# Tags and descriptions

Every analyzed file carries a caption, a description, keywords and tags produced by the
AI. All of it is editable, and editing it is worth the time: search, smart albums and
agents all read these fields.

## Edit a file's analysis

Open a file and go to the **Analysis** tab. Everything the AI found is listed in full,
and every field can be corrected.

Two fields repay the most attention:

**The caption** appears under the file everywhere in the app, and it is what an agent
quotes when it cites the image. If it reads "a group of people at a table" when the
photo shows your executive team at the annual meeting, correct it.

**The keywords** are what search matches. Add the terms your organization actually uses
and that a model could not have guessed: product names, event names, the internal name
for a campaign.

## Tags

Tags are the controlled half of the vocabulary. Keywords are free text; tags come from a
list your workspace maintains, so everyone tagging "Q3 Campaign" spells it the same way.

Add and remove tags from the file's Overview tab, or select several files in the library
and tag them in one action.

Applying tags and defining them are separate permissions, in three steps. Anyone who can
edit media can apply the tags that already exist. Adding a new value to a group is a
permission of its own, which the built-in roles give to administrators. Changing the
structure (adding, renaming or removing the groups themselves) is an administrative
action under **Administration → Media Tags**.

If you want the people doing the filing to be able to add values as they go, grant that
one permission to their role rather than making them administrators. See [users and
roles](/guide/admin/users-and-roles).

Grouping tags keeps related ones together in the picker. Keep the list short enough that
people read it rather than inventing a new tag each time: twenty tags everyone uses beat
a hundred nobody can find.

**Hashtags** are managed separately under **Administration → Media Hashtags** and are
for social posts rather than for organizing the library.

## Flags

Flags mark state rather than content: "needs review", "approved for external use", "do
not use". Configure them under **Administration → Flags**. Use a flag for anything that
records a decision about a file rather than a description of it.

## Analysis schemas

What the AI looks for is configurable. **Administration → Media Analysis** lets an
administrator define what a given kind of file should be analyzed *for*: a product
catalog needs different fields extracted than an event photography library.

Set this up early. Every file analyzed afterward uses it, and files already in the
library can be brought up to date by selecting them and choosing **Reanalyze**.

## Bulk edits

Select several files in the library to tag them, flag them, add them to an album, or
reanalyze them in one action. There is no bulk caption edit, by design. A caption that
fits two hundred files is too generic to be useful.

## Why it matters

Three things improve as soon as the descriptions are correct:

1. **Search finds files.** Most "I can't find it" is a caption problem.
2. **Agents cite the right file.** An agent asked for a photo of the leadership team
   picks the correct one because the caption identifies it.
3. **Generated work uses better images.** The presentation generator and the design
   editor choose from the library by matching descriptions against what the slide or
   layout needs.