Chatbots at the Drive-Thru: What Has Happened and What Remains Unresolved

In 2021, McDonald's installed voice-based ordering systems in dozens of test restaurants across the United States. Three years later, in 2024, it shut down that pilot with minimal fanfare. Wendy's, meanwhile, launched its own system—called Wendy's FreshAI, developed with Google—and continues to expand it. The contrast between these two trajectories captures where conversational AI in fast food actually stands: nowhere near the frictionless rollout that initial demos promised, but also nowhere near the complete failure some claimed when McDonald's pulled back.

This week, The Verge covered in its newsletter The Stepback an analysis of how this technology has evolved from those first pilots to today, and why the drive-thru remains a relevant testing ground for understanding the practical limits of conversational AI systems in high-pressure environments.

Why the Drive-Thru Is Especially Challenging

Taking an order at a window doesn't sound complicated, but it concentrates a notable amount of hostile variables for any speech recognition system: engine noise, wind, regional accents, children in the backseat, customers changing their minds mid-order, promotions that shift weekly and menus that vary by franchise. Current language models handle semantic ambiguity well, but the speech systems feeding them remain sensitive to acoustic context.

McDonald's pilot stumbled right there: transcription errors generated incorrect orders frequently enough that the operational cost of manual corrections wiped out much of the labour savings. It's not that AI doesn't work; it's that it doesn't work well enough under the real-world conditions of a high-volume restaurant.

Wendy's has taken a different approach: more gradual rollout, fine-tuning by location and keeping a human available to step in when the system isn't confident. It's a more conservative approach, but also more honest about what the technology can do today.

Who Benefits and Under What Conditions

For chains with resources to invest in integration and ongoing maintenance, AI systems at the drive-thru make sense in specific scenarios: restaurants with relatively stable menus, predictable order volumes and markets where labour costs are high. It's not a universal solution.

For workers, the pitch has always been that AI would handle order-taking so they could focus on food prep. In practice, what the more mature pilots show is a gradual reduction in shift staffing, not a reassignment of tasks. This isn't new to fast food: it's the standard automation pattern in hospitality when technology reaches sufficient reliability.

For customers, the experience varies widely. Some user studies from operators suggest that some customers prefer interacting with an automated system because they feel less pressure to decide quickly. Others, especially with complex orders or modifications, report frustration when the system doesn't confirm correctly what they ordered.

The Drive-Thru as Laboratory, Not Destination

What's interesting about the fast food case isn't whether voice chatbots will replace window staff—that depends on economic and regulatory variables beyond model capability—but what these deployments reveal about the real requirements of conversational AI in production.

Lab demos always work well. The challenge is noise, unpredictability and the cost of errors when a customer is waiting and five cars are behind them. That environment has forced engineering teams to be far more rigorous about latency, acceptable error rates and escalation flow design to humans. Lessons that, by the way, apply directly to any agent or conversational assistant system in high-volume contexts.

That said, we should be careful not to over-extrapolate. Wendy's expanding its system in 2026 doesn't mean the problem is solved; it means they've found an operational balance point in their specific conditions. Other chains, with different menus, different markets and different margins, may reach different conclusions. AI in the drive-thru is a maturing use case, not an off-the-shelf plug-and-play solution.

---

From ClaudeWave, we value analyses like The Verge's that focus on the actual trajectory of these pilots rather than launch promises. The drive-thru has been the most cited conversational AI use case in corporate presentations for five years now; it was time for a reality check with real data.