Every public note we keep about ad creative.
Four hubs, thirty-plus long-form pieces, nine scored case studies, three methodology pages, eleven free tools. All grounded in the same RoastIQ pipeline. Vertex AI Gemini + public transparency data, no mock-ups, no anonymised numbers.
Pick the angle, not the answer.
Industry research is wide reads across categories. Research insights is the methodology that defends them. Creative analysis is the field guide and per-pattern teardowns. Case studies are nine scored ads, on the record.
Industry research
Cross-category creative pulses, TikTok hooks, brand-cue timing, sound-off performance, UGC vs studio, MENA creative boom, skincare convergence. Built on the same 2,000+ benchmark pool.
Open the hub →Research insights
The methodology pieces. How the benchmark is constructed and defended. Why we use official public APIs instead of scraping. The LLM-as-judge inter-rater reliability work.
Open the hub →Creative analysis
The Field Guide and per-pattern teardowns. First two seconds, brand cue placement, hook-rate vs thumbstop, score interpretation, synthetic-vs-live tests. The long-form behind every verdict.
Open the hub →Case studies
Nine real ads from CeraVe, L'Oréal Paris, Coca-Cola, BMW, Booking.com, Nivea, Nike, Netflix, pulled from public transparency APIs, scored through RoastIQ, every verdict on the record with evidence frames.
Open the hub →"Every score on this site has a trace. If you can't reproduce the verdict from the public ad + the methodology page, we re-did it wrong."
Industry research
Hub home →TikTok Hooks 2026
700 TikTok ads scored against public engagement. The scroll-stop patterns that survived 2026, by hook archetype, opening shot and first-three-second composition.
Brand Cue Timing
First-distinctive-cue latency across 1,420 ads in 8 industries. When the brand actually shows up in the cut, scored against TikTok engagement and YouTube view counts.
Sound-Off Performance
1,110 ads paired audio-only / vision-only. The penalty for sound-on dependency on TikTok, Meta Feed, Reels and Shorts where most users watch muted.
UGC vs Studio
880 ads across 14 category codes and 4 short-form surfaces. The five-signal format classifier (UGC, studio, inflated, hybrid) and which format wins on which surface.
Skincare Category Convergence
312 skincare ads across 19 brands. The 12-code default visual system the category collapses into, and the four brands escaping it.
MENA Creative Boom
Six MENA markets (MA, DZ, TN, EG, KSA, UAE), 296 ads, language-strategy patterns and what local-dialect vs MSA vs EN-localised costs in engagement.
Research insights
Hub home →Benchmark Credibility
500 ads × 43 attributes = 21,500 data points. How the SaliencyLab benchmark is constructed, sourced, and defended against volume-based competitors.
Data Sourcing Defense
We don't scrape. We use official public transparency tools (Meta Ad Library, TikTok Creative Center, Google Ads Transparency). How to prove it and defend against challenges.
LLM-as-Judge IRR
Methodology paper on LLM-as-judge attribute detection: ~85% aggregate observed agreement, formal inter-rater reliability framework. Pre-print under CC-BY 4.0.
Creative analysis
Hub home →The Field Guide
What we learned analyzing 2,000+ ads. The five KPIs, the decision ladder, the seven patterns that kill ads, and the long-form methodology behind every RoastIQ verdict.
First Two Seconds
Twenty openings scored against the same Beat the Skip rubric. Ten that won, ten that lost. The four moves that consistently scored, and the four that consistently failed.
Brand Cue Placement
The single best predictor of Brand Impact in our 2,000+ sample is not how often the logo appears, it's when the first distinctive asset hits the frame. The curve, the audit, and category exceptions.
Hook Rate vs Thumbstop
TikTok's watched-6s and Meta's thumb-stop are not the same number. A clean breakdown of which metric should drive your edit decisions, and which to ignore.
Reading a Verdict
A worked example. An ad scored 62, three reviewers disagreed, here's how the decision trace settled it. Composite math, KPI weights, confidence labels, walked one step at a time.
Synthetic vs Live Test
When 36 simulated buyers tell you what you need, and when you should still pay for a real audience. The decision tree, the Kantar comparison, and the honest limits of LLM-based personas.
What "Benchmarked" Means
A score with no peer is just a number. How big a benchmark pool needs to be by use case, where our 2,000+ ads come from (all official public APIs), and how we defend the methodology.
Case studies
Hub home →CeraVe
Educational-POV TikTok cut. Brand cue lands after second 3, Brand Impact drops to the rule's lower edge.
L'Oréal Paris
The diagnosis behind a beauty brand's mid-band scoring. KPI breakdown + evidence frames + recommended edits.
Coca-Cola
Meta-Feed creative scored through RoastIQ. Composite, verdict, KPI breakdown, evidence frames and recommended edits.
BMW
BMW YouTube creative scored through RoastIQ. Composite, verdict, KPI breakdown, evidence frames and recommended edits.
Booking.com
Booking.com YouTube creative scored through RoastIQ. Composite, verdict, KPI breakdown, evidence frames and recommended edits.
Nivea
What a mid-band beauty cut looks like in the decision trace. Composite, KPI breakdown, evidence frames.
BMW
BMW Meta-Feed creative scored through RoastIQ. Composite, verdict, KPI breakdown, evidence frames and recommended edits.
Nike
Nike YouTube creative scored through RoastIQ. Composite, verdict, KPI breakdown, evidence frames and recommended edits.
Netflix
Netflix Meta-Feed creative scored through RoastIQ. Composite, verdict, KPI breakdown, evidence frames and recommended edits.
Methodology, then the tools.
The methodology pages are the most-cited surface on the site, read them before any RoastIQ score. The product surfaces below are the live tools.
How SaliencyLab scores.
5 KPIs (Beat the Skip, Get Noticed, Brand Impact, Sell Proposition, Build Brand), the decision ladder (Scale · Sharpen · Rebuild), the benchmark cohort and the held-out validation.
Read the overview →RoastIQ scoring.
The Vertex AI Gemini multimodal pipeline. KPI weights, composite math, model versioning, confidence labels and how the decision trace persists.
Read the deep-dive →BuyerLens synthetic users.
36 persona profiles, the scenario-simulation framing, what synthetic panels can and cannot tell you, and the audit trail behind every interview.
Read the deep-dive →Eleven calculators, zero login.
Small focused tools we use ourselves when scoring. Each one solves a single ad-craft question, no signup, no email gate.
Hook Timer
Mark the precise frame your hook lands. Compare against the 2-second decision rule.
Ad Length Helper
Per-platform duration sweet spots and the trade-off curves we've seen in the benchmark.
Ad Specs Lookup
Current platform spec sheet for Meta, TikTok and YouTube, pulled from official docs and kept in sync.
Benchmark Library
Browse the public-API benchmark cohort. Filter by platform, category, region, score band.
Brief Builder
Structured brief template, keeps brand asset, single-message, distinctive cue at the front.
Brand Color Audit
Drop a swatch in, get the contrast scores, distinctive-asset framing, and the category-default compare.
Compressor
Browser-side video compress for upload-size limits. No file leaves your machine.
Copy Check
The single-message, one-CTA, claim-honesty audit we run on every script before greenlight.
Fatigue Window
When to refresh. The frequency curve and the early-warning signals from the public outcome cohort.
Aspect Resize
9:16 / 1:1 / 16:9 re-frame helper. Keeps the brand cue inside the safe zone on every cut.
Thumbnail Audit
The thumb-stop heuristic + a Brand Impact preview applied to YouTube and Meta still frames.