AI Reads Google Reviews to Reveal Hidden Grocery Store Problems

A new study shows AI can turn ordinary Google reviews into a detailed picture of how hard — or easy — it is to buy groceries in a city. Researchers Teresa Groton and Benjamin Rachunok gathered 25,125 reviews of 49 Raleigh, North Carolina grocery stores, then used two AI techniques: one that automatically spots recurring topics in the text, and another that sorts each topic into five standard measures of food access — availability, accessibility, affordability, accommodation and acceptability. The AI matched human coders 85.4% of the time.

Why does that matter? Cities have traditionally judged "food access" by distance: how many people live within a mile of a supermarket. But a store can be close and still out of reach — if it's too expensive, if aisles are hard to navigate with a wheelchair or stroller, if staff are unwelcoming, or if shelves don't carry the food a neighborhood actually cooks.

The reviews exposed exactly those gaps. Strikingly, stores in the same chain, just a short drive apart, were described very differently — even though company policies were identical. The authors say food access is really about fit: how well a store matches the people living around it. Perceived access also followed income and demographic patterns that roughly track, but don't perfectly copy, the patterns seen with geographic measures.

The honest catch: Google reviewers aren't a random sample. People post after a great or terrible trip, so ordinary experiences get lost, and this covers only one mid-sized American city. Still, it offers a cheap, fast companion to costly surveys and store visits — a way for planners and shoppers to see barriers a map alone will never show.