If you’re running Facebook / Instagram ads in 2026 and they’re not working for you, you only need to know 3 things to improve: make it real, test vigorously, make it scale. This isn’t me guessing: this is based on analyzing just over $100M in ad spend in the last 9 months.
Make it real: Quality = native mimicry,
The winning ad doesn’t look like an ad — it looks like organic content the target person already consumes. The rules:
- “The middle ground is where ads go to die.” Ads win only if they’re
- (a) genuinely raw/native (“ugly ads,” “yappers”) or
- (b) obviously an ad but carried by such strong direct-response copy nobody cares. The dressed-up-UGC middle gets scrolled past.
- Match the format to the consumption context — film where the product is really used, and visually telegraph the target audience in the first frame (props/setting) to earn “relevant attention” before a word is spoken.
- Script only the hook; freestyle the body. Over-scripting is what makes UGC read as fake. The irreplaceable, AI-proof asset is a talented human “yapper.”
- Copy is the one constant across every winner — and it’s mined verbatim from real customer language (reviews, comments, Reddit), so it’s “a customer talking to a customer,” not a brand talking at one.
Test Vigorously: Quantity = volume beats perfection.
- “Volume negates luck.”
- You don’t make each ad 10% better; you ship 3–10x more and let the market tell you what works. More ads = more shots at goal = faster learning. When you 10x creative volume you’ll likely 5x revenue ($10M→$50M) in 18 months with ads that weren’t objectively better — they just iterated faster.
- The enabling constraint isn’t talent, it’s creative operations: a tracked pipeline (master DB + filtered views, concept-level tracking, days-to-launch metric, “tie the loop between performance and strategy”).
- Diversity = many angles/sources, not one house style.
- Many streams feed one river: internal team, creators, AI, customer content are separate tributaries, each with its own style, so you hit different pockets of the audience.
- If every ad looks the same, you’ve collapsed diversity to one — hence deliberately using different brief templates and multiple formats (yappers, avatar-specific, AI animation, negative framing/exposé, challenge ads, post-it, apology).
Make it scale: AI creates diversity with the right structure
- AI as a volume-and-context engine, not a magic button. You should think of your role with AI as “context engineering” — LLMs are great reasoners but terrible default copywriters, so you build reusable, purpose-trained projects
- A minimal version of this is just 4 things:
- brand doc + subject domain doc
- top-performer examples
- system prompt
- success definition for creative
- One-shot prompting is for suckers. AI lets you scale quantity and diversity cheaply (AI animations, unlimited image ads, headline generation) while humans hold the line on quality (the yapper, the strategist’s copy sense).
The AI Ads Operating System biggest leverage, most portable:
Whether you’re B2B or B2C, the playbook is the same, and AI is yoru engine for testing the playbook.
- Stand up a master creative database with filtered views (Notion — we already live there). Track at concept level, not ad level, with source, angle, format, status, launch date. We have the backlog/Notion muscle already; this is the same pattern pointed at ad creative.
- Instrument days-to-launch and output-by-source/angle as first-class metrics. His single most useful diagnostic: the gap between asset delivery and asset going live — every day of delay is a day without learning.
- Build a Creative Learnings log (what worked/didn’t, weekly, per ad) so onboarding a client or teammate is “read this,” not a two-hour call. This is a natural fit with our Obsidian/vault practice.
What Quality Means in 2026
- Native/”ugly” ads win over polished wherever the audience is on organic feeds. For founder-led brands this is a gift — the founder ugly ad is exactly what the Content Engine we teach and build around creates:: a credible person talking naturally to camera in a native setting, hook scripted, rest freestyled.
- It’s almost like I’ve been giving you good advice for years. 😛
- Make customer-language mining a standard research step: pull verbatim phrases from reviews, Reddit, LinkedIn comments, sales-call transcripts (we have MeetGeek transcripts!) and write copy in those words. “Customer talking to a customer.”
- Kill the middle ground in our own outputs — either genuinely raw, or obviously-an-ad-with-killer-copy. No half-produced UGC.
Use AI for quantity + diversity, keep humans on quality:
- Build context-engineered AI projects. I love using Hermes or OpenClaw + Obsidian for this. All of the context you need is in one place: a brand doc, a domain doc, and a top-performer examples doc, plus a system prompt. One project per job: headline generator, script writer, script critic. This is very close to how we already structure skills/routers in the vault.
- Lean into AI animation and AI image ads as cheap diversity tributaries — we already have the Nano Banana / Remotion / kinetic-typography stack, so we’re unusually well-positioned to produce these at volume.
Stop trying to make the perfect ad. Build a machine that ships many native, customer-worded ads across diverse angles, measure days-to-launch and what’s winning, and iterate weekly or, better, daily.
