Marcus Burke
Meta ads consultant for subscription apps, 14 years in paid social (ex-Blinkist, Innogames)
Where to hear and read about Marcus Burke1 episode or test
Marcus Burke is a Meta ads consultant for subscription apps with 14 years in paid social, coming from Innogames and Blinkist. The thing most worth learning from him is to write ads in the exact words of narrow audiences, then structure the account so each audience and format gets its own spend.
message plus media plus content type times funnel equals audience.
How to Build a $2M/Month App (Copy Me)
Marcus Burke: the story and the lessons
1Marcus Burke's path from mobile games to subscription apps
Burke learned paid social in mobile gaming at Innogames. He started at Innogames in Hamburg as games moved from browser to mobile, after training as an air traffic controller. Early mobile gaming was trial and error with few examples to copy, and Clash of Clans was everyone's reference point.
His head of marketing told him Facebook is dead when he joined, because performance was degrading.
Big spend bought Burke's team early access to Meta's products. Spending millions a month alongside King and Supercell got them invited to Meta headquarters events and into beta products first. Meta introduced app event optimization and value optimization during this period.
At Blinkist he pioneered sending ads to a blog, not the app store. He joined Blinkist in 2019 through a friend from Innogames, then worked at Study Smarter and Tandem.
He went freelance in 2022 because management work had become hiring, reporting and administration.
Burke says he has scaled one app to $2M a month. He has run Meta ads for apps for 14 years, and his best-performing concept, a sleep-sound ad for Moonate, has passed $1M in spend.
Both figures come from Burke alone, with no spend or revenue data shown.
2How Burke writes Meta ads for subscription apps
Burke makes app ads say plainly that the product is an app. An ad that hides the app makes the App Store landing feel jarring and hurts install rate. His fix is to mention the app in the copy or show a demo.
"you want to prepare people for that they are going to land on the app store otherwise your install rate tends to suffer."
He refuses to copy D2C talking-head formats one to one. The ecom pattern of hook, pain points, then product is not the best format for apps. He forms a hypothesis about why an ad worked for its product before borrowing it.
Organic virality needs broad, non-salesy content, so he borrows only the native feel of a format, not the whole thing.
Burke writes in the audience's own pain-point language. Ad sets run broad, so Meta picks the audience from who engages. Specific phrasing like "rewire toxic attachments" draws the right engagers, while generic copy pulls a bigger audience that converts worse.
The best copy borrows words people use to label themselves, then plugs the product in as the fix. A line like "we made this app" also primes the App Store landing.
He feeds 10 to 20 Reddit threads to Claude for audience language. He asks Claude for sub-audiences, language patterns, pain points, desires and vivid emotional phrases. Reddit works because users write anonymously and bluntly, and big threads signal what matters.
App reviews, competitor reviews and comment sections are other written sources.
Claude gets one context file per sub-audience. Each file sits alongside the app's features, social proof and awards. Burke then asks for several distinct hooks, or feeds in a competitor's text wall as a template.
Without that context, an LLM falls back on generic internet marketing language. Claude writes strong copy but can't tell which formats convert. "The key is context."
Burke lets onboarding do the selling and keeps the ad soft. Many subscription apps have a long onboarding where the sale happens, so emotional, pain-point-led text with a light app mention is enough. His text-wall ads run about 10 seconds.
How much to demo depends on the category. A habit tracker like Me+ needs little explanation, while Picnic uses over-the-shoulder demos with feature captions.
The Moonate concept sells a scenario first and the app second. A five-second story hook about a husband who passed out on the couch leads into a slowly zooming glowing orb with a new line every couple of seconds. It had about 400 iterations.
Variants changed scenes and copy, and tinnitus-specific versions came from learning that many users play noise to mask tinnitus.
He interviews users weekly to learn how they describe the problem. At a past company, everyone on the UA team did one user interview a week, and much of their copy understanding came from it.
The questions must get people talking about the problem. He treats chasing social trends as secondary early on.
3How Burke uses AI to produce ad creative
Burke uses AI to control creative variables, not to cut cost. His freelancers make about 99% of his clients' ads with AI, yet it is not cheaper yet because generations rack up a bill. The gain is that creator, scene and copy become controllable and testable.
The dream is a closed loop where tagged variables link to Meta performance. He expects AI to be cheap and far more testable within half a year.
He does audience and visual research before generating anything. His process has five steps: copy research, visual research, copywriting, generation, editing. Visual research borrows the native look of the feed, and competitor ad libraries are a main source of formats.
Engaging with one audience's content for 5 minutes makes Reels about 80% that audience. The goal is the pattern, such as every third post looking alike, not one viral video.
Burke builds scenes as images first, then turns the best into video. He iterates in GPT Image 2 or Nano Banana until it looks authentic, then uses the image as reference in Seedance 2 or Veo 3. Images are cheaper to iterate on than prompting video directly.
Higgsfield gives access to many models under one subscription. Seedance 2 is probably trained on TikTok data, so it suits authentic UGC, and GPT Image 2 keeps text consistent where Nano Banana is weaker on fine print.
He automates proven formats with node workflows like Weave. Once a format works, he defines each scene's prompt and feeds in variables in Weave or Flora, then generates many iterations at once. Winning ads had 400 to 500 iterations.
Winning concepts need hundreds of iterations to stay alive and move into new placements.
4How Burke decides who his ads reach
Burke treats audience as message plus media plus content type, times funnel. Pain points like anxious attachment or overthinking pull different audiences purely through wording. A street interview style brings a younger Reels audience, and a raw text wall brings an older one.
He finds "creative equals targeting" a bit too simplified. Each variable is its own scaling lever: new messages, proven messages in new formats, or new funnels.
He uses web ads for problem-aware reach and app ads for solution-aware reach. An app funnel starts with a download, so the ad must sit close to the solution. A web funnel can run top-of-funnel content, like Nibble's geopolitics ad with no selling, then walk people to the app.
Static ads deliver poorly in app promotion campaigns, so he starts with video and tests statics later, especially when older Facebook feed users are the most lucrative audience.
Burke reads cost per trial by age against trial-to-paid conversion. He asks for age in onboarding so each age bucket's trial-to-paid rate can be tracked. A user 65 plus may cost about 100% more per trial but convert much better.
Young traffic gives cheap trials that convert worst, and an 18 to 24 user often can't pay about $70 a year. Younger does not mean worse, but the trial goal must be stricter.
He uses Meta value rules to push spend toward older users. Targeting stays broad (US, 18 plus) while value rules tell Meta a 45 plus user is, say, 30 to 40% more valuable. This works around trial-to-paid arriving about 7 days later, too late for the algorithm.
Low spend on older people often means no content attracts them, so he also makes ads for them.
5Burke's account structure for scaling spend
Burke pulls outlier ads into their own ad set. In one ad set Meta can put about 80% of spend on its favorite ad. Ads with good cost per target but 40 to 50% of spend on Facebook feed get starved when mixed with others.
He gives them their own budget with broad targeting and all placements, then checks whether the feed spend really bought an older audience. "you basically pull them out give them their own ads set and basically give them room to breathe"
He splits ad sets by format and message, with budgets per ad set. Static, video, text walls, memes and UGC get separate sets, and each message such as overthinking or anxious attachment gets its own. His example has five ad sets, each with its own budget and cost-per-target goal.
Memes bring lower-quality traffic, so they must deliver at about half the price. "I'm not looking for the very best ad and only like serve to that audience. I'm looking to scale."
With a tight budget, Burke starts with everything in one ad set. More ad sets means doubling, tripling or quadrupling budget. Bootstrapped teams put all ads in one set, see what wins and repeat weekly.
Splitting is much easier near break-even than while losing 30 to 40%, and VC-backed teams can split messages and formats upfront to learn faster.
He leans on cultural trends only once spend is large. At millions a month an account always needs extra profitable pockets, and trend language like ADHD or toxic attachments can supply them. Early on, the goal is copy that resonates, trend or not.
He sees these as multi-year trends, but calls self-diagnosis inflated by content "a bit of a dangerous place."
How to Build a $2M/Month App (Copy Me)
Marcus Burke explains how to build Meta ads for subscription apps: research the audience's own language on Reddit, prime people for the App Store, and structure ad sets so spend reaches distinct audiences. His best-performing ad concept has passed $1M in spend across hundreds of iterations, and he has scaled an app to $2M a month.
Burke learned paid social in mobile gaming at Innogames. He started at Innogames in Hamburg as games moved from browser to mobile, after training as an air traffic controller. Early mobile gaming was trial and error with few examples to copy, and Clash of Clans was everyone's reference point.
His head of marketing told him Facebook is dead when he joined, because performance was degrading.
Big spend bought Burke's team early access to Meta's products. Spending millions a month alongside King and Supercell got them invited to Meta headquarters events and into beta products first. Meta introduced app event optimization and value optimization during this period.
At Blinkist he pioneered sending ads to a blog, not the app store. He joined Blinkist in 2019 through a friend from Innogames, then worked at Study Smarter and Tandem.
He went freelance in 2022 because management work had become hiring, reporting and administration.
Burke says he has scaled one app to $2M a month. He has run Meta ads for apps for 14 years, and his best-performing concept, a sleep-sound ad for Moonate, has passed $1M in spend.
Both figures come from Burke alone, with no spend or revenue data shown.
Burke makes app ads say plainly that the product is an app. An ad that hides the app makes the App Store landing feel jarring and hurts install rate. His fix is to mention the app in the copy or show a demo.
"you want to prepare people for that they are going to land on the app store otherwise your install rate tends to suffer."
He refuses to copy D2C talking-head formats one to one. The ecom pattern of hook, pain points, then product is not the best format for apps. He forms a hypothesis about why an ad worked for its product before borrowing it.
Organic virality needs broad, non-salesy content, so he borrows only the native feel of a format, not the whole thing.
Burke writes in the audience's own pain-point language. Ad sets run broad, so Meta picks the audience from who engages. Specific phrasing like "rewire toxic attachments" draws the right engagers, while generic copy pulls a bigger audience that converts worse.
The best copy borrows words people use to label themselves, then plugs the product in as the fix. A line like "we made this app" also primes the App Store landing.
He feeds 10 to 20 Reddit threads to Claude for audience language. He asks Claude for sub-audiences, language patterns, pain points, desires and vivid emotional phrases. Reddit works because users write anonymously and bluntly, and big threads signal what matters.
App reviews, competitor reviews and comment sections are other written sources.
Claude gets one context file per sub-audience. Each file sits alongside the app's features, social proof and awards. Burke then asks for several distinct hooks, or feeds in a competitor's text wall as a template.
Without that context, an LLM falls back on generic internet marketing language. Claude writes strong copy but can't tell which formats convert. "The key is context."
Burke lets onboarding do the selling and keeps the ad soft. Many subscription apps have a long onboarding where the sale happens, so emotional, pain-point-led text with a light app mention is enough. His text-wall ads run about 10 seconds.
How much to demo depends on the category. A habit tracker like Me+ needs little explanation, while Picnic uses over-the-shoulder demos with feature captions.
The Moonate concept sells a scenario first and the app second. A five-second story hook about a husband who passed out on the couch leads into a slowly zooming glowing orb with a new line every couple of seconds. It had about 400 iterations.
Variants changed scenes and copy, and tinnitus-specific versions came from learning that many users play noise to mask tinnitus.
He interviews users weekly to learn how they describe the problem. At a past company, everyone on the UA team did one user interview a week, and much of their copy understanding came from it.
The questions must get people talking about the problem. He treats chasing social trends as secondary early on.
Burke uses AI to control creative variables, not to cut cost. His freelancers make about 99% of his clients' ads with AI, yet it is not cheaper yet because generations rack up a bill. The gain is that creator, scene and copy become controllable and testable.
The dream is a closed loop where tagged variables link to Meta performance. He expects AI to be cheap and far more testable within half a year.
He does audience and visual research before generating anything. His process has five steps: copy research, visual research, copywriting, generation, editing. Visual research borrows the native look of the feed, and competitor ad libraries are a main source of formats.
Engaging with one audience's content for 5 minutes makes Reels about 80% that audience. The goal is the pattern, such as every third post looking alike, not one viral video.
Burke builds scenes as images first, then turns the best into video. He iterates in GPT Image 2 or Nano Banana until it looks authentic, then uses the image as reference in Seedance 2 or Veo 3. Images are cheaper to iterate on than prompting video directly.
Higgsfield gives access to many models under one subscription. Seedance 2 is probably trained on TikTok data, so it suits authentic UGC, and GPT Image 2 keeps text consistent where Nano Banana is weaker on fine print.
He automates proven formats with node workflows like Weave. Once a format works, he defines each scene's prompt and feeds in variables in Weave or Flora, then generates many iterations at once. Winning ads had 400 to 500 iterations.
Winning concepts need hundreds of iterations to stay alive and move into new placements.
Burke treats audience as message plus media plus content type, times funnel. Pain points like anxious attachment or overthinking pull different audiences purely through wording. A street interview style brings a younger Reels audience, and a raw text wall brings an older one.
He finds "creative equals targeting" a bit too simplified. Each variable is its own scaling lever: new messages, proven messages in new formats, or new funnels.
He uses web ads for problem-aware reach and app ads for solution-aware reach. An app funnel starts with a download, so the ad must sit close to the solution. A web funnel can run top-of-funnel content, like Nibble's geopolitics ad with no selling, then walk people to the app.
Static ads deliver poorly in app promotion campaigns, so he starts with video and tests statics later, especially when older Facebook feed users are the most lucrative audience.
Burke reads cost per trial by age against trial-to-paid conversion. He asks for age in onboarding so each age bucket's trial-to-paid rate can be tracked. A user 65 plus may cost about 100% more per trial but convert much better.
Young traffic gives cheap trials that convert worst, and an 18 to 24 user often can't pay about $70 a year. Younger does not mean worse, but the trial goal must be stricter.
He uses Meta value rules to push spend toward older users. Targeting stays broad (US, 18 plus) while value rules tell Meta a 45 plus user is, say, 30 to 40% more valuable. This works around trial-to-paid arriving about 7 days later, too late for the algorithm.
Low spend on older people often means no content attracts them, so he also makes ads for them.
Burke pulls outlier ads into their own ad set. In one ad set Meta can put about 80% of spend on its favorite ad. Ads with good cost per target but 40 to 50% of spend on Facebook feed get starved when mixed with others.
He gives them their own budget with broad targeting and all placements, then checks whether the feed spend really bought an older audience. "you basically pull them out give them their own ads set and basically give them room to breathe"
He splits ad sets by format and message, with budgets per ad set. Static, video, text walls, memes and UGC get separate sets, and each message such as overthinking or anxious attachment gets its own. His example has five ad sets, each with its own budget and cost-per-target goal.
Memes bring lower-quality traffic, so they must deliver at about half the price. "I'm not looking for the very best ad and only like serve to that audience. I'm looking to scale."
With a tight budget, Burke starts with everything in one ad set. More ad sets means doubling, tripling or quadrupling budget. Bootstrapped teams put all ads in one set, see what wins and repeat weekly.
Splitting is much easier near break-even than while losing 30 to 40%, and VC-backed teams can split messages and formats upfront to learn faster.
He leans on cultural trends only once spend is large. At millions a month an account always needs extra profitable pockets, and trend language like ADHD or toxic attachments can supply them. Early on, the goal is copy that resonates, trend or not.
He sees these as multi-year trends, but calls self-diagnosis inflated by content "a bit of a dangerous place."