A Small Service Can Turn Menu Decisions into a Shopping Cart

Grocery shopping has entered the conversation

On August 5, 2026, Albertsons, a US grocery-store chain, announced a connection that lets people find Safeway products and build a cart inside ChatGPT. A person can say, “Plan pasta meals for a family of four,” or upload a food photo, recipe, or shopping list.

ChatGPT finds relevant products and discount information, then builds a cart. Payment does not finish inside ChatGPT, however. The person moves to Safeway to review the products again and pay.

That distinction matters. Customers can still compare products, change quantities, or remove items after receiving recommendations. The AI is not confirming a purchase; it is reducing the tedious preparation work.

At the time of the announcement, Albertsons operated 2,240 stores and 22 distribution centres. This was not a feature tested in a few neighbourhood locations. It began within a large retail network that must connect conversation to different regional products and inventory. The company said it planned to expand the feature to other banners, including Albertsons, Vons, and Jewel-Osco.

The conditions for failure are already clear. In a Gartner survey of 322 US consumers, 54% of people who had experienced AI shopping said they had to verify all of the information they received, while 62% said it had wasted their time instead. If prices, inventory, or products are wrong, a service meant to reduce grocery work adds another task to verify.

Start by reducing the gap between menu decisions and payment

Consider a dual-income parent living with two elementary-school children and a spouse with an egg allergy. They ask what everyone wants to eat in the family messenger, open the refrigerator to check what remains, make a list in a notes app, then search for products and coupons in a grocery app.

Information moves repeatedly: from the family messenger to the notes app, from the notes app to the search bar, and from search results to the cart. Conditions such as a child not being able to eat spicy food, or onions already being at home, are easy to lose along the way. The originally planned menu can also change while looking for discounted products.

A new service can change this earlier part of the process. A person enters a refrigerator photo, foods their family avoids, the number of dinners needed this week, and a budget. The service first shows menu options, then turns only the ingredients needed for the selected meals into a shopping list. It can then send people to product pages at the grocery store they use most often, or help them place items in a cart.

Some work does not change. Actual stock and final discount amounts vary by delivery address, store, membership tier, and promotional card, so a person must check them just before payment. Choices that are hard to judge from a screen, such as fish size or fruit condition, also remain.

The easiest way to fail is to require people to enter every food item in their refrigerator each time. If inventory records fall behind by only a few days, the service assumes an unavailable ingredient is there, and the person spends more time managing inventory than using recommendations. For a first service, it is better to ask only for “three ingredients to use up this time” than to promise precise refrigerator management.

Take care when mixing paid placements into discount recommendations. If people cannot tell whether an item is cheap or advertised, they will verify every recommendation again. Show a short reason such as “currently discounted,” “bought last time,” or “egg allergy excluded,” and leave the final choice to the person.

In Korea, Samsung Food already connects meal planning and purchases in a limited way. Its Korean purchase guide links 217 ready-made foods and meal kits from 12 food companies from recipes to partners’ direct online stores. That is different from checking out ordinary fresh groceries from multiple sellers in one cart. Rather than trying to combine payment across several grocery stores from the start, there is room to begin with a smaller step: organise family constraints and hand over the required products accurately.

Others connected the decision stage first

Instacart, which operates in the United States and Canada, launched a ChatGPT connection in 2023 that turned recipes into grocery lists. It later expanded the flow so people could review and pay for carts built from local-store inventory inside ChatGPT. Customers pay for products, and delivery fees, service fees, and other charges may apply depending on order conditions and membership status.

The starting point was not to build a new grocery store. It added conversational menu decisions in front of store, inventory, and delivery networks that already existed. Order volume and conversion after the connection have not been disclosed, however, so it is too early to conclude that connecting to payment alone made it successful.

France’s Carrefour launched Hopla in its own online store in 2023. The service suggested products and recipes based on budget and meal conditions. It later added a feature that reflected purchase history, and in 2026 expanded the flow so people could set a cart and delivery method in ChatGPT before paying on the Carrefour website.

It is notable that Carrefour did not hand everything to an external conversational service from the beginning. It first tested the connection between budget, diet, and products in its own store, then widened the entry point. Carrefour said it could reach 26 million ChatGPT users in France, but that is the number of people it can reach, not the number of actual purchasers.

Samsung Food launched in 2023 with more than 160,000 recipes across 104 countries. It offered weekly meal plans, shopping lists, and links to partner retailers by country. Its basic features are free. Its paid service, priced at US$6.99 per month based on the displayed US price, adds personalised meal plans and nutrition tracking.

This model does not rely only on grocery sales margins; it puts a price on planning itself. But for people to pay every week, it must do more than show many recipes. It must actually reduce how often they need to revise a shopping list.

Four things you could build here

1. Dinner planning that uses leftover ingredients first

• What it does: Takes a refrigerator photo and ingredients that need to be eaten first, then organises three days of meals and only the missing ingredients.
• Who uses it: A three-person dual-income household that cooks weekday dinners but often leaves vegetables unused in the refrigerator.
• Why now: It has become possible to read photos and short text together, then turn them into meals and shopping lists.
• First screen: Include a place to upload a refrigerator photo and a field to choose “three ingredients to use first this week.”

2. A family meal-agreement board

• What it does: Collects foods each family member cannot eat and meals they want, then shows only options everyone can eat.
• Who uses it: A household where a child with allergies and adults who like spicy food live together.
• Why now: Scattered family responses can be organised at once, and the selected meal can immediately become a shared shopping list.
• First screen: Place each person’s excluded foods, desired meals for the week, and the final voting result on one screen.

3. A local-supermarket discount shopping helper

• What it does: Takes a shopping list and supermarket promotion screens, divides what to buy where, and compares estimated spending.
• Who uses it: A household that alternates between pickup and in-store shopping at two or three large supermarkets near home rather than online delivery.
• Why now: Even when grocery carts are not connected between stores, it is possible to read product links or promotion screens and create a comparison list.
• First screen: Show estimated totals by supermarket and the time when each price was checked, using the same shopping list as the basis.

4. A grocery-checking service for parents

• What it does: Turns a parent’s spoken item requests into a list an adult child can review, and saves rules for substitute products.
• Who uses it: Adult children who place weekly orders for parents who are not comfortable with online grocery shopping.
• Why now: A request such as “if there is no small tofu, get the large one” can be organised into a list with quantity and substitution conditions.
• First screen: Present the parent’s voice request, whether substitution is allowed for each item, and the adult child’s final approval button in sequence.

Why this matters where you are

You can check whether grocery shoppers in your market move information between family chats, notes, store searches, and carts in the same way. Store inventory, promotions, delivery terms, and payment flows will differ by market, so the handoff to a retailer needs local verification. A useful first step is to test whether a service can preserve household constraints while reducing the number of times people copy information between tools.

What to check today

Ask one dual-income household that bought groceries within the last week to replay their most recent shopping trip for 30 minutes. If they move between family conversations, the refrigerator, notes, and grocery screens three or more times, and if at least one condition was missed or product was chosen again during that process, it is worth testing a service that reduces the copying steps before building a recommendation feature.