AI Methodology
Obadiah writes EmpiresDiary. AI does the heavy first-draft work. The audit catches factual problems. He does the editing, and he decides what publishes.
Text Generation
Each episode begins as an AI draft. The prompts behind those drafts are written by Obadiah. They specify the time period, the reading level, the dramatic shape of each episode, and the kinds of details a serious history piece needs: dates, named people, places, institutions.
A draft is just a draft. It does not become a published episode until it has cleared a factual audit and Obadiah has read the result for himself.
Author Review
After the audit pipeline runs, Obadiah reads the episode end to end. He is the last gate before anything publishes. Anything the audit missed gets caught here: factual mistakes, weak paragraphs, anachronisms, hedge-words used in the wrong place.
Episodes that fall short of the editorial standards are held back. The schedule never overrides the standard.
Factual Audit And Revision
- An automated reviewer reads each draft, looking for false claims, overconfident wording, broken chronology, and anachronisms.
- Drafts that fail get a revision pass that fixes the flagged issues.
- The revised draft is checked a second time before publication.
- Source notes and citations stay with the published episode so readers can see what the article was built on.
Audio And Visuals
Narration is generated from the final audited text of each episode. Artwork is generated from prompts that lock period details: era, geography, architecture, clothing, military equipment, and anything else the scene needs to get right.
Generic “ancient ruins” imagery is off the table. Each image is supposed to reflect the actual time period and region the episode is about, not a vague antique aesthetic.
Limits
AI gets things wrong sometimes. The audit and the corrections process are part of how this site publishes, not a polish layer added after the fact.