Understanding AI Enrichment Column Value Access in Data Analysis

John T. · 2026-01-06T06:34:43.482Z

I'm curious, why sometimes I am able to reference the values inside an AI enrichment column, but sometimes I'm not able too (meaning I have to extract out the values inside the AI enrichment column into a seperate column)

Understanding AI Enrichment Column Value Access in Data Analysis | Clay

Clay Support
APP
·
·
The ability to reference AI enrichment data depends on how the data is structured. You can reference data at different levels: [1]
When you CAN reference directly:
Entire enrichment results
Specific endpoints within structured data
Individual list items when the AI returns organized data [2]
When you need to extract separately:
When the AI returns unstructured text that needs parsing
When you need specific values that aren't clearly defined endpoints
When the data structure doesn't allow direct referencing [3]
For extraction, you can either use AI formulas to pull specific info or click on rows and hover over fields to "Add as column" - this doesn't cost extra credits since you're just reformatting existing data.
Clay Support
APP
·
·
Did that answer your question?
Muhammad S.
·
·
Any example?
Clay Support
APP
·
·
Sure! Here are concrete examples:
When you CAN reference directly:
Website Summary enrichment: You can reference the entire enrichment or specific endpoints like "summary" or "industry" [1]
People data with structured lists: Reference specific employee data like /people[0]/first_name to get the first person's name [2]
Experience data: Reference /experience[0]/url to get the URL from someone's latest job [3]
When you need to extract separately:
Counting experiences: If you want to know "how many jobs has this person had?" you'd use an AI formula to count the total experiences [4]
Concatenating data: To combine all company names into one string like "Google, Apple, Microsoft"
Complex analysis: When the AI returns unstructured text that needs parsing
The key difference is structured vs unstructured data - if it has clear endpoints, you can reference directly. If you need analysis or formatting, extract it.
Clay Support
APP
·
·
Is that what you were looking for?
John T.
·
·
For example this. My AI cell I can't reference the values inside unless i extract out in another column
Clay Support
APP
·
·
Looking at your screenshot, this happens when your AI enrichment returns unstructured text rather than structured data with clear endpoints. [1]
Your "Persona Category (2)" column likely contains plain text output from the AI (like "Marketing Manager" or "Sales Director") instead of structured data with specific fields you can reference. [2]
To extract the values, you can:
Click on a row and hover over the field, then select "Add as column"
Use an AI formula in a new column to parse the text
This extraction doesn't cost extra credits since you're just reformatting existing data.
The key difference: structured enrichments with clear endpoints can be referenced directly, but unstructured text output needs extraction.
Clay Support
APP
·
·
Was that helpful?
John T.
·
·
how to i force it to have struuctured data?
Clay Support
APP
·
·
To get structured data from your AI enrichment, you need to define specific output fields when setting up the column:
1.
When creating your AI enrichment, go to "Add and define outputs"
2.
In the Fields section, enter specific field names where you want the output to appear (like "category", "score", "reason")
3.
Use the dropdown to select the appropriate data type for each field
This forces the AI to return structured data with clear endpoints you can reference directly, instead of unstructured text. [1]
Without defined output fields, AI returns plain text that requires extraction.
Clay Support
APP
·
·
Was that helpful?
Muhammad S.
·
·
Generate the schema via the prompt for consistent output structure.
Clay Support
APP
·
·
Perfect! You've selected JSON Schema which is the best way to get consistent structured output.
Click "Generate from prompt" - it'll analyze your existing prompt and automatically create a JSON schema that matches what you're trying to extract. [1]
This forces the AI to return data in the exact structure you define, so you can reference specific fields directly instead of extracting later. [2]
The schema ensures consistent field names and data types across all rows, preventing variations like "unhappy" vs "dissatisfied" - everything gets mapped to your defined structure.
Try it and let me know how it works!
Clay Support
APP
·
·
Is that what you were looking for?
John T.
·
·
yes i am already doing this But still same issue

18 comments