Skip to content
Agentguiden.no
Submit

Have an agent find and test discount codes first

A ChatGPT user shares a prompt where the agent searches search engines and coupon sites for discount codes for an online store. It then opens the store's own site, adds a test item to the cart, goes to checkout and tries the codes one by one. Finally it reports which codes work, how big the discount is and which conditions apply. He found a valid discount. The prompt does not tell the agent to stop before payment.

Done with ChatGPTModerate

What was done

A ChatGPT user shares a prompt where the agent searches search engines and coupon sites for discount codes for an online store. It then opens the store's own site, adds a test item to the cart, goes to checkout and tries the codes one by one. Finally it reports which codes work, how big the discount is and which conditions apply. He found a valid discount. The prompt does not tell the agent to stop before payment.

How to do it

  1. Paste the prompt into the tool, and replace the text in brackets with your own.
  2. Read the answer through. Do not send, delete or publish anything until you have seen it yourself.
  3. Check the original before you trust a figure, a name or a date.

Try it with ChatGPT

Search search engines and coupon sites for discount codes for [online store]. Open the store's own website, add [item] to the cart and try the codes one by one at checkout. Stop before payment, and do not enter any card details. Show me a table with code, discount, conditions and whether it worked. You are done when every code has been tried and the cart is left without a purchase.

Replace text in brackets with your own.

Read the original

@cutetoxicguy on X ·

More examples

  • Set up a family home chat server by phone

    A Hacker News user has a home server running a local language model, with Open WebUI as the interface. The teenager in the house has their own account on the home network. An agent using the same model set up the server and handled the accounts, all driven from the phone through Hermes Agent. He thinks local models are fine for ordinary questions and troubleshooting. It is a comment in a discussion, not a guide.

    HermesHacker News ·

  • Build a knowledge base from YouTube videos and research papers

    Weijian Zhang runs OpenClaw on its own Mac mini, with an agent he calls Eve. He installed a tool that summarises YouTube videos and the search tool QMD himself, then asked the agent to learn both and build its own knowledge base from the videos. Later he added a tool that trims research papers down, with the output saved as Markdown. He writes that the answers were good, but the post is short and shows few examples.

    OpenClawWeijian Zhang on Substack ·

  • Rename hundreds of scanned letters so they sort by date

    A retiree with no coding experience had long ago scanned hundreds of letters and papers with file names built from initials, day, month, year and page. That gave no chronological order. She asked Codex to plan a rename to year, month and day without doing anything yet, and made a backup first. She ran the change in steps and then on more folders. Otherwise the post is mostly about a dinner app.

    CodexCountries Beginning with I ·

  • Get a restaurant menu as a link with clear headings

    A blind user writes in an AppleVis review that he asks Muse for an accessible menu every time he goes to a restaurant. The agent creates a link he can share, where each category is a heading at one level and each dish a heading at the level below. That lets him move quickly with a screen reader. The review is one person's experience, and the menu is only as accurate as the source the agent found.

    MuseAppleVis ·

Five agents a day

A short daily email with five concrete examples of AI agents. Nothing else.