Deepsolv Blog
Deepsolv Blog
Backed by
Creative Decision Intelligence for Meta growth teams. Know what to test next — before you spend on it.
Follow Deepsolv on InstagramFollow Deepsolv on LinkedInFollow Deepsolv on TwitterAll rights reserved © 2026 deepsolv.ai
10 min read
Published: 1/27/2026
The question keeping performance marketers up at night in 2026 isn't whether AI can run Meta Ads- it's whether they're making the right creative decisions before spending a single dollar.
Meta's new Andromeda algorithm has changed everything. It's 100x faster at matching people to ads and can handle 10,000x more ad variants simultaneously. Translation? The brands winning on Meta in 2026 aren't the ones spending more, they're the ones feeding the algorithm smarter creative decisions, backed by data-driven intelligence.
This is where Claude Code enters the picture—not as another automation tool that blindly runs your campaigns, but as an agentic coding assistant that helps you make confident creative decisions before launch.
In this guide, you'll discover how to set up Claude Code for Meta Ads, automate your most time-consuming workflows, and build a creative research system that tells you what to test next—before you waste budget finding out the hard way.
Before we dive into setup, let's be clear about what Claude Code brings to the table.
Claude Code is NOT:
Claude Code IS:
Think of Claude Code as the analyst you wish you had, one that can pull campaign data at 3 AM, identify creative fatigue instantly, generate testing hypotheses based on competitor research, and create ad copy variations tailored to specific placements- all without waiting for your marketing team to wake up.
Here's the uncomfortable truth about Meta advertising in 2026: creative quality now accounts for up to 70% of campaign results, according to research on creative strategy automation.
As Meta fully automates targeting and bidding through Advantage+ and AI-powered delivery, the only lever left in your control is creative decision-making. The problem? Most teams are still making creative decisions based on:
A recent experiment highlighted in Medium showed what happens when you give Claude Code full autonomy over a Meta Ads account with a $1,500 budget. The results? The system managed creative production, ad creation via Meta API, budget controls, and analytics with compounding memory that made it genuinely autonomous.
But here's the key insight from that experiment: The value wasn't in automation- it was in systematic pattern discovery.
This is exactly what Deepsolv's Adam does at scale. While Claude Code handles the technical execution and file manipulation, Adam acts as your creative research layer—continuously analyzing your brand's ad performance data alongside competitor and market creatives to identify which ICP-angle-message combinations actually drive results.
Let's get you operational. Here's what you need:
Requirements:
Download and install Claude Desktop, then open the app and click the "Cowork" tab at the top. Create a dedicated workspace folder structure:
~/meta-ads-workspace/
├── exports/ ← Campaign performance exports from Ads Manager
├── creatives/ ← Ad copy, creative briefs, testing matrices
├── competitors/ ← Competitor ad research and patterns
├── reports/ ← Automated performance reports
└── uploads/ ← Files ready for Meta upload
This folder structure becomes Claude Code's working environment—the place where it reads campaign data, generates creative variations, builds testing matrices, and produces insights.
Skills are pre-built instruction sets that teach Claude Code how to handle specific marketing tasks with professional-grade quality.
The free marketing skills pack (available on GitHub at irinabuht12-oss/marketing-skills) includes several Meta-specific capabilities:
Installation steps:
You have two options for connecting Claude Code to Meta:
Option A: MCP Integration (Read-Only Analysis)
MCP (Model Context Protocol) lets Claude pull performance data directly from your Meta Ads account without manual CSV exports.
For the easiest setup, use a remote MCP server:
Now Claude can query your account with natural language: "Show me campaigns with ROAS below 2" or "Which ad sets have frequency above 3?"
Option B: Full Automation via Dedicated Platform
For teams wanting Claude to execute changes (not just analyze), platforms like Ryze AI provide direct account connection with full automation capabilities. This unlocks workflows like automatic creative refresh, budget reallocation, and campaign creation—all while maintaining human oversight.
The key difference: MCP gives you read access for analysis. Dedicated platforms give you write access for execution.
Now that you're set up, let's put Claude Code to work. These workflows replace hours of manual analysis with automated intelligence.
The Problem: You're spending $50-200/day on ads that stopped working three days ago, but you didn't catch it because you were focused on other campaigns.
The Claude Code Solution:
Drop your Meta Ads export (Ads → Export → All columns) into the /exports folder, then use this prompt:
I've placed a Meta Ads export in /exports.
TASK: Identify creative fatigue and recommend refreshes.
STEP 1 - FATIGUE SIGNALS:
For each ad, check:
- Frequency > 2.5 in last 7 days (people seeing ad too often)
- CTR dropped >20% vs. first 7 days of delivery
- CPM increased >30% vs. first 7 days
- Conversion rate declining week-over-week
STEP 2 - CATEGORIZE:
- CRITICAL: 3+ fatigue signals, still spending >$50/day
- WARNING: 2 fatigue signals, or 1 signal + high spend
- HEALTHY: 0-1 signals, stable performance
STEP 3 - PRIORITIZE REFRESHES:
Rank fatigued ads by:
1. Current daily spend (highest first)
2. Historical ROAS (prioritize proven performers worth saving)
3. Days since last creative refresh
OUTPUT:
1. fatigue-report.md
- Summary: X ads critical, Y warning, Z healthy
- Critical ads table: Ad name, spend, frequency, CTR trend, days active
- Recommended action for each (pause, refresh copy, refresh image, new angle)
2. refresh-priority.csv
Columns: Ad ID, Ad Name, Campaign, Fatigue Level, Daily Spend,
Historical ROAS, Recommended Action, Priority Score
Sorted by priority score descending
3. creative-brief.md
For top 5 fatigued ads worth saving:
- Current angle that worked
- 3 new angle suggestions
- Copy variations to test
What This Replaces: 2-3 hours of manual campaign review, spreadsheet analysis, and creative strategizing.
Real Brand Example: A DTC fitness brand used this workflow to catch creative fatigue on their top-performing ad (spending $180/day) on day 6 instead of day 12. By refreshing the creative angle early, they maintained a 3.2x ROAS instead of watching it decay to 1.8x while trying to force performance through budget increases.
The Problem: You're running three different campaigns targeting similar audiences, and they're competing against each other in the same auctions—driving up your CPMs and wasting budget.
The Claude Code Solution:
I've placed Meta Ads ad set performance data in /exports.
TASK: Identify audience overlap that's causing self-competition.
ANALYSIS:
1. GROUP AD SETS BY TARGETING:
- Extract targeting criteria (interests, lookalikes, custom audiences, age, gender, location)
- Flag ad sets with >50% estimated targeting overlap
- Identify ad sets competing for same auction
2. FIND OVERLAP SYMPTOMS:
- Ad sets with same/similar targeting but different campaigns
- High CPM ad sets that share audience with lower CPM ad sets
- Lookalike audiences at different % ranges (1%, 2%, 5%) running simultaneously
3. CALCULATE WASTE:
- Estimate budget being wasted on self-competition
- Identify which ad set should "win" based on performance
RULES FOR RECOMMENDATIONS:
- Don't overlap LAL 1% and LAL 2% in same campaign
- Separate prospecting and retargeting audiences
- One interest cluster per ad set, not stacked
OUTPUT:
1. overlap-analysis.md
- Overlap groups found
- Estimated wasted spend from self-competition
- Which ad sets to pause vs. consolidate
2. audience-restructure.csv
Columns: Current Ad Set, Issue, Recommendation, Action (Pause/Consolidate/Keep)
3. new-structure.md
- Recommended campaign structure
- Which audiences to separate
- Budget allocation by audience tier
Real Brand Example: An e-commerce furniture brand discovered they were running LAL 1%, LAL 2%, and LAL 5% audiences simultaneously across different campaigns—all competing for the same high-intent users. After consolidating based on Claude's analysis, their average CPM dropped 23% and their blended ROAS increased from 2.4x to 3.1x.
The Problem: You know which ads won, but you don't know why they won—which makes replicating success a guessing game.
The Claude Code Solution:
This workflow analyzes your top performers to extract actionable patterns:
I've placed Meta Ads ad-level data in /exports.
TASK: Identify what makes winning creatives work and create a testing playbook.
ANALYSIS:
1. DEFINE WINNERS:
- Ads with ROAS > [YOUR_TARGET] and spend > $500
- Ads with CTR in top 25% of account
- Ads that maintained performance for >14 days
2. PATTERN EXTRACTION:
For winners, analyze:
COPY PATTERNS:
- First line hook (question, stat, pain point, curiosity)
- Length (short <50 chars, medium 50-125, long 125+)
- CTA type (direct, soft, urgency)
- Emoji usage (none, light, heavy)
- Social proof inclusion (reviews, numbers, testimonials)
FORMAT PATTERNS:
- Placement performance (Feed vs. Stories vs. Reels)
- Image vs. video vs. carousel
- UGC vs. polished creative
3. LOSER ANALYSIS:
What do bottom 25% performers have in common?
What's different from winners?
OUTPUT:
1. creative-playbook.md
- Winning patterns summary
- Losing patterns to avoid
- Template: "[Hook type] + [Body length] + [CTA type] = Winner"
2. testing-matrix.csv
Columns: Test Name, Variable, Control Version, Test Version, Hypothesis
- Generate 10 specific A/B tests based on patterns found
- Each test isolates ONE variable
3. new-ad-concepts.md
- 5 new ad concepts applying winning patterns
- Primary text, headline, description for each
- Recommended placement
What This Unlocks: Instead of testing random creative variations, you're now testing hypotheses grounded in your actual performance data. This is how you move from creative guessing to creative intelligence.
The Problem: Your budget is distributed based on last month's performance or gut feeling, not current data.
The Claude Code Solution:
I've placed Meta Ads campaign and ad set data in /exports.
My total daily budget: $[YOUR_DAILY_BUDGET]
My target ROAS: [YOUR_TARGET_ROAS]
My target CPA: $[YOUR_TARGET_CPA]
TASK: Recommend budget reallocation to maximize ROAS.
ANALYSIS:
1. CURRENT STATE:
- Total spend by campaign
- ROAS by campaign and ad set
- Which campaigns are budget-limited (delivery issues)?
2. CATEGORIZE PERFORMANCE:
- SCALE: ROAS > target, not budget-limited, stable 7d trend
- OPTIMIZE: ROAS near target (±20%), potential with tweaks
- CUT: ROAS < 50% of target for >7 days, high spend
3. REALLOCATION RULES:
- Don't cut spend >30% in one day (learning phase reset)
- Don't increase >20% in one day (same reason)
- Prioritize proven campaigns over new tests
- Keep minimum viable budget for testing (10-15% of total)
OUTPUT:
1. reallocation-plan.md
- Current vs. recommended budget by campaign
- Expected ROAS improvement
- Timeline (Day 1, Day 3, Day 7 changes)
2. budget-changes.csv
Columns: Campaign, Ad Set, Current Daily Budget, New Daily Budget,
Change %, Reason, Implement Date
3. monitoring-checklist.md
- KPIs to watch after changes
- When to adjust further
- Red flags to pause immediately
Real Brand Example: A SaaS company with $3,000/day budget was splitting it evenly across 5 campaigns "to be fair." Claude's analysis revealed two campaigns were delivering 80% of their conversions at 40% below target CPA, while three campaigns were consistently 60% above target. Reallocating budget increased their monthly qualified demo bookings from 147 to 231—with the same total spend.
The Problem: You're using the same ad copy across Feed, Stories, and Reels—even though each placement has different user behavior, attention spans, and optimal formats.
The Claude Code Solution:
I've placed my product info and top-performing ad examples in /exports.
Product/Service: [DESCRIBE IN 1-2 SENTENCES]
Target audience: [WHO ARE THEY]
Main pain point we solve: [THE PROBLEM]
Key differentiator: [WHY US VS COMPETITORS]
TASK: Generate ad copy optimized for each Meta placement.
GENERATE FOR EACH PLACEMENT:
1. FEED (Facebook + Instagram):
- Primary text: 3 variations (short/medium/long)
- Headline: 5 variations (max 40 chars)
- Description: 3 variations (max 30 chars)
- Optimized for: scroll-stopping, detail-friendly
2. STORIES (Facebook + Instagram):
- Overlay text: 5 variations (max 20 chars - must be readable fast)
- CTA text: 3 variations
- Optimized for: vertical, fast consumption, swipe-up friendly
3. REELS:
- Hook text (first 3 seconds): 5 variations
- Caption: 3 variations (short, punchy)
- Optimized for: native feel, not "ad-like"
RULES:
- No generic CTAs ("Learn More", "Shop Now" only as last resort)
- Include specific numbers/outcomes where possible
- Match the tone of each placement (Feed=informative, Stories=urgent, Reels=casual)
- Vary the angles: pain point, benefit, social proof, curiosity, urgency
OUTPUT:
1. feed-copy.csv
Columns: Primary Text, Headline, Description, Angle, Character Count
2. stories-copy.csv
Columns: Overlay Text, CTA, Angle, Character Count
3. reels-copy.csv
Columns: Hook Text, Caption, Angle, Character Count
4. testing-plan.md
- Which variations to test first
- Recommended budget per test
- Success metrics by placement
Why This Matters in 2026: As noted in recent Meta advertising analysis, 90% of Meta inventory will be vertical in 2026, and ads not designed for 9×16 are leaving CPM efficiency on the table.
If you've followed along this far, you've seen how powerful Claude Code can be for executing Meta Ads workflows—pulling data, identifying patterns, generating variations, and producing reports.
But here's the question that determines whether you're just automating busy work or actually getting smarter about creative decisions:
What patterns should Claude Code be looking for in the first place?
This is where most teams hit a wall. They can automate the analysis, but they can't automate the intelligence about which creative patterns actually correlate with performance across different ICPs, industries, and market conditions.
This is precisely the gap that Deepsolv's Adam fills.
Think of the relationship this way:
Adam continuously analyzes:
Then Adam does something Claude Code alone cannot:
It maps creative patterns directly to performance outcomes and generates clear hypotheses about what will work before you test it.
For example, instead of just telling you "Ad #17 has creative fatigue," Adam tells you:
This is creative decision intelligence—and it's what separates teams that test randomly from teams that test strategically.
Here's how performance teams are combining these tools in 2026:
The result? You're not just automating Meta Ads workflows—you're building a compounding intelligence system that gets smarter with every campaign you run.
Before we wrap up, here are the lessons learned from teams already running this workflow:
Export frequently. Meta data changes fast—use exports from the last 7 days maximum. Older data leads to outdated insights.
Test one variable at a time. When Claude suggests multiple changes (new angle + new format + new targeting), implement them separately so you know what drove the improvement.
Watch frequency first. High frequency = dying creative. Refresh before it completely tanks your performance.
Separate prospecting from retargeting. Never mix them in the same campaign. They have different success criteria, different creative needs, and different optimization timelines.
Creative > Targeting in 2026. Put 80% of your optimization effort into creative decisions. Meta's algorithm handles targeting better than you can manually.
Build custom skills for your brand. Create a skill file with your brand voice guidelines, approved phrases, banned words, and performance thresholds. This ensures Claude Code's outputs match your standards every time.
Don't skip the human review. Claude Code should accelerate your decisions, not replace your judgment. Use it to eliminate blind spots and surface insights, then apply your strategic thinking to the final call.
As we've seen throughout this guide, Meta's advertising platform in 2026 is fundamentally different from even 18 months ago. The platform is moving toward Goal-Only campaigns where advertisers declare objectives and let AI handle optimization, and creative diversity and campaign simplicity are no longer best practices—they're survival tactics.
In this new environment, your competitive advantage doesn't come from manual campaign optimization or trying to outsmart Meta's algorithm with targeting tricks. It comes from making better creative decisions faster than your competitors.
Claude Code gives you the workflow automation and execution speed.
Deepsolv's Adam gives you the creative intelligence and pattern discovery.
Together, they transform Meta Ads from a platform where you're constantly reacting to performance changes into a system where you're proactively testing the right things, backed by data and competitive intelligence.
If you're spending ₹30L+ monthly on Meta Ads and struggling with creative decision confidence, wondering what angle to test next, which ICP to design for, or whether that performance drop is creative fatigue or market shift- you're exactly who Deepsolv's Adam was built for.
Book a free trial and discover how creative decision intelligence transforms your Meta Ads strategy from reactive optimization to proactive experimentation.
See how Adam identifies the ICP-angle-message combinations that drive results in your account, before you spend budget testing blind.
FAQs
Frequently asked questions related to this blog post
