If i write a prompt to scrape websites and determine which one has online checkout in place, but the prompt returns obviously wrong output, what can i do?
To improve your prompt for detecting online checkout functionality: 1. Use the "Help me" button (metaprompter) to refine your prompt - it translates your plain English request into an optimized prompt. 2. Be more specific in your prompt - clearly define what indicates "online checkout" (shopping cart icons, payment fields, checkout buttons). 3. Iterate on your prompt by adjusting wording or structure - this is normal and part of the process. 4. Define clear column outputs for structured results (e.g., "Has Checkout: Yes/No", "Checkout Type", "Payment Methods"). 5. Preview your results before applying to ensure accuracy.
I am already very specific
Also have given 5-6 examples of website with and without online checkout. I've mapped out everything an online checkout could mean, add to cart, proceed to checkout, etc.
If your specific prompt for checkout detection returns incorrect results, try: 1. Iterate on your prompt by adjusting wording or structure - this is normal and part of the process. 2. Use the "Help me" button (metaprompter) to refine your prompt. 3. Preview your results before applying to ensure accuracy. 4. Break down your request into smaller, more focused prompts. If issues persist, request human assistance by clicking "Talk to a person."
If your detailed prompt for detecting online checkout isn't working: 1. Iterate on your prompt - this is normal and expected when working with AI. 2. Try different AI models - switch between Claygent Neon (best for data extraction and formatting), GPT-4, or Claude Opus (better for complex reasoning). 3. Use the preview feature before applying to verify results. 4. Consider modifying the generated code directly if needed. 5. Break your request into smaller, more focused prompts.
And the prompt is failing with very easy website that has online checkout, while hallucinating with websites that clearly don't have online checkout, stating "contact us"
To improve your checkout detection accuracy: 1. Try different AI models - switch between Claygent Neon (best for data extraction), GPT-4, or Claude Opus (better for complex reasoning). 2. Use the "Help me" button (metaprompter) to refine your prompt structure. 3. Iterate on your prompt - this is normal and expected when working with AI. Adjust wording or structure until you get better results. 4. Break your task into smaller, more focused prompts that handle specific aspects of checkout detection. 5. Use the preview feature before applying to verify results.
Have you included examples directly in the prompt? That usually helps guide the model more clearly. Also, make sure you’re using the best available model for the task. If the prompt still fails on obvious cases, it could be due to the complexity—like if the site has a lot of pages or dynamic content. That can consume too many tokens and limit accuracy. Feel free to share more details and we can help refine the setup.
This it the clay table where it was giving me wrong output, you can check the "Has Online Checkout" column, i gave examples and what i thought is good guidance but still https://app.clay.com/workspaces/507944/workbooks/wb_0sxi4lyzg54vtsdb4go/tables/t_0sxi4naPEN7WaFG2i9x/views/gv_0sxi4naNruecbedDKjV
Hey,
I can see your prompt is detailed, but Claygent might be struggling with the complexity. Here are a few things to try: ** Remove the examples and make it shorter - sometimes too much detail confuses the AI ** Focus on just the core task: "Visit this website and determine if it has an online checkout system where customers can buy products directly" ** Add "Take your time and be thorough" ** Try "Browse the entire website systematically before deciding" ** Use "If uncertain, default to NO" *But more than everything here: Switch to GPT-4 or Claude - they will handle this task better than GPT-4o mini as this is a very small model not made for this.
Which specific rows are giving you the worst results?
Let me know if you have more questions.
Thanks for the input! I will try it out next time, as per the rows with worst results, 75, 76, 77 were all the same company but it returned 2 NO's and 1 YES, 24 was wrong just to name a few examples
i've moved past this now but when i was going over it i was noticing many wrong ones
Glad you were able to move past it — and sorry for the inconsistent results. Let me know if it comes up again or if you need help digging into specific rows.
Hey I am trying to write prompt with the Clay's feature of generating and explaining it super simple + adding the advice you gave me. But for this table I am getting an obviously wrong result in rows 3,5,7,8,9 - I am asking the AI if that website is actually reachable and representing a business or if it gives an error, or domain for sale for example. https://app.clay.com/workspaces/507944/workbooks/wb_0sxhq21nAeh4ns9pP3z/tables/t_0sxhqc6SQPWpVdW9Svs/views/gv_0sxhqc7tUZj7449cV84
Hey, If the AI is returning incorrect results in rows 3, 5, 7, 8, and 9, it’s likely due to how the prompt is written. The output really depends on how clearly you describe what you want it to check. That said, for this type of check — a better approach would be using the Check if URL is valid integration. It’s free and more reliable for this use case. You can also add the Find Redirect Page integration to catch domains that redirect somewhere else. I’ve added those to your table so you can compare the results. Let me know if you have more questions.
hey thank you so much! I wasn't aware of this integration!
Will be checking it out and ask if any new questions pop up
Great !!! Cheers ! :)