AI Marketing · Selected Work
I write the marketing and build the machine that ships it.
Most marketers hand you a strategy and leave. Most engineers hand you a tool nobody uses. I do both, alone, and everything below is live work I made and can walk you through line by line.
Ellyn Shirk
Operator who builds. I run my own product company on systems I wrote myself, which is the same thing I would be doing for you.
Case Studies
Three things I shipped, start to finish.
No committee, no agency layer, no handoff. In each of these I did the positioning, the copy, the design, and the build.
A product page that had to earn its own conversions
ProcessForge is a SaaS product I built and shipped at getprocessforge.com. The first version of the landing page had the problems most founder-built pages have: two competing calls to action in the hero, a headline that described the category instead of the promise, and a cold navy interface that fought the warm brand of the page it sat behind.
I rebuilt it against what actually converts rather than what I liked looking at. One call to action instead of two. A promise in the subhead instead of a category description. A hero that agreed with what the product was claiming instead of quietly arguing with it. Then all the small things most people skip, which are usually the ones deciding whether a page feels finished.
Shipped April 2026. The point is not that I know these principles. It is that I applied them to my own product, where the cost of being wrong lands on me.
Add before sending
Drop in your real before and after numbers here: signups per week, or conversion rate, or time on page. One honest metric beats three vague claims. If you do not have clean numbers yet, delete this box and let the work speak.
What this proves
- I can diagnose why a page is not converting and say so plainly
- I write copy that carries a promise, not a category description
- I ship the fix myself instead of filing a ticket for it
- I have skin in the game on my own product
A children's book, and the thing that makes people pick it up
Penelope June is a picture book series I write. Four manuscripts drafted, a locked character system so the art can never drift between books, and a defined palette and voice that every asset is held against.
The marketing problem with a picture book is that it sits on a shelf next to a hundred others that look the same. So I built the differentiator instead of buying ads for it: a read-along app where the child reads the book out loud and the illustration animates as they say each word. Live microphone, forgiving enough for a four-year-old's pronunciation, no narration doing the reading for them.
It costs nothing per copy, it turns the physical book into the product and the app into the reason to choose it, and it is the kind of thing a parent tells another parent about. That is the whole marketing strategy, built rather than bought.
Add before sending
Once the book is out, add sales, preorders, or the response you got when people tried the read-along. Until then this reads honestly as a product and brand case study, which is what it currently is. Do not imply results you do not have yet.
What this proves
- I find the differentiator instead of defaulting to paid spend
- I hold a brand system tight enough that nothing drifts
- I can take an idea to a finished, working artifact alone
- I think about distribution while I am still making the thing
The marketing engine I built to run my own company
Everlight is my digital products business, and I run it solo. Rather than doing the marketing by hand every week, I built the system that does it and then supervised the system.
That stack includes an AI chief of staff that briefs me every morning on what needs attention, listing and content pipelines that produce on-brand assets from a single input, and an advisory panel I can put a decision in front of before committing to it.
This is the piece most people cannot show you. Plenty of marketers now use AI tools. I build the tools, wire them to real data, and keep them running. If you want AI doing actual marketing work rather than generating drafts someone still has to fix, this is what that looks like when it is finished.
Add before sending
Add one or two screenshots of the morning briefing and a content pipeline output. Reseed with demo data first so nothing real appears. Screenshots make this case study land far harder than the description does.
What this proves
- I build AI systems, not just prompts
- I automate the repetitive half so the judgment half gets attention
- I keep systems running against live data
- I use my own company as the proving ground
What I Do
Where I am actually useful.
Short version: the work between deciding what to say and having it reliably go out every week without anyone remembering to do it.
Positioning and copy
Landing pages, product pages, and launch copy that answer why someone needs this and what they get, in that order.
Conversion work
Diagnosing why a page is not converting, then rebuilding it. One call to action, a specific promise, no clutter.
AI content pipelines
Systems that turn one input into on-brand assets across channels, running on a schedule rather than on willpower.
Brand systems
A palette, a voice, and rules tight enough that everything produced later still looks like it came from one place.
Next Step
Tell me what is not working.
You probably do not need a full-time marketer. You need someone to build the thing properly and then stay to keep it running.
Send me the page, the funnel, or the thing you keep meaning to automate and have not. I will tell you what I would change and whether I am the right person to do it. If I am not, I will say that too.