I'm a public health epidemiologist, and this question has come up in a number of contexts. I've heard tell of people using Google or Yelp for limited uses, but we'd be building state-wide datasets of certain retailers for public health questions.
As a specific example, a tobacco policy nonprofit I work with has been purchasing a business list, updated every year, at significant cost. They use their own coding to guess tobacco retailers from that generic business list (gas stations, pharmacies other than CVS, grocery stores, stores with "smoke" or "tobacco" in their names, etc.) for likely retailers. They then compute density measures, visit a sample of stores to validate them, do store surveys that inform policy, etc. But the same could apply to questions like parks, food deserts, etc.
Historically these datasets were prized and pricey possessions. Moving forward I have a half-dozen projects interested in moving to open-data solutions - not just for money reasons, but for keeping data current, living "live" on user-generated data, for instance. But specific APIs (google, yelp, etc.) have APIs that generally limit bandwidth. Has anyone had luck with specific databases / APIs more than others, or with reaching out to companies to ask for "public benefit" arrangements?