How AI fields work
When you create a field with autofill enabled:
You provide instructions describing what the field should contain
The AI analyzes available data from the record and related records (meetings, emails, associated companies/people). Comments on records are not included as a data source.
The AI populates the field following your instructions
Autofill writes the value straight to the field. There is no suggestion to review and nothing to accept, which is what separates it from Deal suggestions and suggested deal field updates. To check how a value was reached, use AI reasoning on the field.
Context limits: When you reference related records (like meetings or emails), the AI uses the 10 most recent records. If 100 meetings exist, only the latest 10 are analyzed.
Fields with autofill enabled show a ✨ icon next to the field name in lists.
Create a field with AI autofill
Two ways to create fields with autofill:
Generate with AI (recommended)
This approach writes the field configuration and autofill instructions for you.
From any list view, click Add new field
On the Generate tab, describe the field you want or select a smart suggestion
Click Generate
Clarify creates the field name, selects the field type, and writes autofill instructions based on your description.
Then,
Review the configuration on the Configure tab
Edit if needed, then click Create
Choose an option:
Create and autofill first 10 rows (recommended for testing)
Create and autofill all rows
Create and don't autofill
Configure manually
Use this when you know exactly what field type and instructions you need.
From any list view, click Add new field
On the Configure tab:
Enter the field Name
Select the field Type
Toggle on Autofill with AI
Enter autofill Instructions (see Writing instructions below)
(Optional) Click Add context to reference related records
(Optional) Enable Internet research if the AI needs to look up public information
Click Create and choose an autofill option
💡 Relationship fields (early access): AI autofill for relationship fields (many-to-one, one-to-many, many-to-many) is rolling out and may not be available in your workspace yet. When enabled, the AI picks and links the most relevant candidate records automatically.
Configure autofill on existing fields
To enable or modify autofill on an existing field:
From any list view, hover over the field header and click the dropdown icon
Select Edit field
Toggle on Autofill with AI (if not already enabled)
Update the Instructions
Configure Add context or Internet research if needed
Click Save
To delete a field, follow the same steps but click Delete field in the edit window.
⚠️ Built-in fields can't autofill. Fields that ship with Clarify, including Job title, Phone numbers, Primary location and Email addresses, have no Edit field option, so autofill can't be turned on for them. Create a new field to hold the AI value and keep the built-in field for what you enter yourself.
✨ Rewrite with AI: If your existing instructions use plain language instead of context variables, click Rewrite with AI in the instructions box. Clarify restructures your prompt to include the right variables while preserving your intent.
Writing autofill instructions
Good instructions are specific and include the expected output format.
Format matters for text fields:
If you want names: "Return the person's full name, like 'Sarah Chen'"
If you want yes/no: "Return 'Yes' or 'No'"
If you want lists: "Return a comma-separated list, like 'Competitor A, Competitor B'"
Handle missing data explicitly: When the AI cannot determine a value, tell it what to output instead of leaving the field empty.
Example: "If no churn reason is found, return 'No reason identified'"
Reference additional context: Click Add context (or type {) to add related records to your instructions. This gives the AI access to data beyond the main record.
Example: To analyze meeting history with a company, insert
{People > Meetings}to include the last 10 meetings with people at that company
⚠️ Instructions alone don't grant data access. Writing "analyze meetings and emails" in your prompt doesn't give the AI access to that data. Insert the variable with Add context (or type {). Without it, the AI can only see fields on the record itself.
Which variables are available
Email context is called Messages. Which variables you see depends on the object the field is on:
Field is on | Variables |
People | Company, Meetings, Messages, Company > Deals |
Companies | People, Deals, People > Meetings, People > Messages, Deals > Tasks |
Deals | Company, People, Company > Deals, People > Meetings, People > Messages |
Custom relationship fields on the object appear as variables too.
The prefix tells you whose data you get. On People, Messages means that person's own emails. On Companies and Deals, People > Messages means emails from everyone linked to that record, still capped at 10 in total rather than 10 each. Pick the object that matches the value you want to store: per-person details belong on People, where the field reads that person's own email.
Be specific about data sources: Tell the AI where to look.
Good: "Analyze the last 5 emails and meetings to identify competitors mentioned"
Vague: "Find competitors"
Run autofill
From list views
Click the header of any field with autofill enabled (✨ icon) and select:
Autofill first 10 rows - Test and validate before running on all records
Autofill all rows - Process all records in the current list
Autofill empty rows - Only fill records where the field is currently empty
💡 Filter first: Apply filters to your list before running autofill to process only specific records.
From individual records
Open a Deal, Company, or Person record
Click the field with autofill enabled
Select Run autofill
This processes only the current record.
Field processing states
After triggering autofill, records progress through these states:
Queued: Request received, waiting to process
Processing: AI is analyzing and generating output (displays "Shaping," "Spinning," "Ideating," or "Pondering")
Complete: Field is populated
View AI reasoning
To see how the AI populated a field:
Hover over any autofilled field
Click 🗒️ AI reasoning
This shows the logic and data sources the AI used to generate the value.
What this costs
AI autofill consumes credits per record per field:
Internet Research | Credits per record |
Disabled | 3 credits |
Enabled | 30 credits |
Example: Autofilling 100 records on one field = 300 credits (no internet research) or 3,000 credits (with internet research).
When to enable internet research: Use it when the AI needs publicly available information not in your CRM (company funding data, headcount, news). Disable it when analyzing data you already have (meeting notes, emails, deal history).
Internet research cannot access authenticated pages (LinkedIn profiles, paywalled content).
Use cases
Use cases
Deals
Identify competitors
Track which alternative solutions your prospects are evaluating.
Field type: Multiline text
Instructions: "Analyze recent meetings and emails to list competitors or alternative solutions mentioned. Return a bulleted list. If none are mentioned, return 'No competitors identified.'"
Internet research: Disabled
Cost: 3 credits per record
Find deal blockers
Surface product gaps or feature requests preventing deals from closing.
Field type: Multiline text
Instructions: "Review recent emails and meetings for items explicitly described as blockers (feature requests, product limitations, pricing concerns). List each blocker. If none exist, return 'No blockers identified.'"
Internet research: Disabled
Cost: 3 credits per record
Capture win/loss reasons
Automatically document why deals closed won or lost.
Field type: Multiline text
Instructions: "Analyze the most recent 5 emails and meetings to determine why this deal was won or lost. Provide a concise 1-2 sentence explanation. If no reason is evident, return 'Reason not documented.'"
Internet research: Disabled
Cost: 3 credits per record
Calculate derived values
Compute numeric results directly from other field values - like weighted pipeline, deal scores, or margin calculations. AI fields handle arithmetic accurately; this is a calculation, not inference.
Field type: Number
Instructions: "Multiply Amount by Close Probability divided by 100. Return only the number, no symbols or text. If either value is missing, return 0."
Internet research: Disabled
Cost: 3 credits per record
Automatic re-run when records change is in early access and may not be available in your workspace yet; otherwise, re-run autofill manually.
Companies
Score ICP fit
Evaluate whether companies match your Ideal Customer Profile criteria.
Field type: Single select (options: "Strong fit," "Moderate fit," "Weak fit")
Instructions: "Evaluate this company against our ICP criteria: [insert your specific criteria like industry, employee count, revenue range, tech stack]. Return 'Strong fit' if all criteria match, 'Moderate fit' if 2-3 match, 'Weak fit' if fewer than 2 match."
Internet research: Enabled
Cost: 30 credits per record
Research company funding
Track funding rounds and total capital raised for VC-backed companies.
Field type: Number
Instructions: "Research and return the total funding raised in millions (USD). If no funding data is available, return 0."
Internet research: Enabled
Cost: 30 credits per record
Document churn reasons
Capture why customers canceled based on exit interviews or emails.
Field type: Multiline text
Instructions: "Review recent emails and meetings where cancellation was discussed. Summarize the stated reason for churn in 1-2 sentences. If no reason is mentioned, return 'No reason provided.'"
Internet research: Disabled
Cost: 3 credits per record
Limitations
Context is limited to 10 recent records
When your instructions reference related records (meetings, emails, etc.), the AI analyzes only the 10 most recent. If you have 50 meetings with a company, only the latest 10 are considered.
Emails are trimmed before the AI reads them
Quoted reply history is removed from each email, so earlier messages in a thread are only read as their own separate records. Signature blocks are often removed as well, depending on the email client that sent the message. Contact details typed into the body of a message come through reliably; details that appeared only in a signature often don't. Very long emails are shortened, and the AI is told when that happened.
Meeting transcripts are not available
The AI cannot access full meeting transcripts. It uses meeting summaries only.
Comments are not available as a data source
AI fields cannot access comments on records. If you need the AI to analyze specific context, ensure it's captured in meetings or emails rather than comments. Note: Deal Summaries and Rep can access comments - this limitation is specific to AI fields.
Auto-run requires configuration
By default, autofill does not run automatically - you run it manually. Automatic re-run on record create or update is in early access and may not be available in your workspace yet.
Internet research is limited to public information
The AI cannot access:
Pages requiring authentication (LinkedIn profiles, internal systems)
Paywalled content
Private databases
Troubleshooting
Field populated with incorrect or irrelevant data
The instructions may be too vague. Revise to be more specific about:
Which data sources to use
What format the output should take
How to handle missing data
Run autofill on 10 rows first to validate before processing all records.
Field left empty when data should exist
Check if:
The AI has access to the necessary data (use Add context to add related records)
Your instructions tell the AI what to output when data is missing
Internet research needs to be enabled for external data lookups
Field shows a warning icon instead of a value
Autofill failed on that record. Hover the icon to see why:
A temporary AI provider error - run autofill again on that record.
Too much data to process for that record - trim the field’s inputs before running again. Retrying without changing anything will fail the same way.
You can always edit the value by hand instead. If the field already had a value, it stays visible next to the warning icon.
This is different from a field left blank because no matching data was found - see above.
Autofill consumes more credits than expected
Each record processed consumes credits per field:
3 credits per record (no internet research)
30 credits per record (with internet research)
If you autofilled 100 records on 3 fields with internet research enabled, that's 9,000 credits (100 × 3 × 30).
Review which fields have internet research enabled and disable it if not necessary.
FAQ
FAQ
Why is variable context limited to the last 10 records?
To balance AI performance with context quality. The AI analyzes the most recent data to ensure relevance.
Can the AI see meeting transcripts?
No. The AI uses meeting summaries only. Full transcripts are not provided.
Can I automatically re-run autofill?
Automatic re-run on record create or update is in early access and may not be available in your workspace yet. Otherwise, autofill runs manually.