Why Can’t AI Return a List of Users I Can See in the UI?

If I can see the users on the screen, why can’t the AI return that same list?

Short Answer: AI can only return what’s structured and queryable—not everything rendered in the UI.

Many platforms generate user lists dynamically based on: permissions, groups and roles. That view is assembled in real time—it’s not always stored as a clean, queryable dataset.

The core disconnect

There are two different layers at play:

• UI layer: builds views dynamically for humans

• Data layer: stores structured objects AI can access

AI usually operates on structured, indexed, or API-exposed data—not the rendered UI itself. So even if something is visible, it may not exist as:

• a defined object

• a retrievable list

• a stable relationship

Why this varies by platform

Different systems handle this differently:

Microsoft (Graph-based ecosystem):

• Strong identity model (users, groups, org structure)

• Many user relationships are queryable via APIs

• Still limited when views are permission-derived or context-specific

Atlassian (Teamwork Graph model):

• Focused on work objects (issues, pages, tasks)

• User relationships are more indirect

• Some UI-level groupings are not first-class queryable entities

Glean (enterprise search layer):

• Aggregates across systems

• Can sometimes surface user-related insights via indexing

• Still depends on what source systems expose and allow

Across all three, the same rule applies: If the data isn’t explicitly modeled, AI won’t reliably return it.

Where AI struggles most

AI interfaces tend to fall short when data is:

• permission-derived at runtime

• embedded in UI components

• generated through logic rather than stored directly

This includes:

• “who has access” views

• dynamic membership lists

• contextual user groupings

Can you fix it?

Sometimes. Options include:

• exposing user data through APIs or directories

• structuring lists explicitly in content

• using systems where identity is a first-class object

But if the platform doesn’t model or expose it, AI should not invent it.

Takeaway

AI doesn’t “see your screen.” It queries structured data models.

If something is:

• dynamically generated

• context-specific

• not stored as a defined object

It may be visible to you—but invisible to AI. The gap isn’t intelligence—it’s data design. If you want AI to return it, it has to exist in a way the system can query.

Use this Prompt: Act as an enterprise AI data-access reviewer. Review this user-list scenario: [describe the platform, screen, visible user list, roles, groups, permissions, and what the AI is being asked to return]. Determine whether the list is likely stored as a queryable object, generated dynamically from permissions or roles, embedded in the UI, or available through an API. Return the answer as a table with: visible list, likely data source, whether AI can retrieve it, why it may fail, permission or API requirement, and test needed to confirm.