AI Has Entered the Drive-Thru.
Here's What Your Business
Should Learn From It.

By Nick Maurer  ·   ·  8 min read  ·  Voice AI & Automation

Wendy's, McDonald's and Taco Bell have spent three years and millions of orders testing AI at the speaker box. Some of it worked. Some of it went viral for the wrong reasons. Both halves hold lessons for any business that answers a phone.

The next time you pull up to a drive-thru and a friendly, slightly too-consistent voice asks if you'd like to make that a combo, there's a real chance you're talking to software. America's largest fast-food chains have been putting AI in charge of the ordering speaker since 2023, and in 2026 the effort is picking up speed.

It's easy to treat this as a fast-food curiosity. It isn't. The drive-thru is one of the hardest customer conversations to automate: background noise, bad microphones, kids shouting from the back seat, regional accents, and a menu with thousands of possible customizations, all against the clock. If voice AI can work there, it can work on your phone line. And where it has failed there, it will fail for you in exactly the same ways.

Where the Big Chains Actually Stand

Headlines tend to flatten this into "AI is taking fast-food orders now." The real picture is more useful: three chains, three very different outcomes.

Wendy's: FreshAI

Scaling

Built with Google Cloud and launched at a single Columbus, Ohio location in June 2023. Wendy's spent months at that one site before expanding, and publicly committed to resisting the urge to deploy too quickly. By early 2025 it was in 100+ restaurants, with a target of 500–600 by year end. Leadership reported a higher average check and an 80-basis-point margin improvement at company-operated locations, driven largely by the AI never forgetting to suggest an add-on. Wendy's positions it as an assistant to crew, not a replacement.

McDonald's: ArchIQ ("Archy")

Second Attempt

McDonald's first try, an IBM-built system tested at 100+ U.S. restaurants from 2021, was shut down in July 2024 after viral videos of wrong orders, including one that reportedly piled on more than $250 of chicken nuggets. In June 2026 the company unveiled ArchIQ, a Google-powered replacement now being tested at five U.S. locations. The CEO has said the new system is about 90% accurate, and a franchisee account reported roughly nine in ten orders completing without staff stepping in. It also watches the restaurant itself, flagging problems like a failing freezer.

Taco Bell: Voice AI

Recalibrating

Rolled out to more than 500 U.S. drive-thrus starting in 2023 and handled over two million orders. Then a customer ordered 18,000 cups of water to force the system to hand him to a human, and the clip went everywhere. In late 2025, Taco Bell's technology chief admitted the system sometimes lets him down, and the company shifted to keeping humans "in the loop," especially at peak hours, with usage decided restaurant by restaurant.

2M+
drive-thru orders handled by Taco Bell's voice AI across 500+ locations
// WSJ via Benzinga, 2025
80 bps
margin improvement Wendy's attributed to FreshAI at company-operated restaurants
// Restaurant Business, 2025
~90%
accuracy McDonald's CEO cited for its new Google-built drive-thru AI
// Newsweek, June 2026
55%
of Americans polled said they'd still rather order from a human
// YouGov via Mashed, 2025

Hold both of those last numbers in your head at once. The technology is getting genuinely good, and more than half of customers would still prefer a person. The businesses that win with voice AI are the ones that respect both facts.

Why They Keep Coming Back to It

McDonald's pulled the plug once and is trying again. Taco Bell took a public embarrassment and adjusted rather than quit. That persistence tells you the economics are compelling. Three reasons stand out:

  • Perfect consistency on the upsell. A tired crew member at 11pm skips the add-on suggestion. The AI never does. That single behavior is where most of Wendy's reported check-size lift came from.
  • Staff freed for the work only people can do. Franchisees report their crews actually asked for it. Trying to hear orders through a noisy headset while making change is miserable. Take that away and the person at the window can focus on speed, accuracy and hospitality.
  • The conversation becomes data. Every AI-handled order is structured, logged and measurable. Taco Bell could study two million orders to find exactly where the tool helps and where it hurts. A human-only speaker box produces none of that.

Where It Went Wrong, and Why It Matters to You

None of the viral failures happened because the AI couldn't understand English. They were design failures, the kind any business can make when it automates a customer conversation without thinking through what happens when things go sideways.

No sanity checks. A system that will accept 18,000 waters, or keep stacking nuggets, has no concept of a reasonable order. The language model was never the right thing to decide what's plausible. A simple business rule would have been.

No easy exit to a person. The water-cup customer wasn't breaking the system for fun. He was trying to reach a human, and the absurd order was the only way he could find. If the escape hatch is hidden, customers will build their own.

Annoying by design. Customers complained about being cut off mid-sentence after a half-second pause, and about being asked again and again to add a drink. Upselling that never takes no for an answer stops being revenue and starts being friction.

Scaled before it was ready. The contrast is instructive: Wendy's deliberately stayed at one location for months before expanding, while the rollouts that stumbled publicly went wide first and learned in front of customers.

The lesson from the drive-thru isn't that voice AI doesn't work. It's that AI talking to your customers needs the same engineering discipline as anything else that touches your revenue.

Your Business Has a Drive-Thru, Too

You may not sell burgers, but you almost certainly have a moment where a customer arrives, says what they want, and waits to be served. For most small businesses, that moment is the phone. It's also where a surprising amount of revenue quietly disappears: calls that ring out after hours, voicemails nobody returns until tomorrow, and callers who simply try the next company on the list.

// At the drive-thru // In your business
The speaker box
Your phone line, web chat and contact form, the first point of contact
The menu
Your services, pricing rules and service area, what the AI is allowed to offer
"Would you like to make that a combo?"
The add-on your team forgets to mention: the maintenance plan, the second service, the upgrade
"Please pull ahead"
The booked appointment, landed straight on your calendar and in your CRM
Crew stepping in at the window
The warm handoff to you or your staff for anything unusual, sensitive or high-value
18,000 cups of water
The edge case you didn't plan for, and the reason testing comes before launch

The same kind of technology these chains are paying to develop is now within reach of a 10-person business: AI that answers calls around the clock, handles common questions, qualifies the inquiry, books the appointment and passes anything complicated to a person, with a written record of every conversation.

Five Rules Borrowed From the Speaker Box

01

Let rules, not the AI, decide what's allowed

The AI handles the conversation. Plain business logic handles the limits: valid services, realistic quantities, your actual service area, open time slots. That's what would have stopped 18,000 waters.

02

Make the human handoff obvious

"Can I speak to someone?" should work the first time, every time. Customers who want a person should never have to trick the system into giving them one.

03

Offer once, then take no for an answer

One well-timed add-on suggestion lifts revenue. Three in a row costs you the customer. Consistency is the advantage; persistence is the liability.

04

Say it's AI, and follow the recording rules

Tell callers they're speaking with an AI assistant. It builds trust and sets expectations. It also matters legally: Florida and several other states require every party's consent before a call is recorded, so any voice AI that records or transcribes calls needs a clear disclosure at the start. Check the rules for every state you serve.

05

Test it like someone's trying to break it

Before a real customer hears it, run it through the mumbled request, the wrong-service inquiry, the caller who changes their mind three times, and the one who's deliberately trolling. Wendy's spent months at one site for a reason.

How many calls did your business miss last month: after hours, during a job, while everyone was busy? And how many of those callers tried a competitor next?

If you don't know the number, that's the first thing to find out. It's usually larger than owners expect, and it's one of the most direct places AI can pay for itself.

At VibeLab Group, this is how we build. The "Lab" is where automations get stress-tested against the strange, the rude and the unexpected before they ever speak to your customers. The chains learned these lessons in public, at scale. You don't have to.

Find Out Where AI Fits in Your Customer Conversations

A free consultation maps how inquiries reach you today, where they slip through, and what a tested, human-backed AI setup would look like for your business.

Get Started