Skip to main content
RoastIQBuyerLensHugoPricingBlogAbout
Book a demoSign inStart free →
Six reports · updated 18 May 2026 · 28 min read

Industry research, six
reports from 2,047 ads.

Market insights, industry trends, and data-driven reports on advertising performance, each anchored to the same benchmark pool, each with its sample size, each with its confidence label. The opposite of a trends deck.

6 reports 2,047 ads in pool 1,200+ with public outcomes 22 markets 20 categories
Oussama Nakhil portrait
Oussama Nakhil · Founder & CEO
Multiple years buyer-side: NielsenIQ insights, then L'Oréal Groupe in global marketing insights. Every chart in these reports is reproducible from the model_version and benchmark_pool_version pinned in the footer.
+0.31
Held-out OOS Spearman ρ · TikTok engagement
6.5×
Top-vs-bottom quintile lift on public outcomes
CC-BY-NC
Dataset licence · GitHub validation code

Market insights · industry trends · data-driven reports.

Two reports on market structure (MENA creative boom, skincare convergence), two on craft trends (UGC vs studio, sound-off survival), and two pure data-driven benchmarks (TikTok hooks, brand cue timing across industries). Every report carries its sample size and its confidence label.

01
Data-driven report · TikTok · 22 markets · n=700
The state of TikTok hooks, 2026.
Median scroll-stop times across markets and categories. What survives the first two seconds, and what does not. From the held-out 700-ad TikTok cohort that anchors our +0.31 Spearman ρ.
Median scroll-stop
1.9s
across 22 markets (was 2.4s in 2024)
Hook window
0–2.0s
the value, the stakes, or the face
Penalty if missed
−22
composite points (Beat the Skip)
Cohort outperformance
6.5×
top-quintile engagement vs bottom

What is happening

The TikTok scroll-stop is faster than it was 18 months ago, and significantly faster than most creative teams assume. In our 700-ad TikTok cohort (2026-Q1 to 2026-Q2 capture), the median scroll-stop sits at 1.9 seconds, with the 25th percentile at 1.3s and the 75th at 2.6s. Two-second hook briefs are now late by default.

// Median scroll-stop by market (TikTok cohort)

Lower = thumbs move faster. The MENA cohort scrolls slower because the cultural-ritual hook archetype lingers; the US cohort scrolls fastest in the world.
United States
1.4s
Brazil
1.6s
United Kingdom
1.7s
France
1.9s
Germany
2.0s
UAE
2.2s
Morocco
2.4s
Indonesia
2.5s

By category, where attention is hardest to win

CategorySample (n)Median scroll-stop% reaching 2sVerdict mix (Scale)
Beverage821.6s59%38%
Beauty / Skincare1181.7s54%22%
Gaming741.8s61%41%
Telco912.1s68%28%
Finance632.3s72%19%
Auto522.4s74%11%
Fashion981.9s62%26%
Retail / E-com1221.8s63%34%
Headline finding

"Auto is the hardest category on TikTok, slowest scroll-stop, lowest Scale rate. The creative tradition (long establishing shots, drone reveal, badge-at-the-end) is structurally hostile to the surface."

The four hook moves that worked

  • Visual stakes on frame one. Face mid-expression, hand mid-action, result mid-reveal. Static product shots cost 14 points on average.
  • Spoken promise before the brand. +28% completion vs brand-first openings.
  • Pattern-break audio. Sub-bass drop, dialect shift, unexpected silence at <1s. +6–9 points on Beat the Skip.
  • Caption-first composition. Top-third pinned caption, sound-off readable in <600ms.
// Three things to change tomorrow
01
Move your hook brief to 1.5s, not 2.0s
The median moved; your brief should match it.
02
Stop opening on the brand sting
Open on the value, attach the brand inside the first 1.5s.
03
Audit your auto and finance cuts first
The two categories with the worst Scale rate on the surface they get the most volume from.
Method. 700 TikTok ads sourced from Creative Center and TikTok Ad Library between 2026-Q1 and 2026-Q2. Each scored through the RoastIQ pipeline against benchmark pool v.2026-05 (n=2,047). Scroll-stop is derived from public TikTok engagement signals (likes, shares, comments, completion proxies) and modelled per-ad. Confidence label HIGH (n_cohort > 500, ρ +0.31 OOS).
02
Data-driven report · 8 industries · n=1,420
Brand cue timing, across industries.
When the first distinctive asset hits the frame is the single strongest predictor of Brand Impact in our pool. Here is the curve, by industry.
Ads with cue < 1.5s
71
median Brand Impact · 62% Scale
Ads with cue > 6.0s
31
median Brand Impact · 71% Rebuild
Gap
+40
Brand Impact points (top vs bottom band)
Best industry
FMCG
distinctive assets at 0.8s median

The curve

The single most useful chart in our archive: the relationship between first-distinctive-asset latency and median Brand Impact. The shape is consistent across categories, what changes is the slope's steepness and the absolute Y-intercept.

First distinctive asset appears at…Median Brand ImpactVerdict mixCohort lift
0.0 – 1.5s7162% Scale+1.8σ
1.5 – 3.0s5854% Sharpen+0.6σ
3.0 – 6.0s4448% Rebuild−0.4σ
> 6.0s / end-card only3171% Rebuild−1.2σ

By industry, median first-distinctive-asset latency

// Median first cue (s) by industry

FMCG and beverage have decades of distinctive-asset investment to draw on (packaging, mascots, colour). Auto and finance lean on logo reveals instead, and pay for it.
FMCG (food / beverage)
0.8s
Retail / E-commerce
1.3s
Beauty / Skincare
1.7s
Telco
2.2s
Fashion
2.6s
Gaming
3.0s
Auto
4.5s
Finance
5.3s
Operating principle

"Distinctive asset, not logo. A wordmark is one cue; a colour, a mascot, a sonic motif, a typographic system are the rest. Brands with five owned cues clear Sharpen even when the logo arrives late."

// What to do with this
01
Audit your category against the table
If your median cue lands worse than category benchmark, you are leaving Brand Impact on the table by default.
02
Invest in non-logo distinctive assets
The logo is one cue. Mascots, colour-codes, sonic motifs and recurring talent compound across exposures.
03
Stop briefing "end-card brand reveal"
71% of end-card-only cuts in our pool land Rebuild on Brand Impact. The pattern is consistent across categories.
Method. n=1,420 ads across 8 industry codes (mapped to our 20-category taxonomy). First-distinctive-asset latency is annotated by Gemini 2.5 against the Ehrenberg-Bass distinctive-asset definition (colour, character, packaging, sonic, recurring talent, anything mappable to brand without the wordmark). Inter-rater reliability with three human coders on a 500-ad subset: ~85% agreement (Paper 04 in pipeline).
03
Industry trend report · 4 platforms · n=1,110
Sound-off performance, quantified.
The majority of feed scrolling is muted. The minority of ads are built for it. The exact composite penalty, by platform, for cuts whose hook, claim or proof requires audio.
TikTok feed muted
52%
of scrolls (US/UK/FR cohort)
Meta-Feed muted
64%
of scrolls (global cohort)
YouTube Shorts muted
38%
of scrolls (lowest of feed surfaces)
Penalty if dependent
−11 to −15
Get Noticed points

The pattern

Sound-on dependency is one of the seven structural failure modes we catalogue. It is also the most expensive one a brand can ignore, because the cost is silent. The ad runs, the spend posts, the engagement underperforms the brief by 10–15%, and no post-mortem identifies the cause because nobody is watching the muted version internally.

// Composite penalty for sound-on-dependent cuts, by surface

Penalty is the gap between the sound-on and sound-off scored version of the same cut on the same surface (matched n=240, 2026-Q1).
Meta-Feed
−15
TikTok
−12
Reels
−11
YouTube Shorts
−7

What "survives muted" actually means

  • The promise lives in the caption. Top-third pinned, contrast-heavy, readable in <600ms.
  • The brand cue is visual, not verbal. Mascot, packaging, colour code, not the sonic logo.
  • The stakes are on the face. Mid-expression frames carry meaning across the audio layer.
  • The audio is a bonus, not a load. Sub-bass drops, dialogue moments and music payoffs add lift; nothing structural depends on them.
Sharpest finding

"Cuts whose joke, twist or reveal does not survive a muted scroll are paying full media rate for half the impressions. The fix costs zero, caption the promise."

// Three immediate moves
01
Watch every cut muted before approval
If the value, the proof and the brand are not legible muted, the ad is structurally weak, on any feed surface.
02
Caption the promise, not the dialogue
Captioning the script is a transcript. Captioning the promise is a hook.
03
Audio is a bonus layer, not a load layer
Brief audio as additive lift. Reserve sound-on dependency for surfaces where it earns the cost.
Method. Muted-scroll rates are platform-published or industry-published where available; modelled where not. Penalty deltas come from a matched n=240 subset where the same cut was scored against the benchmark with audio enabled and audio nulled. Confidence label HIGH on TikTok and Meta-Feed; MED on Reels and YouTube Shorts.
04
Industry trend report · 12 categories · n=880
UGC vs studio, the 2026 gap.
When UGC wins, when studio wins, and when "inflated UGC", studio production wearing UGC clothing, silently collapses Build Brand. Category-by-category from 880 cuts.
UGC composite advantage
+9
points on TikTok · DTC/wellness
Studio composite advantage
+11
points on YouTube In-Stream · luxury/auto
Inflated-UGC penalty
−14
Build Brand points (Pattern 03)
% of UGC briefs inflated
31%
in our 2026 cohort

The headline

The aggregate UGC-vs-studio question is the wrong frame. UGC wins on TikTok in DTC, wellness and food categories; studio wins on YouTube In-Stream in luxury, auto and finance. The story is the match between surface, category and format, not a global preference.

// Median composite by format & surface

"UGC-native" = filmed on phone, native creator cadence. "Inflated UGC" = studio crew wearing UGC clothing. "Studio" = full production.
UGC-native · TikTok
71
Studio · TikTok
62
Inflated UGC · TikTok
48
UGC-native · YouTube In-Stream
56
Studio · YouTube In-Stream
68
UGC-native · Meta-Feed
61
Studio · Meta-Feed
63

The inflated-UGC trap

Pattern 03 in our failure-mode catalogue, polished studio production wearing UGC clothing, is the single fastest-growing failure mode in 2026. The cut reads as inauthentic to the platform-native viewer; Build Brand collapses because nothing reads as native. We tagged this pattern in 31% of TikTok cuts briefed as "UGC-style" this year.

CategoryUGC winStudio winInflated-UGC hit rate
DTC supplements / wellness+1238%
Food delivery+931%
Beauty / skincare+642%
Fashion (fast)+728%
Fashion (luxury)+811%
Auto+1114%
Finance+922%
Telco+426%
The mistake to stop making

"Briefing 'shoot it like UGC' to a studio crew. The polish leaks through every frame, the platform native-ness collapses, and the cut spends UGC media rates to deliver studio-feel, the worst of both."

// Rules to brief by
01
DTC + TikTok = real UGC, not styled
Hire a creator. Trust the cadence. The +12 in supplements is from cuts that read native, not curated.
02
Luxury / auto + YouTube = full studio
The category's distinctive assets need production craft to compound. Studio still wins where the surface rewards it.
03
Never brief "UGC-style"
If you cannot afford real UGC, ship a studio cut. The hybrid is the most expensive mistake in the table.
Method. n=880 ads, 12 category codes, 4 surface codes. Format classification is annotated per ad against three classes (UGC-native, Inflated-UGC, Studio) using rendered-frame analysis plus production-cue detection (camera motion profile, lighting hierarchy, audio mastering quality). MED confidence on the studio-luxury cohort (n=82); HIGH elsewhere.
05
Market insight report · 6 MENA markets · n=296
The MENA creative boom.
Arabic-first ads outperform English-localised cuts in MENA on every platform we measured. Casablanca, Riyadh, Cairo and Dubai are showing the rest of the region, and the rest of the global brands, what local-native looks like.
Arabic-first advantage
+14
composite points vs EN-localised
Darija TikTok lift
+1.7σ
predicted engagement (MA cohort)
MSA on YouTube
+0.9σ
vs English-VO equivalents
Markets analysed
6
MA · DZ · TN · EG · KSA · UAE

What the data shows

Across 296 MENA-targeted ads sourced from Meta Ad Library and TikTok Creative Center between 2026-Q1 and 2026-Q2, ads scripted in the local Arabic register (Darija for the Maghreb; Khaleeji for the Gulf; Egyptian for Egypt; MSA for pan-regional) outperformed English-localised cuts by a composite gap of +14 points. The gap is largest on TikTok (Darija-first cuts in Morocco at +1.7σ above pool median) and smallest on YouTube In-Stream (where MSA's pan-regional reach narrows the local-dialect advantage).

// Composite by language strategy, MENA cohort

"Local dialect" = scripted, voiced and captioned in the market's specific Arabic register. "MSA" = pan-regional Modern Standard. "EN-localised" = English ad with Arabic subtitles.
Local dialect · TikTok
74
MSA · TikTok
62
EN-localised · TikTok
48
Local dialect · Meta-Feed
68
MSA · YouTube In-Stream
66
EN-localised · YouTube In-Stream
55

Why this is happening now

  • Platform algorithms reward native dialect. TikTok's recommendation surface is denser for Darija content in Morocco than for global English content, distribution is structurally biased toward the local register.
  • Cultural ritual hooks travel. Iftar, Eid, family gatherings, market scenes, the MENA cohort scrolls slower on these openings (median scroll-stop 2.4s vs global 1.9s), giving the brand handshake more room.
  • The MENA creator economy matured. Local UGC talent is now production-fluent in a way that did not exist three years ago; the inflated-UGC trap is rarer here than in Western markets.
  • Global brands are catching up. Inwi, STC, Etisalat, Mr Beast Arabic, Anghami have ridden it for years; FMCG and beauty multinationals started shipping dialect-first cuts in 2025 and are seeing the lift.
The pattern other regions can copy

"Local-dialect-first is not a localisation tactic. It is a platform-native creative discipline. The same logic explains why PT-BR-first beats EN-localised in São Paulo and why Bahasa-first beats EN in Jakarta."

// What to do if you brief MENA work
01
Script in the dialect, not in English
English with Arabic subtitles is the worst-performing format in our cohort. Script natively or ship MSA.
02
Use MSA for pan-regional YouTube, dialect for in-market TikTok
The platform-surface match matters as much as the language choice.
03
Hire local UGC, not regional agencies pretending to be local
The cadence reads. The platform reads it back. The algorithm rewards it.
Method. n=296 MENA-targeted ads, 6 markets. Language strategy classified per ad as Local Dialect / MSA / EN-Localised based on transcription (Google Speech-to-Text) and caption analysis. MED confidence (n_cohort under 350 per language band); pool densification active for Q3 2026.
06
Market insight report · Beauty / Skincare · n=312 · 19 brands
The skincare ad convergence crisis.
When every skincare ad looks the same, the category gets the attention and the brand pays for it. We quantified the convergence, and identified the four brands escaping it.
Convergence rate
67%
of skincare cuts share >6 visual codes
Median Build Brand
52
lowest of our 20 category codes
Pattern 07 hit rate
43%
"generic-category opener" detected
Brands escaping
4 / 19
in our 2026 cohort

What convergence looks like

Open a skincare reel at random in 2026 and you will see: a clean studio, a face mid-application, a pipette in slow-motion, a soft pink-or-beige palette, a sans-serif lower-third claim, a clinical badge in the corner, a packaging beauty shot at the end. We tagged this code-set in 67% of the 312 skincare ads in our 2026 cohort, across 19 brands. Pattern 07 (generic-category opener) fired in 43% of them.

// Distinctive code density vs Build Brand score

"Distinctive" = brand-owned codes that another category buyer would not confuse with a competitor. Pipettes and pink palettes are category codes, not distinctive codes.
≥ 5 distinctive codes
78
3–4 distinctive codes
62
1–2 distinctive codes
46
0 distinctive (category only)
28

The four brands escaping

From our 19-brand cohort, four are running cuts that read distinctively in the first 2 seconds. Each owns at least three non-pipette, non-pink visual codes that map back to the brand without a logo.

  • The Ordinary, tabular ingredient typography, lab-canister minimalism, no faces. Convergence rate inside the cohort: 8%.
  • CeraVe, dermatologist-led narrator, white-and-blue palette, ingredient close-ups, no pipette tropes. Inflated-UGC hit rate: 11% (cohort: 42%).
  • Glossier, pastel-but-distinctively-Glossier palette, recurring talent, recurring kerning. Build Brand 71 vs cohort 52.
  • Typology (FR), apothecary-brown packaging, no model close-ups, label-first composition. Distinctive code density: 6.
The thesis

"Pipettes and pink palettes are category codes. They tell the viewer this is a skincare ad, which the viewer already knew from the algorithm. The job of the cut is to add the brand layer on top, and 67% of the cohort never does."

// What to do if you are inside the cohort
01
Audit your distinctive-code list
Write down the five visual codes a viewer should map back to your brand without seeing the wordmark. If the list is "pipette, pink, face, lower-third claim", you are inside the convergence.
02
Brief the cut against the category, not the competitor
If the cut would still read as your brand if the wordmark was removed, you are out of the crisis.
03
Invest in non-face codes
Faces are category-default in skincare. Packaging, palette, typography and recurring motifs compound across exposures faster.
Method. n=312 skincare ads, 19 brands, 4 markets (US/UK/FR/MA), 2026-Q1 to 2026-Q2 capture. Convergence rate is the share of cuts sharing ≥6 of a 12-code "category default" set, scored against the Ehrenberg-Bass distinctive-asset definition. Confidence label HIGH (n_cohort > 300).

Sources. All ads in the underlying pool are sourced from official public transparency tools (Meta Ad Library API, TikTok Creative Center, TikTok Ad Library, Google Ads Transparency Center) or manual editorial curation with documented licence terms. We do not scrape. Pipeline. Vertex AI Gemini 2.5 Flash/Pro (multimodal) + Google Video Intelligence (shot/label) + Speech-to-Text (transcription) + TranSalNet-class saliency, all scored against benchmark pool v.2026-05 (n=2,047) via Zod-validated structured output. Validation. Held-out OOS Spearman ρ +0.30–0.32 (TikTok engagement, TikTok CTR, YouTube view counts); Meta-Feed flagged DIRECTIONAL until brand-ad cohort densifies. Reproducibility. Every chart is pinned to a model_version and benchmark_pool_version stored against the report. The dataset paper (Paper 05) will release the underlying benchmark CC-BY-NC with validation code on GitHub.

FAQ.

Are these reports peer-reviewed?
Not yet. Five papers are in the publication pipeline (JMR flagship on AI-vs-survey concurrent validity, Journal of Advertising counter-intuitive findings, Marketing Science methodological critique, ISR LLM-as-judge reliability, Scientific Data dataset paper). We publish the misses alongside the hits.
How is "convergence" actually measured?
In the skincare report, by tagging each ad against a 12-code "category default" set (pipette, pink palette, lab badge, lower-third claim, face mid-application, etc.) and computing the share of cuts sharing ≥6 codes. The Ehrenberg-Bass distinctive-asset definition is the operating reference.
Why does the Meta-Feed cohort always carry a DIRECTIONAL flag?
Because the Meta Ad Library exposes the asset but not the engagement signal at the granularity we need for brand ads. We honour that with every report, the scores are directional defaults until our brand-ad cohort is dense enough to publish a ρ. We will not invent precision.
Can I get the underlying data?
Paper 05, the Scientific Data dataset paper, will release the benchmark under CC-BY-NC with validation code on GitHub. Until then, every chart in these reports is pinned to a model_version and benchmark_pool_version that we will reproduce on request.
Will you publish updates?
Yes. Each report carries a benchmark_pool_version pin. As the pool densifies (currently growing ~80 ads/week across categories and markets), we will republish with the new version stamp and a dated changelog. Industry-research reports are versioned, not silently rewritten.
How does this compare to Kantar Link AI or System 1?
Different construct. Survey-based methodologies measure recall and stated response; we predict engagement and click intent against public behavioural outcomes. Both have a place; only one fits inside a creative sprint at €0.005 per ad. The JMR flagship paper (in pipeline) will run a head-to-head on a 200-ad held-out cohort.