I just saw Target put checkout inside ChatGPT—big baskets, bigger platform risk

Executive Summary

Target will launch a ChatGPT-powered shopping app in beta next week, enabling idea generation, multi-item baskets, grocery shopping, and in-chat checkout. At the same time, Target is rolling out ChatGPT Enterprise to 18,000 HQ employees and wiring OpenAI models into employee tools, supply-chain forecasting, and customer-facing assistants. For retailers, this signals a meaningful shift toward conversational commerce as an acquisition and conversion channel-promising larger baskets and faster discovery, but introducing platform dependency, governance, and margin risks.

Key Takeaways

• In-chat, multi-item checkout is the step-change: expect higher average order values and better attach rates versus single-item chat flows.
• Target gains distribution to ChatGPT’s user base, but cedes part of discovery to a third-party platform with uncertain unit economics and data access.
• Deploying ChatGPT Enterprise to 18,000 employees can compress cycle times across merchandising, forecasting, and content-but demands strict guardrails.
• Operational risks: hallucinated recommendations, price/inventory mismatches, PCI/PII exposure during checkout, and unclear liability in misorders.
• Buyers should treat this as a new channel test with rigorous measurement, not a wholesale shift from owned web/app experiences.

Breaking Down the Announcement

The beta experience embeds a full shopping journey inside ChatGPT: the assistant can propose ideas (e.g., “host a fall dinner for eight”), assemble a multi-item basket across categories, include grocery items, and complete checkout in the chat. Planned additions include loyalty (Target Circle) linkages for personalized offers and same-day delivery based on location and inventory. Separately, Target is deploying ChatGPT Enterprise to 18,000 HQ employees and integrating OpenAI models into internal tools, store workflows, and demand forecasting.

Technically, this relies on real-time APIs for product, price, and inventory; function calling/structured outputs to keep the assistant grounded; and a secure transaction handoff for payment. Expect a mix of curated prompts, retrieval over the product catalog, and policy constraints to reduce off-brand or non-compliant responses. For enterprise data, ChatGPT Enterprise typically disables training on customer content by default; the consumer ChatGPT surface requires explicit configuration to avoid training on user interactions. That distinction matters for compliance posture.

What This Changes for Retail Operators

Conversational, multi-item shopping inside a general-purpose assistant expands “top of funnel” discovery beyond search and retailer apps. The upside is higher attach and curated bundles-particularly for occasions, recipes, back-to-school, and dorm setups where shoppers want guidance. If Target can keep latency to a few seconds per step and maintain accurate grounding, a chat-led session could drive a meaningful lift in AOV and conversion versus traditional search/browse. Early benchmarks to watch: multi-item attach rate, basket size, assisted conversion, refund/return deltas from AI-driven orders, and time-to-checkout compared to site/app.

Unit economics are the catch. A rich session can consume tens of thousands of tokens; at enterprise rates that’s often low tens of cents per completed session, plus engineering and platform fees. Gains must outweigh LLM and referral costs, and the experience cannot be slower than a well-optimized native app. For checkout, ensure payment flows avoid passing PAN or CVV through the model and meet PCI DSS and, where relevant, SCA/3DS requirements. Inventory must be real-time at the store/SKU level to avoid substitutions that erode trust.

Competitive Angle

Amazon’s Rufus and Walmart’s GenAI features concentrate conversational discovery inside their own properties. Instacart’s “Ask Instacart” and Shopify’s Sidekick focus on their ecosystems. Target’s move is different: it places a transactional assistant inside a third-party AI platform. That could unlock incremental reach—especially among shoppers who start with ChatGPT for ideas—but it also hands OpenAI influence over ranking, merchandising context, and potentially fees. If OpenAI becomes a shopping aggregator, retailers face the same strategic tradeoffs they encountered with marketplaces and mobile app stores.

Constraints and Risks

• Grounding and accuracy: LLMs can hallucinate or miss constraints like size, dietary needs, or compatibility. Strict schema validation and rule-based checks are mandatory for SKUs, prices, and availability.
• Latency and abandonment: Multi-turn chat can feel slow versus tap-to-cart. Target will need aggressive caching and narrow tool calls to keep responses under 2-3 seconds.
• Governance: For consumer ChatGPT, clarify whether user interactions are used for model training. Configure “no training” where possible and disclose data flows in privacy notices.
• Liability and CX: Establish policies for misorders and substitutions initiated by the assistant, including automatic make-goods and streamlined returns.
• Cost control: Rate-limit long sessions, cap tool calls, and prioritize high-intent paths (bundles, replenishment, list-building) to keep per-order LLM costs predictable.

Why Target Is Doing This Now

With pressure on comparable sales, Target needs channels that improve discovery and margin mix without discount-heavy tactics. Search is shifting toward AI summaries that compress links and ads; meeting customers “in the chat” hedges against that traffic shift. Internally, deploying ChatGPT Enterprise to 18,000 employees aims to cut cycle times in forecasting, content, and decision support—areas where even single-digit efficiency gains produce material P&L impact at Target’s scale.

Recommendations for the Next 90 Days

• Instrument the funnel: Track assisted revenue, AOV, attach rate, session cost, time-to-checkout, overflow to web/app, and post-order defect/return rates specifically for ChatGPT-originated orders.
• Harden the rails: Enforce schema-validated outputs, price/inventory locks at add-to-cart, and secure payment handoff. Do not allow card data in free text. Run red-team tests for prompt injection and data exfiltration.
• Define channel economics: Set a clear CAC/LTV model for ChatGPT as a paid/partner channel. Cap spending until you see sustained unit economics versus your app and web.
• Align privacy and policy: Turn off training on user content where possible. Update privacy disclosures, retention policies, and DSR workflows for chat-originated orders.
• Train 18,000 employees with guardrails: Provide sanctioned prompts, retrieval over approved knowledge bases, and automatic PII redaction. Monitor usage and set quotas to manage cost and risk.

Bottom line: Target’s ChatGPT app is a credible step toward true conversational commerce. Treat it as a new, measurable performance channel with strict governance. The prize is higher-intent baskets and faster decisions; the price is platform dependency and operational complexity you must actively manage.