How to Create and Edit Notion Database Properties With AI

Notion offers two useful AI routes for database structure: Build with AI when creating a new database, and Notion Agent when working with an existing one. They are not the same as an assumed “Generate with AI” button in every Properties panel.

Specify field names, types and boundaries before asking for changes. Work on a test database or retained copy first, especially when existing formulas, relations or automations depend on the schema.

Define the Properties Before You Ask AI to Add Them

For a project tracker, write a short specification: Project name as the title, Due date as a date, Owner as a person, and Priority as a select with Low, Medium and High. State that existing fields and records must be preserved. Avoid “make a complete project system” if you only want three new properties.

Check the workspace’s AI access and your editing permissions. Business/Enterprise include Notion AI access subject to current usage rules; do not assume a separate historical add-on is always required. See the current AI FAQ.

AI may infer the wrong type or create a near-duplicate field. The requested schema is your acceptance checklist, not a guarantee of the generated result.

Create a New Database With Build With AI

1. On the destination page, enter /database and choose Build with AI. A new page’s Table starter is another documented entry point.
2. Describe the records, field names and types you want. Include the intended view if it matters.
3. Inspect the database preview. Ask for specific changes and remove unnecessary fields before continuing.
4. Select Continue, choose the visualization and finish with Done.

This is the new-database builder described in Notion’s database AI guide. Do not treat that builder’s limitations as a description of every capability of the separate Notion Agent.

Before adding real records, create two harmless examples. Confirm dates sort correctly, person fields reference the right people, and select options match your process.

Add Properties to an Existing Database With Notion Agent

1. Open the intended database and open Notion Agent using the face icon. Explicitly mention the database so similarly named trackers are not confused.
2. Ask for a plan first: “List the properties currently present and propose only the missing fields from this specification. Do not edit anything yet.”
3. Check the plan for duplicate names and unintended type changes. Then authorize the exact additions to your test copy or approved database.
4. Inspect the actual property list and sample records after the agent completes the work. Do not assume every operation has a separate Add to database confirmation screen.

For example: “Add a Date property called Review due and a Select property called Review state with Pending, In review and Complete. Preserve all existing properties and page contents; do not populate or delete records.”

Notion Agent’s documentation supports creating and editing database properties, including relations, and says the agent has your permissions. It can use existing context; the former claim that it cannot read the database while changing its schema is incorrect.

Validate Types, Dependencies and Advanced Fields

Open each new property’s settings and verify the actual type, name and options against your specification. A field labeled “Email” may still be plain text if created incorrectly. Check relations point to the intended database and that no existing property was renamed or converted accidentally.

Handle advanced fields individually. Notion’s current Agent documentation is inconsistent about formula creation, so this guide does not promise either universal formula support or a blanket inability. You can configure the property manually and ask AI to explain a proposed formula, then test it against known results. Review Notion’s property reference for the correct type and behavior.

Test any dependent views, formulas, relations, rollups and automations before applying the same change to a production tracker. A successful-looking chat reply or a faster setup time is not evidence that the schema is correct.

For a small, precise change, manual property creation may be simpler. Use AI when a clear specification helps draft multiple related fields, then keep the final structure under human review.

Official documentation checked: October 14, 2026.