AI Content Pipeline
Used AI workflows to create UGC ads and content assets, then sent them to Perplexity unsolicited. The quality spoke for itself. They became a paying client.
Perplexity became a paying client
Replaced with AI workflows
Distinct ad format experiments
Content sold itself
The Problem
We wanted to prove that AI content creation workflows could produce professional-grade marketing assets — not just "good for AI" quality, but genuinely competitive with human creative teams. The test: could we create content good enough that a leading AI company would pay for it?
Separately, we were working with a client whose 5-person content team was a bottleneck. They needed the same volume of output but couldn't hire fast enough.
What We Built
For the Perplexity engagement:
- AI-generated UGC-style video ads
- Product content and social media assets
- Created everything using AI tools first, then sent to Perplexity unsolicited
- No pitch deck. No sales call. The work was the pitch.
For the content team replacement:
- End-to-end AI content workflows covering ideation, scripting, generation, and editing
- Same output quality as the 5-person team
- Fraction of the time and cost
We also built and tested 8 distinct ad format categories:
- Normal conversation ad (sales + value-based subcategories)
- Podcast-style character interview (without founder)
- Selfie-style video with conversational native tone and slang
- AI founder selfie video
- Street-style interview (news anchor format, streets of India)
- Past self vs future self emotional drama conversation
- Workplace scenario ads
- Testimonial-style narratives
For each format, we wrote multiple scripts and tested variations. This wasn't "generate one video and ship it." It was systematic creative experimentation.
The Numbers
| Metric | Value | Context |
|---|---|---|
| Client acquired | Perplexity | Without a pitch deck or sales call |
| Content team replaced | 5 people | Same output quality maintained |
| Ad formats tested | 8 categories | Each with 2+ script variations |
| Scripts per format | 2+ | A/B testing creative approaches |
| Production pipeline | Full-stack AI | Video, voice, and copywriting |
Key Engineering Challenges
- Consistency Across AI Characters: Getting AI-generated video characters to maintain visual consistency across multiple scenes and ad formats. The founder selfie video format was particularly challenging and couldn't be fully pulled off due to consistency limitations.
- Native Tone Replication: Creating AI characters that speak with authentic Indian conversational slang and cadence. The selfie-style format specifically required natural phrases and tone breaks that feel human, not scripted.
- Format-Market Fit: Each of the 8 formats performed differently with different audience segments. Had to build a testing framework to identify which format resonated with which demographic before scaling ad spend.
Want to build something like this? Let's talk.