Comet to test grocery quantities and cart totals
Thorben Sanktjohanser tested Perplexity Comet on a family grocery order at Knuspr.de. He asked for five types of food for four people and prioritized cheaper options, expecting the assistant to turn that request into a sensible basket.
Original prompt
Shared by the source author. See source notes for context.
You need to buy the weekly groceries for a family of four. Prioritize the cheapest options available for each item.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Thorben Sanktjohanser
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Ten trials exposed inconsistent quantities and totals; one claimed €16.89 basket actually contained €39.82 before delivery.
- Human involvement
- The author inspected the real baskets, recorded results across ten runs and challenged a subtotal that excluded extra products.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Request five grocery categories for four people with a preference for cheaper options.
- Repeat the task with clearer quantities and record the basket totals.
- Open the actual cart and compare its products and subtotal with Comet's claim.
Thorben Sanktjohanser tested Perplexity Comet on a family grocery order at Knuspr.de. He asked for five types of food for four people and prioritized cheaper options, expecting the assistant to turn that request into a sensible basket.
The initial results exposed missing quantity assumptions. Comet chose too little food and sometimes selected sliced chicken intended for sandwiches rather than the requested type of chicken. Repeating the task produced different choices, so a cart that looked plausible once was not a stable shopping result.
He expanded the instructions and ran ten trials. The recorded basket totals varied between €15.40 and €31.91. In one example, Comet reported €16.89 while the actual cart held €39.82 of goods before delivery. Extra products from checkout recommendations had entered the basket without being requested.
Sanktjohanser opened the cart himself and challenged the discrepancy. His article compares the assistant's statements with the store's contents rather than treating a generated total as a receipt. These were shopping tests; he does not report paying for ten orders. The useful result is a concrete set of failure conditions: unsuitable package sizes, product substitutions and additions near checkout can undermine a cheapest basket instruction.
Prompt
Original prompt
Source: sanktjoh.medium.com · Thorben Sanktjohanser
Frequently asked questions
Why did the first basket fall short?
Quantities were inadequate for four people, and some chicken choices were sandwich slices rather than the intended product.
What made one subtotal misleading?
Unrequested checkout recommendation items appeared in the basket, making the actual total exceed Comet's stated figure.
Were these ten paid grocery orders?
The article reports repeated basket tests, rather than ten completed purchases.
More details
About this account
Comet searched the retailer and assembled grocery carts, but quantities, substitutions and checkout additions varied across trials.
Firsthand source opened and read. Prompt is a verbatim excerpt of the published original instruction.
Credit: Thorben Sanktjohanser
Source published
Added to Agents In Use
Source checked
Tools and topics
Perplexity Comet · Knuspr.de
Grocery quantities · Cart totals · Knuspr · Checkout extras
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.