Blog

Benchmark Reports on Ad Creative Automation

Things we measured while building ad creative automation — numbers and methodology included.

Why Advertise on Multiple Channels — How to Pick Media and Fix the Creative Bottleneck

Why Advertise on Multiple Channels — How to Pick Media and Fix the Creative Bottleneck

Ad channels feed each other — cut Meta and search CPA climbs (+19% incremental branded search in Brainlabs' lift studies). Six criteria for picking channels, the 70-20-10 budget frame, and how Expedia and Nike solved the creative versioning bottleneck that channel expansion creates.

2026-07-03
Big edit (budget 20%+ · targeting · creative swap · bid) = learning resetLearning — until 50 conv/weekOptimized · stable delivery
An ad set must gather ~50 conversions a week to exit learning. A budget change over 20%, or a targeting/creative/bid change, sends that counter back to zero.

The Meta Learning Phase: How to Exit Fast (50-Conversion Rule & Learning Limited Fix)

Meta's learning phase needs ~50 conversions per ad set per week to exit. Here's exactly what edits reset it, how long it takes, how to fix "Learning Limited," and how to refresh creative without panicking the algorithm.

2026-06-18
6-10+ creativesASC — auto targeting/placement/budgetauto delivery & audiences
ASC automates targeting, placement and budget. So the lever you control shifts to creative volume and diversity (6-10+).

Meta Advantage+ Shopping (ASC): When to Use It and How to Set It Up

When Meta's Advantage+ Shopping Campaign (ASC) beats manual setup, the catalog/Pixel/CAPI prerequisites, how many creatives it wants (6-10+), and how the existing-customer budget cap works now.

2026-06-18
Purchase (action)Pixel (browser)CAPI (server)Meta — dedupe by event_id (1)
A browser Pixel alone leaks 30-40% of conversions. Send Pixel + server CAPI together; a shared event_id lets Meta dedupe into one.

Meta Pixel + Conversions API (CAPI) Setup: A Checklist for When Conversions Won't Track

Pixel alone misses 30-40% of conversions post-iOS 14. Here's the order to check: run Pixel + Conversions API together, set up event_id deduplication, and raise Event Match Quality (EMQ). Plus why marketplace pages only capture clicks.

2026-06-18
Disabled · rejectedFind cause · auditRequest Review (once)Recover · prevent
Not more ads or repeated appeals — identify the reason, audit the account, submit one accurate Request Review, then recover and prevent.

Meta Ad Account Disabled or Ad Rejected — Causes and Recovery Steps (2026)

Why Meta (Facebook/Instagram) ad accounts get disabled or restricted and ads get rejected, plus the exact Request Review / appeal steps to recover and how to prevent it — based on Meta's official policies.

2026-06-18
No labelmisleading · risky"Virtual / AI" labelnear person · up front
If a virtual person appears, label it so consumers can tell (mandatory in Korea from Jun 2026). Even labeled, a first-person testimonial can be deceptive.

AI-Generated Model Ads: Disclosure Rules and How to Use Them Well (2026)

Running AI-generated virtual-model creative on Meta and unsure about disclosure? A practical guide to Meta's AI labeling, FTC guidance, Korea's new virtual-person labeling rule, and how to use AI models responsibly.

2026-06-18
1xsoft on dense screens2x (retina)crisp · compresses well
Build at 2x (retina) of display size to stay crisp on dense screens; 2x also compresses well. Reserve 4K for large/full-screen.

Ad Banner Resolution and Quality: How High Should You Go? (2026)

Deliver ad banners at 2x (retina) of display size as a baseline. Here is when 2K/4K matters, per-platform file-size limits, and how to compress without losing crispness.

2026-06-18
100 creatives9:161:116:9launch together · budget per concept
Meaningfully different creatives learn faster launched together. Group by ratio/platform/concept and split budget per group.

How to Operate 100 Ad Creatives: Launch All at Once or Stagger? A Bulk Setup & Management Workflow

You produced a pile of creatives — now do you launch them all at once or roll them out slowly, and how do you split the budget? Here's the operations workflow for running 100+ creatives: simultaneous vs staggered launch, grouping by ratio/platform, naming, budget split, and test structure.

2026-06-18
CBO — algorithm distributesCampaign budgetSet A · 60B · 30C · 10ABO — fixed & equal per set33Set A33Set B33Set C
CBO lets the algorithm pour one campaign budget into the winning ad set (for scaling). ABO fixes equal budgets per ad set for a clean comparison (for testing).

CBO vs ABO: Should Your Meta Budget Live at the Campaign or Ad Set Level?

A clear answer to whether you should add new creative to the same ad set or a new one, and when CBO beats ABO. Test with ABO, scale with CBO — here's why, with an actionable checklist.

2026-06-18
Broad / Advantage+default with enough dataLookalikemid-funnel bridgeRetargeting~3x conversion · safety net
With enough data, start broad (Advantage+) and let the algorithm work. Lookalikes are the mid-funnel bridge; retargeting is the ~3x-converting safety net. The broader you go, the more creative IS the targeting.

Meta Targeting in 2026: Broad, Detailed, Lookalike & Retargeting — When to Use Each

Confused about how to set Meta audiences? Here's the difference between broad (Advantage+), detailed targeting, lookalikes, and retargeting — when each wins in 2025-2026, and why creative now does the targeting once your audience goes broad.

2026-06-18
First-2s hookstop the scroll · payoffProblemOne benefitCTA (action verb)
The hook promises, the body proves, the CTA converts. Core message in the first 2 seconds, one benefit, one CTA.

How to Write Ad Hooks & Copy by Principle, Not Gut (2026)

Stop writing ad copy on instinct. A clear framework — first-2-second hook, problem to solution, one benefit, one CTA, and platform differences — plus a checklist to decide your ad copy by principle.

2026-06-18
Safe (<2.0)Warning · prep refresh (2–3)Danger · swap now (3+)Refresh pointFrequency ↑ → CTR ↓ · CPM ↑
As frequency rises, CTR falls and CPM climbs. Prep the next creative at 2.0–2.5 and swap before it crosses 3.0.

When to Refresh a Winning Ad: Creative Fatigue Signals and Refresh Cadence

Frequency 2.0-2.5 warnings, CTR decay, CPM rise — exactly when, how often, and how many variants to refresh your winning ad creative, with 2025-2026 benchmarks.

2026-06-18
Sudden jump (×2 at once)100×2200Learning resetPerformance dropsGradual (+20% every 2–3 days)100120144173Learning kept · steady growth
Jump more than 20% at once and the algorithm resets its learning, so performance collapses. Raise ~20% every 2–3 days to grow while keeping the learning.

When and How Much to Scale Ad Budget — The Scaling Rules

The core ad budget scaling rules — never raise more than 20% at once (to avoid resetting machine learning), when to increase (3–7 days, frequency check), vertical vs horizontal scaling, and the creative supply scaling actually needs.

2026-06-18
Ad Creative Styles: 9 Types and Which One to Use on Each Platform

Ad Creative Styles: 9 Types and Which One to Use on Each Platform

Nine ad creative styles (cutout, studio, model, lifestyle, offer and more) shown with real examples, plus research on how the effective visual differs by platform (Meta, TikTok, Google) — and why running several styles lifts ROAS.

2026-06-18
Turn a Product Photo into a Styled Ad Visual — No Photoshoot Needed

Turn a Product Photo into a Styled Ad Visual — No Photoshoot Needed

How to turn one cutout product photo into a styled ad visual — scene, props and headline — with AI and no shoot, while keeping the product and its label unchanged. Real before/after example, the common mistakes, and a checklist.

2026-06-18
RegenerationMasterSALE 30%regenerateper size →SALE 30%SAIE 3O%SALE 30%SLAE 30%… N chances to breakPreservationMasterSALE 30%preserve text,rebuild background →SALE 30%SALE 30%SALE 30%SALE 30%… original intact
Regenerating per size multiplies the chances of garbled text by the number of sizes. Preservation never redraws the text, so there is no path to breakage

Making Ad Banners with AI: Image Generators and Banner Tools Are Not the Same

Most AI image tools (Midjourney, DALL·E, Gemini/nano banana) generate one image — not a ship-ready ad banner. The difference between generating a visual and completing/adapting it into a campaign, which tool to use, and how to avoid broken text, distortion and safe-zone clipping.

2026-06-18
Other AI (full regeneration)Move logo to top-right… others drift / duplicateImageFactory (move only)Move logo to top-right… everything else intact
A general model repaints the whole image for one edit, so other elements drift or duplicate. ImageFactory moves only the selected element and keeps the rest intact

How to Edit Just One Part of an AI Image — Without Redrawing Everything

Five ways to fix only one part of an AI image without regenerating the whole thing — masking (inpainting), anchor/pixel/percentage instructions, and layer editing — with how-to, prompt examples and limits, plus drag-and-drop element moving.

2026-06-17
RegenerationMasterSALE 30%regenerateper size →SALE 30%SAIE 3O%SALE 30%SLAE 30%… N chances to breakPreservationMasterSALE 30%preserve text,rebuild background →SALE 30%SALE 30%SALE 30%SALE 30%… original intact
Regenerating per size multiplies the chances of garbled text by the number of sizes. Preservation never redraws the text, so there is no path to breakage

Why AI Breaks Korean Text in Ads — and Why the Real Fix Is Preservation

Why generated images garble Korean (pattern drawing, 11,172 composed syllables), where the four known fixes fall short, and the structural fix for ad work: preserve the text, regenerate only the background.

2026-06-12
LOGORequest: 300×100 (3:1)What you wantedLOGOGenerated: snapped to 21:9Model rounds to a supported ratioLOGOForce-fit: width +27%Circle becomes ellipse — distortion
Ratio-snap distortion: a 3:1 request is generated at 21:9, then force-fitted — turning a circular logo into an ellipse (27% measured)

Why AI Resizing Squashes Your Banners — 27% to 0–2% Distortion, Measured

AI resize distortion comes from aspect-ratio snapping: a 300×100 request snapped to 21:9 measured 27% distortion. How we got it down to 0–2%, plus a practical QA checklist.

2026-06-12
Prompting — "keep it inside"Safe zoneBUY NOWSome outputs: CTA covered by UIGeometric guarantee — pinned by structureSafe zoneBUY NOWAll outputs: pinned inside the safe areavs
Prompt-based placement (left) is probabilistic — some outputs push the CTA outside the safe zone. Geometric guarantee (right) structurally pins key elements inside

Benchmarking 11 Safe-Zone Resize Strategies — QC Scores Published

We benchmarked 11 ways to convert banners into safe-zone placements like Stories and Reels under identical conditions. The quality-vs-guarantee trade-off in numbers: the top-quality method scored 70 on safe-zone compliance vs 100 for the guaranteed approach.

2026-06-12

Try it on your own creative

One master creative becomes every placement size — with the text preserved exactly. 14-day free trial.

Get started free