companies toward responding more fairly on the first try, because someone is finally measuring whether the response matched the harm. It won't fix your individual complaint tomorrow, but it makes systemic neglect easier to catch.