Inconsistent Results with Claygent Research on Large Datasets

Olaitan F. · 2026-01-12T12:12:44.270Z

Hi Support, I'm getting inconsistent and poor results with Claygent research. I've done a lot of iterations but sometimes it finds some information and sometimes it doesn't. I've spent thousands of credits testing with even the most expensive models. I usually have to run it more than once to get the desired result. It is quite frustrating because I want to run it on 70k rows

Inconsistent Results with Claygent Research on Large Datasets | Clay

9 comments

· Sorted by Oldest

Clay Support
APP
·
·
I understand how frustrating inconsistent Claygent results can be, especially with thousands of credits at stake.
Here are key steps to improve consistency:
Prompt optimization:
Make your search queries more specific and well-defined
Use our metaprompters to craft better prompts
Test with 10 rows first before scaling to your full dataset [1]
Model selection:
Try GPT-5 in Claygent - it shows 8-17% better research accuracy with 38% fewer tokens [2]
Consider Claygent Neon for complex research tasks - it has the highest accuracy with less null/inaccurate replies [3]
Data structure:
Break large requests into smaller batches
Use column outputs to structure complex results properly [4]
Before running on 70k rows, I'd recommend trying Sculptor (our free AI tool in the top-right corner) to help optimize your Claygent setup. It can help build better workflows and automations.
Have you tried using the metaprompter or testing with GPT-5/Neon models yet?
Muhammad S.
·
·
Would you be able to let me have a look at your prompt and what you are trying to achieve?
Clay Support
APP
·
·
I'd be happy to help optimize your Claygent setup for better consistency across 70k rows. [1]
To review your prompt effectively, I'll need to see the specific prompt you're currently using and understand what type of research you're trying to accomplish. [2] Our Clay University has examples of 11 different AI prompts for automating prospect research that might give us a good starting point for comparison. [3]
Before we dive into prompt optimization, I'd also recommend trying Sculptor (our free AI tool in the top-right corner) - it can help build better Claygent workflows and might spot optimization opportunities we haven't considered yet.
Could you paste your current Claygent prompt here along with a brief description of what specific information you're trying to research?
Clay Support
APP
·
·
If you still need help optimizing your Claygent prompt or workflow, I’m here to assist. Would you like to share your current prompt and the research goals you’re aiming to achieve?
Olaitan F.
·
·
I can't share the full prompt here as it is public but I'm trying to get a detailed company dossier
Clay Support
APP
·
·
Got it - for detailed company dossiers, structure is key to consistency. Use column outputs to break down your research into specific fields like funding stage, headcount, tech stack, recent news, etc. [1] This prevents Claygent from cramming everything into one messy response.
For complex company research, try the 3-credit model and be very specific about what goes where in your outputs. [2] Test with 10 rows first to nail the structure before your 70k run. [3]
Since you can't share the full prompt publicly, have you tried Sculptor yet? It's our free AI tool in the top-right corner that can help optimize complex Claygent workflows like company dossiers. It might spot ways to improve consistency that we haven't considered.
What specific data points are you trying to capture in your dossier?
Muhammad S.
·
·
huddle? Olaitan F.
Clay Support
APP
·
·
For consistent company dossier results at scale, structure is everything. Use column outputs to break your research into specific fields like funding stage, headcount, tech stack, recent news. [1] This prevents Claygent from cramming everything into one messy response and gives you predictable data structure. [2]
Make your prompts specific and well-defined, then break large requests into smaller batches. Always test with 10 rows first to verify quality before scaling to your full 70k dataset. [3] [4]
Since you mentioned trying expensive models already, have you tried Sculptor yet? It's our free AI tool in the top-right corner that can help optimize complex Claygent workflows like company dossiers and might spot consistency improvements we haven't considered.
What specific data points are you trying to capture in each dossier - funding info, team size, tech stack, or something else?
Clay Support
APP
·
·
If you still need help improving the consistency of your Claygent research for company dossiers, I’m here to assist. Would you like to share more details about the specific data points or challenges you’re facing so we can continue troubleshooting?

Clay Support
APP
·
·
I understand how frustrating inconsistent Claygent results can be, especially with thousands of credits at stake.
Here are key steps to improve consistency:
Prompt optimization:
Make your search queries more specific and well-defined
Use our metaprompters to craft better prompts
Test with 10 rows first before scaling to your full dataset [1]
Model selection:
Try GPT-5 in Claygent - it shows 8-17% better research accuracy with 38% fewer tokens [2]
Consider Claygent Neon for complex research tasks - it has the highest accuracy with less null/inaccurate replies [3]
Data structure:
Break large requests into smaller batches
Use column outputs to structure complex results properly [4]
Before running on 70k rows, I'd recommend trying Sculptor (our free AI tool in the top-right corner) to help optimize your Claygent setup. It can help build better workflows and automations.
Have you tried using the metaprompter or testing with GPT-5/Neon models yet?
Muhammad S.
·
·
Would you be able to let me have a look at your prompt and what you are trying to achieve?
Clay Support
APP
·
·
I'd be happy to help optimize your Claygent setup for better consistency across 70k rows. [1]
To review your prompt effectively, I'll need to see the specific prompt you're currently using and understand what type of research you're trying to accomplish. [2] Our Clay University has examples of 11 different AI prompts for automating prospect research that might give us a good starting point for comparison. [3]
Before we dive into prompt optimization, I'd also recommend trying Sculptor (our free AI tool in the top-right corner) - it can help build better Claygent workflows and might spot optimization opportunities we haven't considered yet.
Could you paste your current Claygent prompt here along with a brief description of what specific information you're trying to research?
Clay Support
APP
·
·
If you still need help optimizing your Claygent prompt or workflow, I’m here to assist. Would you like to share your current prompt and the research goals you’re aiming to achieve?
Olaitan F.
·
·
I can't share the full prompt here as it is public but I'm trying to get a detailed company dossier
Clay Support
APP
·
·
Got it - for detailed company dossiers, structure is key to consistency. Use column outputs to break down your research into specific fields like funding stage, headcount, tech stack, recent news, etc. [1] This prevents Claygent from cramming everything into one messy response.
For complex company research, try the 3-credit model and be very specific about what goes where in your outputs. [2] Test with 10 rows first to nail the structure before your 70k run. [3]
Since you can't share the full prompt publicly, have you tried Sculptor yet? It's our free AI tool in the top-right corner that can help optimize complex Claygent workflows like company dossiers. It might spot ways to improve consistency that we haven't considered.
What specific data points are you trying to capture in your dossier?
Muhammad S.
·
·
huddle? Olaitan F.
Clay Support
APP
·
·
For consistent company dossier results at scale, structure is everything. Use column outputs to break your research into specific fields like funding stage, headcount, tech stack, recent news. [1] This prevents Claygent from cramming everything into one messy response and gives you predictable data structure. [2]
Make your prompts specific and well-defined, then break large requests into smaller batches. Always test with 10 rows first to verify quality before scaling to your full 70k dataset. [3] [4]
Since you mentioned trying expensive models already, have you tried Sculptor yet? It's our free AI tool in the top-right corner that can help optimize complex Claygent workflows like company dossiers and might spot consistency improvements we haven't considered.
What specific data points are you trying to capture in each dossier - funding info, team size, tech stack, or something else?
Clay Support
APP
·
·
If you still need help improving the consistency of your Claygent research for company dossiers, I’m here to assist. Would you like to share more details about the specific data points or challenges you’re facing so we can continue troubleshooting?