What It Is
Kimi K2.5 is an AI model from Moonshot AI. Released January 27, 2026.
The short version: it handles large files that crash ChatGPT and Claude.
Free tier. Open source. 256K context window.
For media buyers, that means you can upload your full Meta or Google Ads export and actually get analysis back.
Why This Matters for Ad Analysis
ChatGPT and Claude have a problem with big CSV files.
You export 30 days of ad data from Meta. 4,000 rows. Upload it. ChatGPT either cuts it off, summarizes poorly, or just crashes.
Kimi takes the whole file.
The 256K context window means it can hold roughly 200,000 words in memory at once. Your typical Meta export is nowhere near that limit.
What Kimi K2.5 Actually Is
1 trillion parameters. Mixture-of-experts architecture (only activates 32B parameters per query, so it runs fast despite the size).
Native multimodal — trained on text and images together from the start. Can read screenshots, charts, PDFs.
Four modes:
Instant: Fast responses, good for simple questions
Thinking: Deeper reasoning, better for analysis
Agent: Can use tools, browse, execute tasks
Agent Swarm: Spins up multiple sub-agents for complex tasks
For ad analysis, you'll mostly use Thinking mode.
How to Access It
Free tier gives you enough for regular use. Paid plans for heavy usage.
Kimi vs ChatGPT vs Claude for Ad Data
ChatGPT (GPT-4o)
- —128K context window (but often chokes on large uploads)
- —Good at summarizing, weaker on detailed row-by-row analysis
- —Paid required for large files
Claude (3.5 Sonnet)
- —200K context window
- —Better with structured data than GPT
- —Still struggles with very large exports
- —Paid required for extended context
Kimi K2.5
- —256K context window
- —Handles raw CSV exports without preprocessing
- —Free tier works for most ad analysis tasks
- —Thinking mode gives detailed, row-level analysis
For a typical media buyer doing weekly audits: Kimi handles files that break the others.
What It's Good At
Data analysis: Upload your export, ask specific questions, get specific answers with campaign names and numbers.
Pattern recognition: Finding which ads fatigued, which audiences saturated, which campaigns are quietly wasting money.
Comparisons: Cross-platform analysis if you upload both Google and Meta exports.
Calculations: Budget reallocation math, CPA projections, efficiency rankings.
What It's Not Good At
Real-time data: It can't connect to your ad accounts. You need to export and upload.
Execution: It analyzes and recommends. You still make changes in Ads Manager.
Context it doesn't have: Your margins, upcoming promos, inventory issues, seasonality. Tell it or it can't factor it in.
Platform changes: If Meta changes attribution windows or reporting tomorrow, Kimi doesn't know until you tell it.
How to Use These Prompts
Step 1: Export your data from Meta Ads Manager (or Google Ads)
Select last 14-30 days, all columns, download as CSV. The more data, the better the patterns Kimi can find.
Step 2: Go to kimi.ai
Free tier works for most exports. No preprocessing needed.
Step 3: Select "Thinking" mode
Click the mode selector and choose "Thinking" for deeper analysis. Takes longer but gives you row-level detail and catches edge cases.
Step 4: Upload the CSV
Drag and drop or click to upload. Kimi handles the parsing automatically.
Step 5: Copy-paste the relevant prompt
Replace all [X] placeholders with your actual numbers. These include your target CPA, target ROAS, monthly budget, and minimum spend thresholds.
Step 6: Review the analysis
Kimi returns structured output with tables, specific campaign names, and actionable recommendations. Cross-check anything that looks off.
Step 7: Ask follow-up questions
Dig deeper on anything unclear. The prompts below include suggested follow-ups.
About the [X] placeholders
Every prompt uses $[X] or [X] for values you need to customize. Don't skip these — they calibrate what "good" and "bad" performance looks like for your specific account.
When to use each prompt
- Weekly Audit:Every Monday. Your starting point for pause/scale decisions.
- Creative Deep Dive:When performance drops, before scaling, or when planning creative refresh.
- Audience Analysis:When you see saturation signals (rising CPMs, frequency above 3).
- Budget Optimization:End of month, budget cuts, or when reallocating across campaigns.
- Cross-Platform:Monthly review of Google vs Meta allocation decisions.
- Account Takeover:First week with a new client or inherited account.
Detailed Prompts for Meta Ads Analysis
Six production-ready prompts covering weekly audits, creative analysis, audience targeting, budget optimization, cross-platform comparison, and account takeovers. Copy, customize the [X] values, and run.
Full Weekly Audit
Your Monday morning starting point. Identifies what to pause, scale, watch, and flags fatigue and anomalies across the entire account.
# Copy this prompt
You are a senior media buyer analyzing Meta Ads performance data.
I'm uploading a CSV export from Meta Ads Manager covering the last 14-30 days.
My targets:
- Target CPA: $[X]
- Target ROAS: [X]
- Monthly budget: $[X]
Analyze this data and give me:
1. PAUSE LIST
Campaigns or ad sets spending money with CPA more than 1.5x my target or ROAS below 50% of target.
For each one show: name, spend, conversions, CPA, ROAS, and how much money it wasted vs if it hit target.
Sort by wasted spend descending.
2. SCALE LIST
Campaigns or ad sets with CPA under target and ROAS above target that aren't budget-capped.
For each one show: name, current spend, current CPA, current ROAS, and recommended new daily budget with reasoning.
Only recommend scaling if there's evidence of headroom (not maxed out, stable or improving metrics).
3. WATCH LIST
Anything borderline - within 20% of targets either direction.
For each one show: name, metrics, and what signal would push it to pause or scale.
4. FATIGUE ALERTS
Any ad or ad set where frequency is above 3 AND either CTR dropped week-over-week or CPM increased week-over-week.
Show: name, frequency, CTR trend, CPM trend.
5. ANOMALIES
Anything unusual - sudden spikes or drops in CPM, CTR, conversion rate, spend.
Show: what changed, when, possible cause.
6. SUMMARY
Total spend this period.
Total wasted spend (on campaigns above target).
Top 3 actions I should take this week, prioritized by impact.
Use tables where helpful. Be specific with campaign and ad set names. Show your math.
What you'll get
- Prioritized pause list with wasted spend calculations
- Scale recommendations with headroom analysis
- Early warning on borderline campaigns
- Fatigue detection before performance tanks
Creative Performance Deep Dive
Ad-level analysis for creative strategists. Identifies top performers, diagnoses failures, detects fatigue, and compares format effectiveness.
# Copy this prompt
You are a creative strategist analyzing ad performance data.
I'm uploading Meta Ads data. Focus on ad-level performance.
Analyze the creatives and tell me:
1. TOP PERFORMERS
Ads with CPA under $[X] and at least $[X] spend (enough data to trust).
For each one:
- Ad name
- Spend, conversions, CPA, CTR, CPM
- What's working: Is it the hook, the format, the offer, the audience? Look at which placements and audiences it performs best in.
2. UNDERPERFORMERS
Ads with CPA above $[X] and at least $[X] spend.
For each one:
- Ad name
- Spend, conversions, CPA, CTR, CPM
- Why it's failing: Low CTR (creative problem), high CTR but low conversion (landing page or offer problem), high CPM (audience saturation or competition).
3. FATIGUE DETECTION
Ads showing signs of creative fatigue:
- Frequency above 3
- CTR declining over time
- CPM increasing over time
For each one: name, current frequency, CTR week 1 vs week 2+, CPM week 1 vs week 2+, recommendation (kill, reduce budget, refresh).
4. FORMAT ANALYSIS
Compare performance by ad format (single image, video, carousel, etc).
Which format has best CPA? Best CTR? Best scale potential?
Show averages and sample sizes.
5. HOOK ANALYSIS
If ad names indicate different hooks or angles, compare their performance.
Which messaging themes are working? Which aren't?
6. RECOMMENDATIONS
Based on this analysis:
- Which 3 ads should I turn off today?
- Which 3 ads should I scale?
- What new creative angles should I test based on what's working?
What you'll get
- Diagnosis of why creatives succeed or fail
- Fatigue detection with specific recommendations
- Format and hook effectiveness analysis
- Concrete next steps for creative testing
Audience Analysis and Targeting
Ad set level analysis for targeting optimization. Ranks audience efficiency, detects saturation, identifies overlap, and suggests expansion opportunities.
# Copy this prompt
You are a media buyer specializing in audience strategy.
I'm uploading Meta Ads data. Focus on ad set level performance to analyze audiences.
My targets:
- Target CPA: $[X]
- Target frequency cap: 3
Analyze:
1. AUDIENCE EFFICIENCY RANKING
Rank all audiences by CPA, best to worst.
Show: audience name, spend, conversions, CPA, ROAS, CTR, CPM, frequency.
Flag any audience spending more than $[X] with CPA above target.
2. SATURATION SIGNALS
Find audiences showing saturation:
- Frequency above 3
- CPM increasing week over week
- CTR declining week over week
- Conversion rate declining while spend is flat or up
For each one: name, metrics, severity (mild/moderate/severe), recommendation.
3. OVERLAP ANALYSIS
If audience names suggest potential overlap (e.g., multiple lookalikes, interest stacks with similar themes), flag them.
Show which audiences might be competing for the same users and cannibalizing each other.
4. SCALING OPPORTUNITIES
Audiences with strong performance (CPA under target) that aren't maxed out.
For each one: name, current daily budget, recommended new budget, expected incremental conversions.
5. AUDIENCE GAPS
Based on what's working, suggest:
- Lookalike variations to test (different seed audiences, different percentages)
- Interest combinations that might work based on top performer patterns
- Exclusions to add to prevent overlap
6. BUDGET REALLOCATION
If I kept total budget the same but reallocated optimally across audiences, show me:
- Current allocation (table with audience name and daily budget)
- Recommended allocation (table with audience name and new daily budget)
- Expected impact on overall CPA
What you'll get
- Efficiency ranking across all audiences
- Early saturation warnings before performance drops
- Overlap detection to prevent self-competition
- Budget reallocation recommendations with expected impact
Budget Optimization and Forecasting
Budget efficiency analysis with scenario planning. Breaks down spend efficiency, models cut/scale scenarios, and provides pacing checks.
# Copy this prompt
You are a performance marketing analyst focused on budget efficiency.
I'm uploading Meta Ads data.
Current situation:
- Total monthly budget: $[X]
- Target CPA: $[X]
- Target ROAS: [X]
Analyze and tell me:
1. CURRENT STATE
- Total spend this period
- Total conversions
- Blended CPA
- Blended ROAS
- How much of spend went to campaigns above target CPA?
2. EFFICIENCY ANALYSIS
Break down spend into buckets:
- Efficient spend (CPA under target): $X, X conversions
- Borderline spend (CPA within 20% of target): $X, X conversions
- Wasted spend (CPA more than 20% above target): $X, X conversions
Show percentage of budget in each bucket.
3. CUT SCENARIO
If I had to cut budget by 20%, what would I cut?
Show specific campaigns/ad sets to reduce or pause.
Calculate new expected blended CPA after cuts.
4. SCALE SCENARIO
If I had 20% more budget, where would I put it?
Show specific campaigns/ad sets to increase.
Calculate expected incremental conversions and CPA at higher spend.
5. REALLOCATION SCENARIO
If I kept budget the same but moved money from worst to best performers:
- What moves would I make? (specific amounts from X to Y)
- What would new blended CPA be?
- How many more conversions would I expect?
6. PACING CHECK
Based on current daily spend rate:
- Am I on track to spend full monthly budget?
- Am I overspending or underspending?
- If I'm going to overspend, what should I cut?
- If I'm going to underspend, what should I scale?
7. RECOMMENDATIONS
Top 3 budget actions for this week, ranked by expected impact on CPA.
What you'll get
- Efficiency breakdown by spend category
- Cut/scale scenario modeling with CPA projections
- Reallocation recommendations with expected impact
- Pacing analysis against monthly budget
Cross-Platform Comparison (Google + Meta)
Side-by-side analysis of Google and Meta performance. Compares efficiency, identifies platform-specific issues, and recommends budget allocation.
# Copy this prompt
You are a performance marketing strategist comparing Google and Meta Ads.
I'm uploading two CSV files:
- Meta Ads export
- Google Ads export
Both cover the same time period.
My targets:
- Target CPA: $[X]
- Target ROAS: [X]
- Total budget across platforms: $[X]
Analyze:
1. PLATFORM SUMMARY
For each platform show:
- Total spend
- Total conversions
- CPA
- ROAS
- CTR
- CPM (Meta) / CPC (Google)
2. EFFICIENCY COMPARISON
Which platform is more efficient right now?
- By CPA
- By ROAS
- By volume (which is driving more conversions)
- By cost trend (which is getting more expensive over time)
3. CAMPAIGN TYPE COMPARISON
Compare similar campaign types across platforms:
- Prospecting/cold traffic
- Retargeting
- Brand
Which platform wins for each type?
4. BUDGET ALLOCATION ANALYSIS
Current split: X% Meta, Y% Google
Is this optimal based on performance?
Recommended split with reasoning.
5. PLATFORM-SPECIFIC ISSUES
Meta: Any audience saturation, creative fatigue, frequency issues?
Google: Any keyword waste, search term issues, quality score problems?
6. CHANNEL STRATEGY
Based on this data:
- Should I shift budget between platforms? How much?
- Which platform should I test scaling first?
- Are there campaign types I should add or cut on either platform?
7. 30-DAY PLAN
Specific actions for each platform over next 30 days, prioritized by expected impact.
What you'll get
- Side-by-side platform performance comparison
- Campaign type effectiveness by platform
- Optimal budget allocation recommendation
- Platform-specific issue identification
New Account Takeover Audit
Full diagnostic for inherited or new client accounts. Assesses account health, identifies quick wins, and creates a 30-60-90 day action plan.
# Copy this prompt
You are a senior media buyer taking over a new client account.
I'm uploading their Meta Ads data. I need a full diagnostic to understand what's happening and what to fix first.
Run a complete audit:
1. ACCOUNT HEALTH SCORE
Rate the account 1-10 based on:
- Efficiency (what % of spend is at or below target CPA)
- Structure (are campaigns organized logically)
- Creative diversity (how many active creatives, variety of formats)
- Audience health (frequency levels, saturation signals)
Show your reasoning for the score.
2. QUICK WINS
Things I can fix in the first 48 hours that will have immediate impact.
Be specific: "Pause Campaign X (wasted $Y this month at $Z CPA)"
3. STRUCTURAL ISSUES
Problems with how campaigns/ad sets are organized.
- Too many campaigns?
- Overlapping audiences?
- Budget spread too thin?
- Missing campaign types (no prospecting, no retargeting, etc)?
4. CREATIVE ISSUES
- How many active ads?
- Creative diversity score (formats, hooks, angles)
- Any obvious fatigue?
- What's the best performing creative and why?
- What creative tests should I run?
5. AUDIENCE ISSUES
- How many audiences?
- Overlap problems?
- Saturation problems?
- Missing audience types?
- What audiences should I test?
6. BUDGET ISSUES
- Is budget allocated efficiently?
- Are good campaigns budget-capped?
- Are bad campaigns still spending?
- Recommended reallocation.
7. TRACKING ISSUES
Any red flags in the data suggesting tracking problems:
- Conversion delays
- Missing data
- Unusual patterns
8. 30-60-90 DAY PLAN
First 30 days: What to fix immediately
Days 30-60: What to optimize
Days 60-90: What to scale/test
Prioritize by impact. Be specific with campaign names and numbers.
What you'll get
- Overall account health assessment with score
- Quick wins you can implement immediately
- Structural, creative, audience, and budget issue identification
- Prioritized 30-60-90 day action plan
Follow-Up Questions to Dig Deeper
After Kimi gives you analysis, use these follow-ups to get more actionable detail. The first response is rarely the final answer — interrogate the recommendations.
"Why do you think Campaign X is underperforming? What specific metrics led to that conclusion?"
Use when: You want to understand the reasoning behind a recommendation before acting on it.
"If I only had time for one action today, which would have the biggest impact?"
Use when: You're short on time and need to prioritize ruthlessly.
"What would happen if I cut Campaign X budget by 50% instead of pausing completely?"
Use when: You want to test a softer intervention before killing a campaign entirely.
"Are there any patterns in the top performers I should replicate in new campaigns?"
Use when: You're planning new campaigns and want to build on what's working.
"What data am I missing that would help you give better recommendations?"
Use when: You suspect the analysis is limited by incomplete data export.
"Walk me through the math on the reallocation recommendation."
Use when: You want to verify the calculations before presenting to a client or stakeholder.
"What's the risk if I scale Campaign X too fast? What warning signs should I watch for?"
Use when: You're about to increase budget significantly and want to set monitoring criteria.
Pro tip: Challenge the output
Kimi can be wrong. If a recommendation seems off based on your experience, push back. Ask "What if the conversion data is delayed by 2 days?" or "Does this account for the fact that we launched a sale last Tuesday?" The model responds well to constraints and context you add in follow-ups.
Limitations to Know
It's not magic. Kimi analyzes the data you give it. If your export is missing columns or limited timeframes, the analysis will be limited.
It doesn't know your business. Margins, LTV, cash flow, competitive landscape — you have to factor these in yourself.
It can be wrong. Double-check recommendations that seem off. Ask it to explain its reasoning.
The free tier has limits. Heavy users will eventually need to pay.
Bottom Line
Kimi K2.5 handles the file sizes that break other models. For media buyers doing regular audits, that's the main thing.
Free. Works with raw exports. Returns specific recommendations with actual numbers.
Not a replacement for judgment. A tool that makes the analysis part faster.
Need Help?
Email hello@get-ryze.ai for setup assistance or custom prompts.
