M1 thesis: do video ads made with AI perform differently?
M1 thesis on 350+ video campaigns (YouTube, Meta, TikTok, connected TV): comparing the performance of ads with and without generative AI.
ExcelPythonAds Transparency CentersTikTok Creative CenterMeta Ad LibraryTrade Desk- 1
Context
This thesis comes out of my apprenticeship at Google France (2023-2024), with the program's agreement. The question: does using generative AI algorithms to produce video creatives dynamically change advertising and creative performance indicators on the main social video platforms and on connected TV?
- 2
What I did
I built the dataset myself: more than 350 campaigns collected from the Ads Transparency Centers, TikTok Creative Center, Meta Ad Library and Trade Desk. Every campaign follows a shared data dictionary (advertiser, creative, AI disclosure, AI tools, VTR, CPV, CPM, ROAS, Brand Lift), cleaned and structured in Excel and Python. I added five semi-structured interviews with advertisers, agencies and platforms, Google France among them, to understand how those teams judge AI creative work and its ethics.
- 3
What came out of it
One base to compare campaigns with and without AI across four platforms, with aligned metrics, and a cross-reading of the VTR and CPM gaps depending on whether AI was used. The interviews bring what the numbers do not show: how AI-assisted creative is perceived. The thesis was handed in on 17 September 2025.