diff --git a/migrations/005_transcription_segments.sql b/migrations/005_transcription_segments.sql
new file mode 100644
index 0000000..0aa0a75
--- /dev/null
+++ b/migrations/005_transcription_segments.sql
@@ -0,0 +1,26 @@
+-- Migration 005: Transkripsjonssegmenter
+-- Ref: docs/concepts/podcastfabrikken.md § 3
+--
+-- Master-kopi av alle transkripsjoner. SRT og ren tekst er avledede
+-- eksportformater som genereres fra denne tabellen.
+-- Universell for all lyd: podcast, møter, voice memos.
+
+CREATE TABLE transcription_segments (
+    id              BIGINT GENERATED ALWAYS AS IDENTITY PRIMARY KEY,
+    node_id         UUID NOT NULL REFERENCES nodes(id) ON DELETE CASCADE,
+    transcribed_at  TIMESTAMPTZ NOT NULL,  -- grupperer segmenter fra samme kjøring
+    seq             INT NOT NULL,
+    start_ms        INT NOT NULL,
+    end_ms          INT NOT NULL,
+    content         TEXT NOT NULL,
+    edited          BOOLEAN DEFAULT false,
+
+    UNIQUE (node_id, transcribed_at, seq)
+);
+
+-- Hovedindeks for oppslag: hent segmenter for en node, sortert
+CREATE INDEX idx_segments_node ON transcription_segments (node_id, transcribed_at, seq);
+
+-- Fulltekstsøk på tvers av transkripsjoner (norsk)
+CREATE INDEX idx_segments_fts ON transcription_segments
+    USING gin(to_tsvector('norwegian', content));