When should you use it?
- A consumable is about to run out
- The exact model or size is printed on the package
- You want the cheapest identical listing
Not for: Stores that block automated shoppers; Items where a near match is not good enough (medication, parts)
What do you need, and what do you get?
You need
- Photo of the package
- Preferred stores
- Payment approval rule
You get
- Identified product and size
- Price comparison across listings
- A cart ready for your OK
How does it work, step by step?
- Snap the labelPhotograph the side with brand, size and model.
- Confirm the matchHave the agent state exactly what it thinks the item is before searching.
- Compare identical listingsAsk it to check that pack size and model match, then compare delivered price.
- Use a store that allows agentsIf a site blocks it, move to another retailer rather than forcing it.
- Approve the purchaseReview the cart and pay yourself or approve the agent's checkout.
The prompt
Our own prompt, inspired by pants2 on Hacker News and written to work in ChatGPT agent, Grok Bot and Claude. Values in lime are placeholders.
Fill in the blanks (optional; the copied prompt uses your values)
Identify the product in this photo: {photo}. State brand, model and size, then wait for my OK. Then find listings for that exact item at {stores}, confirm pack size matches, and list delivered price, seller and return terms. Build the cart at the cheapest acceptable store and stop before payment.
Inspired by pants2 on Hacker News: Hacker News comment on buying from a photo with ChatGPT Agent. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Lookalike listings differ in count or size; insist on an exact match.
- Some retailers ban agent purchases; respect their terms.
Our take
The workflow is tiny and that is its charm: a photo carries the brand, model and size more reliably than your memory. The real work is matching listings exactly, since marketplaces are full of near-duplicates with the wrong size. A photo of the wrapper carries the brand, model and size more reliably than memory, which is why pants2's habit works. The hard part is matching listings exactly; marketplaces are full of lookalikes with a different count or size. Have the agent state what it thinks the item is before searching, then confirm pack size and model match before comparing delivered prices. When Amazon started blocking agents, he moved some purchases to retailers that still allowed them, which is the right response; respect a store's terms rather than forcing it. Review the cart before paying, or approve the checkout yourself. Medication, replacement parts and anything where a near match is not good enough should not go through this.
Where the time goes back: Reordering becomes a photo and a confirmation instead of a search session.
FAQ
Why did it stop working on Amazon?
Is the cheapest listing safe?
Source and credit
Inspired by pants2 on Hacker News. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- Hacker News comment on buying from a photo with ChatGPT Agent
- Creator
- pants2 on Hacker News
- Published on
- Hacker News
- Original date
- Nov 4, 2025
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- ChatGPT agent
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends or purchases.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Oct 24, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.