Critical Issue: High Failure Rate in Claygent AI LinkedIn Columns | Clay

Critical Issue: High Failure Rate in Claygent AI LinkedIn Columns | Clay

Clay Support
APP
·
·
To troubleshoot your Claygent failures with LinkedIn profiles:
Immediate steps:
Click into the specific cells showing "unable to verify" errors to see detailed error messages
Check if your prompts are specific enough—vague queries can cause failures
Verify your API usage limits and tiers aren't being exceeded
Common causes:
Rate limits on your API tier (especially with heavier prompts)
Prompt design issues—if the same failure appears 5+ times, it's typically a prompt problem [1]
To improve accuracy:
Review failed validations and identify patterns
Update prompts with specific instructions to handle failure patterns
Break large requests into smaller batches
Please share the detailed error messages from the failed cells so we can troubleshoot further.
Clay Support
APP
·
·
Is that what you were looking for?
Clay Support
APP
·
·
If you still need help resolving the "unable to verify" errors with your Claygent AI columns, I’m here to assist. Would you like to provide more details about the specific error messages or patterns you’re seeing?
John D.
·
·
Hi Clay Support Team, I am formally requesting to be connected with a live support agent or a technical specialist to discuss a high-priority issue. We have been corresponding with your automated support system, but the generic troubleshooting steps provided are not applicable to our situation, and we are not making progress. We have already determined that this is not a prompt issue or an API limit issue. The core problem is a systemic, ~50% failure rate of Clay's native LinkedIn scraping function. We are using our own OpenAI API key (GPT-4o Mini), which confirms the failure is happening within Clay's data retrieval step, before our API is even called. This is a critical issue that is cutting our lead volume in half. We need to speak with a person who can look beyond the standard script and investigate a potential technical problem with your platform's infrastructure. Please arrange for us to connect with a live agent
Clay Support
APP
·
·
I understand how critical this issue is for you, and I’m routing you to one of our human agents who can assist further.
Intercom
APP
·
·
Clay Support is sharing a file with you.
Our support team has got your message and we'll get back to you soon!
If you’re dealing with a specific table, drop the URL below so we can help you quicker. Otherwise, someone from our team will be in touch soon!
Loading attachments
Sayli G.
·
·
Hello John! Thanks for reaching out! To summarise the issue you are facing high error rates with Linkedin scraping AI columns with a 'Unable to Verify Error'. You said you confirmed that the error is in our data retrieval step and that the API is not even being called. Does that mean that API spend does not show up because it fails beforehand? Would like the table link URLs of the tables which were affected by this. Upon collecting all context I can loop in an engineer if needed. Also to confirm you are saying these used to work and don't work anymore?
John D.
·
·
To answer your other questions:
1.
Yes, these exact workflows and prompts were working with a very high success rate until recently. This is a new and sudden degradation in performance.
2.
Here are the URL links for the tables
https://app.clay.com/workspaces/42993/workbooks/wb_0swzkc7x7XNzRN2Sm7z/tables/t_0t2bxm7CTbTtie53uaT/views/gv_0sy2t0174mntNokBUB4 https://app.clay.com/workspaces/42993/workbooks/wb_0swmxk1MaRjSRBRunm6/tables/t_0t432gyRsG6bxn8WuZj/views/gv_uuYKwiKmbdE3 https://app.clay.com/workspaces/42993/workbooks/wb_0t0isfaNa42TQw9vv5G/tables/t_0t3248fHrB2hgF9iW42/views/gv_uuYKwiKmbdE3 https://app.clay.com/workspaces/42993/workbooks/wb_0syvi5kh3fFdsUrwEVD/tables/t_0t3q7aznFStk4xw6TZo/views/gv_0sy2t0174mntNokBUB4 Given that we are paying for API calls that are failing due to the poor quality of the scraped data, this strongly points to an issue with the reliability of your data retrieval step. Please proceed with looping in an engineer to investigate why the LinkedIn scraper is now consistently returning bad data.
Sayli G.
·
·
Thankyou for the additional context. I have a few more questions here: 1. When did this start happening? You said it started happening a few weeks ago. 2. To confirm you are getting charged on your API key? I see you mentioned before that it is failing before even calling the API. Want to confirm the exact details of the situation. 3. Do you have a table link URL of a table where this was working beforehand? This will be good to do a comparative analysis on intended vs current behavior. I have reached out to the eng team on this. Would like this additional context while I sync with them.
John D.
·
·
Thanks for getting back to us and for looping in the engineering team. Here is the additional context you requested:
1.
When did this start happening? We first noticed this significant increase in the failure rate approximately two weeks ago.
2.
Confirmation of API Charges: My apologies for the earlier confusion on this point. I can now definitively confirm that yes, we are being charged an API call for every row that returns the "Unable to Verify" error. My initial assumption was incorrect. This reinforces our belief that Clay is successfully passing something to our API, but the data being passed is unusable, causing the call to fail and waste our credits.
3.
Comparative Table URL: Unfortunately, we do not have a preserved "before" state of the tables. As a standard part of our operational workflow, we clear out old, processed data from our tables to stay within our 50,000-row limit. But those are the same tables we ran before having this error
Sayli G.
·
·
No worries! The team is aware of the issue and are working on the fix. Some changes were made on Linkedin's end that affected us as well and hence led to the bad data you are talking about! Will keep you updated!
1
John D.
·
·
Hello Sayli G., Any updates regarding this issue?
LuisArturo
·
·
Hey there John jumping in here for Sayli, apologies the team is still working on this. Although it looks like they may have found what the source of the issue is and are working to address it
John D.
·
·
Alright thank you for that update