Intro:
At Growth Hackers, we run 100s of experiments every month across Meta Ads, SEO, landing pages, creative, and automation workflows. Instead of letting those insights sit in internal dashboards and Notion docs, we’ve decided to make them public.
Welcome to the first edition of our new series: Marketing Experiments That Actually Work.
In each post, you’ll get:
- 4 field-tested experiments you can replicate this week
- 3 actionable insights that shift how you think about performance
- A breakdown of the tools and stacks we use to build, test, and grow faster
Let’s dive into what worked this month.
🧪 Experiments That Delivered
1. Cutting Wasted Spend in Display Campaigns by Blocking Bad Placements
Many brands unknowingly let their Google Display budget leak into junk placements — think random apps, toolbars, or niche clickbait sites.
We audited a Display campaign and found that over 30% of impressions were going to such low-intent sources. By using Google Ads’ placement reports, we excluded the bottom-tier performers.
Result: Conversions went up. CPA dropped. CTR became more stable.
How to try it:
- Google Ads → Display Campaign → Reports → “Where ads showed”
- Sort by Bounce Rate or CPA
- Exclude sites or apps with high spend but no conversions
Efficiency Gain: +15-25% in ROAS from cleaner placement targeting
2. Using Reddit for SEO and LLM Visibility
Reddit is now heavily favored in Google search results and is one of the most cited sources by LLMs.
We tested keyword-rich answers on niche subreddits, using branded profiles and real engagement (no links initially). Within days, our responses began showing up in SERPs — and we suspect they’re feeding into LLM training too.
How to replicate:
- Find subreddits where your audience hangs out
- Post authentic, value-driven responses
- Use primary and long-tail keywords
- Monitor visibility in Google via incognito and Ahrefs
Bonus: Improves trust, builds community, and likely earns LLM mentions
3. Reels in Local Languages = 3x-5x Organic Reach
We worked with an SME client and created Instagram Reels in local dialects instead of formal Hindi or English.
Three of those reels went viral.
Audience connected faster, shared more, and even competitors started mimicking the format.
Try this:
- Create Reels in regional language (not just subtitles)
- Focus on relatability, emotion, and humor
- Prioritize cultural cues over polish
Lesson: Relatability > Grammar. Emotion = Virality.
4. Broad Match Campaigns: Use With Caution
Google continues to push Broad Match, but it can destroy efficiency if left unchecked.
In our internal tests:
- Broad Match worked best only when paired with exact/phrase match
- Negative keyword sets were essential
- Monitoring for conversions (not just clicks) was non-negotiable
If you’re experimenting with Broad Match:
- Never use it alone
- Set strict performance thresholds
- Build out negatives weekly
Framework: Explore with Broad. Scale with Precision.
🔍 Strategic Insights from the Field
Insight #1: Negative Keywords in PMax Are More Critical Than Ever
PMax pulls from search inventory, which means your ads might show for irrelevant or low-converting queries.
We’ve seen major wins just by regularly auditing Search Term Insights and excluding wasteful queries.
Outcome: Cleaner traffic, lower cost, more control
Insight #2: Bidding on Your Own Brand Name Is No Longer Optional
AI overviews are occupying top SERP real estate.
Competitors can easily outbid you and push your organic results down even further. We’ve seen this firsthand.
Solution: Bid on your brand terms. Control your message. Own the shelf space.
Insight #3: AI Image Quality Depends Entirely on Prompt Structure
Generic prompts = generic output.
We’ve been using structured JSON-style prompts to guide AI tools like Midjourney or DALL·E. By defining attributes like color, lighting, composition, and emotion — the results are drastically better.
Rule: Treat it like a design brief, not a keyword search.
🛠️ Behind the Stack: Tools That Power Our Workflow
AI-First SEO Monitoring (Ahrefs + Semrush)
We’re tracking not just keyword rankings, but:
- Visibility in AI Overviews
- Citation frequency in AI answers
- Ranking shifts in AI-enabled SERPs vs traditional SERPs
We’ve built dashboards to measure:
- Share of voice
- Impression loss due to AI boxes
- Total traffic change post-AI rollout
Takeaway: SEO is now AI Optimization. Adapt accordingly.
AI-Powered Creative Production
We’re using Freepik’s AI stack to:
- Generate ad-ready visuals in minutes
- Remove backgrounds and apply brand styling
- Build visual consistency across campaigns
Bonus: We’re also testing tools like Nano Banana for hyper-realistic image generation (think Photoshop-level visuals at AI speed).
Creative Edge: Studio-quality output without studio cost
Video Content Workflow
Every video in this series is:
- Scripted using Claude
- Narrated with ElevenLabs
- Shot via HeyGen
- Edited in CapCut & Premiere Pro
This hybrid setup gives us speed, control, and polish — without a production crew.
Bonus Insight: Repurpose your videos into blogs, reels, and ads.
Dev + Automation Stack
Internally, our tools and analytics systems are built in VS Code using Python, FastAPI, and AI libraries.
We test locally, deploy smartly, and use Claude to validate pipelines, debug, and optimize.
Philosophy: Don’t just automate — audit and accelerate.
Bonus Insight: Recent AI Advancements in Visual Generation
AI image generation has reached a pivotal moment, with new models achieving near-photographic quality that’s reshaping creative workflows.
Tools like Nano Banana now enable hyper-realistic image generation and editing that rivals traditional software like Photoshop.
What this means for creators:
- Generate campaign-ready hero images and product shots without expensive studio setups
- Test multiple creative variations in minutes instead of days
- Build moodboards and experimental concepts with unprecedented realism
The gap between AI-generated and photographed content is rapidly closing, opening new possibilities for rapid prototyping and creative exploration.
📆 What’s Next
This is just the first drop.
We’ll send out one edition every month — concise, field-tested, and 100% experiment-backed.
If you liked what you read:
