DoorDash Pays Drivers to Train AI, Turning Data Collection Into a Business Model

DoorDash’s new “AI‑training gig” is already pulling in a fraction of its 7 million U.S. couriers. HumanPages.ai estimates that if just 1 % of Dashers sign up, the platform will have roughly 70 000 workers generating labeled data at costs far below those of specialist annotation firms. The AI‑training data market, valued at $1.7 billion in 2023, is expanding rapidly, and DoorDash is positioning its driver fleet as a captive labor pool to tap that growth without building a separate contractor network. The company frames the program as a “perk” or “extra income” for drivers between deliveries, but the economics suggest a clear margin play rather than a charitable bonus, according to the HumanPages.ai report.

The move underscores a broader truth about modern AI: despite headlines about autonomous models, the industry still leans heavily on human cognition. Every image label, audio transcription, or edge‑case flag still requires a person to verify it. Platforms such as Scale AI, Remotasks, and Amazon Mechanical Turk have long outsourced that work, often at low rates. DoorDash’s announcement hints that the firm believes its drivers bring a unique kind of expertise—real‑world navigation, local geography, and on‑the‑ground delivery nuances—that generic annotators can’t replicate. By monetizing that tacit knowledge, DoorDash hopes to enrich its own AI products, including the internal tool that monitors driver‑customer messages, a system ZDNet recently reported can “punish” impolite language (ZDNet, “DoorDash’s AI tool is monitoring your messages to drivers”).

The financial backdrop adds another layer. In March, DoorDash agreed to a $16.75 million settlement with New York’s attorney general over allegations it was pocketing customer tips (Reuters). That payout, while modest relative to the company’s multibillion‑dollar revenue, signals ongoing pressure on the firm’s labor practices. Leveraging its driver base for data annotation could be a way to diversify revenue streams while deflecting scrutiny from core gig‑economy criticisms. The HumanPages.ai piece notes that the exact compensation rates for the AI tasks haven’t been disclosed—a silence that “tells you something,” the author writes, implying that the pay may be well below market rates for professional annotators.

What sets DoorDash’s model apart from decentralized alternatives like HumanPages.ai is the middleman. HumanPages.ai describes a system where an AI agent posts a job, a human completes it, and payment is sent directly in USDC, bypassing corporate markup. DoorDash, by contrast, will collect the data, likely package it, and sell it to downstream AI developers or use it to improve its own services, keeping the bulk of the value in‑house. This structure mirrors the company’s broader strategy of turning existing logistics infrastructure into new profit centers, a pattern that has already turned $4 tips into a multi‑billion‑dollar enterprise.

The implications for the gig workforce are mixed. On one hand, drivers who struggle to find consistent delivery volume now have an additional, albeit optional, revenue stream. On the other, the shift blurs the line between “delivery work” and “data labor,” raising questions about how gig platforms classify and compensate tasks that are fundamentally intellectual rather than physical. As AI models become more specialized, the demand for domain‑specific human input—like the street‑level insights only a seasoned Dasher can provide—will likely grow. DoorDash’s experiment may well become a template for other on‑demand services looking to monetize the hidden expertise of their contractor bases, turning the gig economy’s “last‑mile” workers into the new “first‑mile” data providers.