diff --git a/database/schema/schema.sql b/database/schema/schema.sql
index 38d4c6d3..9370a117 100644
--- a/database/schema/schema.sql
+++ b/database/schema/schema.sql
@@ -174,7 +174,7 @@ CREATE TABLE IF NOT EXISTS media_items (
     language VARCHAR(10) DEFAULT 'en', -- Language code (ISO 639-1)
     edition VARCHAR(100), -- Edition information
     page_count INTEGER, -- Total page count
-    genre VARCHAR(100), -- Genre classification
+    genre TEXT, -- Genre classification (smart-genre shelf OR raw tag free text — Audible category strings exceed 100 chars)
     copyright_year INTEGER, -- Copyright/publication year
     goodreads_id VARCHAR(50), -- Goodreads identifier
     openlibrary_id VARCHAR(50), -- Open Library identifier
@@ -279,6 +279,14 @@ CREATE INDEX IF NOT EXISTS idx_media_items_contributors_gin ON media_items USING
 ALTER TABLE media_items ADD COLUMN IF NOT EXISTS tags_search TEXT[];
 ALTER TABLE media_items ADD COLUMN IF NOT EXISTS contributors_search TEXT[];
 
+-- Widen genre for existing databases: the audio tag fallback stores the
+-- raw file genre atom (Audible category strings exceed 100 chars —
+-- e.g. "Politicians, Presidents & Heads of State, United States, ..."),
+-- and VARCHAR(100) rejected the whole import/update at the database
+-- layer (SQLSTATE 22001). VARCHAR -> TEXT is a metadata-only change,
+-- safe to re-run (media_highlights position widening precedent).
+ALTER TABLE media_items ALTER COLUMN genre TYPE TEXT;
+
 -- Audiobook editorial facts (A2a): narrators are user-editable first-class
 -- (search/filter/badge surface); abridged rides along; audio_metadata keeps
 -- the scan-derived tech record (duration/codec/chapters).