DoorDash launches Ask DoorDash AI assistant

The most striking thing about Ask DoorDash is not the chatbot itself but what early usage data shows. Nearly half of takeout orders placed through the tool came from restaurants the customer had never used before, and co-founder Andy Fang says it drives more new-restaurant trial than anything else the company has tried in the app ^[2]. On the grocery side, customers build carts roughly five times faster and produce baskets about 35% larger than the standard flow ^[2]. Those are not marginal funnel improvements; they point to the interface changing what people order, not just how fast they order it. For a marketplace whose economics hinge on discovery and basket size, a feature that simultaneously expands the long tail of restaurants and inflates grocery subtotals is close to ideal.

Ask DoorDash works by absorbing messy, real-world inputs and resolving them into a structured order. A user can type 'filling dinner for a family of 4,' and the app surfaces restaurants with a personalized blurb, then refines on dietary preference, budget, group size, and past orders in real time ^[1]. For groceries, the unlock is multimodal input: upload a cookbook photo, snap a handwritten list, or paste a recipe link, and the app builds a cart with correct ingredients and quantities, even prompting about pantry staples like sugar and butter so you don't double-buy ^[4][5]. The technical backbone is a blend of frontier models from OpenAI, Anthropic, and Google, with open-source models layered in to control cost ^[2]. That cost discipline matters at DoorDash's scale: running large models against every grocery photo is expensive, and the open-source supplement likely helps keep per-order inference affordable as the feature expands across the U.S.

Behind the consumer polish is an existential bet. CEO Tony Xu's argument is that even if external AI agents start handling the first steps of food delivery, DoorDash can orchestrate the entire flow end-to-end and keep customers from drifting to third-party assistants ^[3]. Analysts have been pressing exactly this question: MoffettNathanson's Michael Morton has flagged the risk that AI-driven intermediaries disrupt DoorDash's direct customer relationship ^[3]. Shipping its own agent is how DoorDash tries to stay the destination rather than become a backend that some other AI calls. The ambition extends further: Fang has signaled DoorDash wants to license the same agentic technology so grocers, restaurants, and retailers can power their own experiences ^[2]— turning a defensive consumer feature into a potential platform business.

Ask DoorDash is also where DoorDash's recent dealmaking surfaces as a product. The forthcoming reservations capability — describe 'a table for two downtown for a date-night dinner around 8 PM' and see restaurants with availability — runs on SevenRooms, the reservations and CRM platform DoorDash bought for $1.2 billion ^[6]. The roughly $4 billion Deliveroo acquisition extends the global footprint and ties Deliveroo's consumer platform to that same reservation tech ^[7]. Stitched together, restaurant search, grocery, and reservations become a single conversational surface rather than three separate products. That unification is the payoff DoorDash needs to justify spending north of five billion dollars on M&A, and it explains why the company plans to put several hundred million dollars into new products and technology this year ^[3].

Not everyone is convinced the behavior change sticks. An eMarketer analyst notes that food delivery is often impulse-driven — users beeline to a known restaurant and place a familiar order — and 'a prompt-based interface may not save time for someone who already has a craving for a pizza with mushrooms and green peppers from their go-to spot' ^[3]. The skeptic's point is real: the impressive discovery and basket metrics may reflect early-adopter exploration rather than the median habitual orderer, for whom typing a sentence is slower than tapping a saved favorite. Notably, the launch's own framing concedes this — DoorDash positions Ask DoorDash for the moments when you don't already know what you want, not as a replacement for the one-tap reorder. Whether the tool converts habitual orderers or settles in as a discovery layer for the undecided is the open question that the launch metrics, drawn from early users, can't yet answer.