Marcus Burke - How to Build a $2M/Month App (Copy Me)

In app-install ads, mention 'app' in the copy (e.g. 'we made this app...') or show an app demo so the click to the App Store isn't a surprise. Don't pretend not to be an app.

Concept: absurd hook scene (e.g. 'I showed my husband the app and he passed out instead of picking up the kids' - wife opens door, husband asleep on couch, ~5s) then a slowly zooming glowing green orb with on-screen lines on benefits/features of green noise. ~100 versions of this ad, ~400 iterations of the concept: new copy per discovered use case (e.g. tinnitus masking) in users' own language, and different sleep scenarios (inside/outside, more or less absurd).

Short (~10s, ~3 clips, odd low camera angle) raw video mimicking content someone in their 30s-40s would post, with a text block on screen the whole time. Text uses identity/pain buzzwords the audience self-identifies with (e.g. 'Gentle reminder that if you're waiting on their text, delaying your response, over-analyzing a non-existent future... we made this app to rewire toxic attachments and support your emotional regulation'). Copy comes from Reddit research. Text walls can also be turned into statics on varied backgrounds (post-it notes, fonts).

1) Find 10-20 high-engagement Reddit threads in the niche (anonymous, candid writing; comment volume signals importance); also your and competitors' app reviews and social comment sections. 2) Feed them to Claude and ask it to identify sub-audiences by language patterns, pain points, desires, and to note vivid emotional language and transformations. 3) Pick ~5 sub-audiences and have Claude write a separate context file for each. 4) Add a product context file (what it does, features, social proof, awards). 5) Prompt for e.g. three highly distinct text hooks, or paste a competitor text wall and ask for one in the same style/length using your audience's pain points. Without this context LLM copy is generic 'marketing blah blah'.

Freelancers make ~99% of client ads with AI. Use Higgsfield as a model aggregator (one subscription, credits). Generate the scene as an image first (GPT Image 2 for consistent text; Nano Banana, weaker on fine print) until it looks authentic, then use it as the reference frame for a video model (Seedance 2 for authentic UGC style, Veo 3, Kling) - far cheaper than prompting video directly. Once a format is proven (e.g. text wall: two scenes + app demo in the middle + text block on top), encode it as a node-based workflow in Weavy or Flora with variables to generate hundreds of iterations. Rationale is control over variables and future self-learning loops, not cost savings; creators on retainer can still be cheaper today.

Over-the-shoulder ad ('my wife is on this app every time she has free time') showing someone using the app with feature captions (Picnic). For a habit tracker (Me+): content-led 'weekly cleaning schedule for ADHD', each day's chore clip interleaved with a demo clip of checking off the list. Match demo depth to how much explanation the app needs and show how it fits into a day.

Web-to-app ads can sit far from the product: Nibble runs an educational geopolitics piece ('every intelligent person should know geopolitics') with a soft 'learn now' CTA; Liven runs a bare image that looks like a user post, text wall in the post copy, no product or marketing message. The web funnel then educates and sells before revealing the app. Statics work much better for web campaigns than app-promotion campaigns.

Around 2019-2020 at Blinkist, scaled Meta ads sending users to a website such as the blog rather than directly to the App Store.

Chart spend share across the six core placements (Facebook/Instagram feed, Reels, Stories) over time, and spend and cost per target by age bucket. Ask age in onboarding so you can compute trial->paid per age group from product analytics. Young traffic gives cheap trials but converts worst; older buckets cost more per trial but convert better, so a cheap-looking trial CPA can be the worst investment.

Chart each ad's placement mix with cost per target overlaid. Ads below target cost that deliver unusually (e.g. 40-50% Facebook feed vs 50/50 Reels for the rest) get starved when mixed in one ad set, since Meta concentrates ~80% of budget on one ad. Pull them into their own ad set with budget, still broad targeting and all placements, then check whether they bought an older audience and work out what made them deliver there (static vs video etc.).

One campaign (app promotion or web); ad sets per bucket: media type (static vs video), style (text walls, memes, UGC video) and message/pain point (overthinking, anxious attachment...). Budgets set at ad-set level, not campaign level, so Meta can't pour 90% into one bucket. Each bucket gets its own cost-per-target goal, e.g. memes bring engaging but lower-quality traffic so they must hit about half the target cost.

Cash-constrained/bootstrapped accounts: put all ads in one ad set, see what wins, repeat weekly until no longer losing money; split into more ad sets only once they can run at break-even. VC-backed teams can split by hypothesis from the start to learn faster.

At a previous company, each UA team member did one user interview per week, asking questions that get users to describe the problem in their own words.

Posts on LinkedIn 2-3 times a week; recently started a Substack (~2 pieces a month).

As a Meta ads consultant also works on ASO, onboarding and paywall/activation, since a minimal 'welcome, register, paywall' funnel optimized for a purchase event no longer converts paid traffic well enough.