GARM Brand Safety Scores: State of the Web

Every ad placement is a brand safety decision. When an ad appears next to harmful content, the advertiser's reputation takes a hit — sometimes a career-ending one. The Global Alliance for Responsible Media (GARM) created its Brand Suitability Framework to give the industry a shared language for content risk. It defines 11 risk categories, from adult content to terrorism, each with floor, high, medium, and low risk tiers.

We built LLMSE's GARM analyzer to map our 58-category website taxonomy onto the GARM framework, producing a brand safety score (0-100) and grade (A-F) for every classified URL. We've now scored over 31,400 websites. Here's what the data shows.

How GARM Scoring Works

Unlike our other analyzers (SEO, EEAT, WCAG) which parse HTML for dozens of signals, GARM scoring is a direct mapping from website category and content sentiment to risk level. This is by design: brand safety is about what the content is, not how it's technically implemented.

The framework defines four risk tiers:

Sentiment then adjusts the score within each tier: positive sentiment adds up to +15 points, negative sentiment deducts up to -20 points. Floor-level content always scores 0 regardless of sentiment — there's no positive framing that makes adult content brand-safe.

The Big Picture: 31,424 Websites Scored

93.8% of the web is brand-safe. That's the good news. The overwhelming majority of websites fall into categories — Business, Technology, Education, Health, Shopping, Sports — that carry no GARM risk flags whatsoever.

The remaining 6.2% (1,937 sites) carry some level of brand risk. That sounds small until you consider the scale: 6.2% of the web's millions of active websites is hundreds of thousands of pages where ad placements require careful evaluation.

Where the Risk Lives: The GARM Categories

Our data maps LLMSE's website categories to the 11 GARM risk categories. Here's where the flagged content concentrates:

Floor Risk: Adult Content

566 domains (1.8% of all scored sites) hit the GARM floor — content that is universally unsuitable for advertising. Every single one is categorized as Adult content.

The GARM framework is unambiguous here: floor-level content scores 0 regardless of context, sentiment, or framing. No advertiser brand tolerance level changes this assessment. These 566 domains represent the hard boundary of the ad-supported web.

High Risk: Crime and Sensitive Topics

379 domains fall into high-risk GARM categories:

Sensitive Topics (266 domains) is the largest high-risk category. These are sites dealing with controversial social issues, discrimination, or content that could be perceived as promoting hate speech or aggression. With positive sentiment (e.g., advocacy organizations), they score 45 (C grade). With neutral sentiment, they drop to 30 (D grade). With negative sentiment, they fall to 10 (F grade).

Crime content (113 domains) follows the same pattern. A crime prevention site with positive framing scores 45 (C), while a sensationalist true crime site with negative sentiment drops to 10 (F). Sentiment is the swing factor that determines whether crime-related content is cautiously advertisable or a brand safety risk.

Medium Risk: The Gray Zone

991 domains fall into medium-risk categories — the most nuanced tier for advertisers:

Religion is the single largest medium-risk category at 371 domains. The GARM framework classifies religious content as a "Debated Sensitive Social Issue" — not because religion is harmful, but because advertisers vary widely in their willingness to appear alongside it. A church's website with positive sentiment scores 75 (B grade); the same content type with neutral sentiment scores 60 (B grade, just at the threshold).

Gambling (191 domains) is the second-largest medium-risk category. Legal sports betting sites like DraftKings or FanDuel score 75 (B) with positive sentiment — potentially suitable for brands that accept the association. The GARM framework acknowledges that gambling is regulated and legal in many jurisdictions, making it a suitability decision rather than a safety floor.

Politics (133 domains) illustrates how sentiment transforms risk. A balanced political news site with good sentiment scores 75 (B). A partisan site with negative sentiment scores 40 (C). Same topic, different framing, dramatically different advertiser risk.

Low Risk: News and Government

Two content categories carry minimal GARM risk:

News and Media is mapped to the "Spam or Harmful Content" GARM category at low risk. This means news content is flagged but at the lowest tier — reflecting the reality that mainstream journalism is brand-safe for nearly all advertisers. Good-sentiment news scores 95 (A). Even neutral-sentiment news scores 80 (A).

Law and Government sites similarly carry low risk. Government agency sites with positive sentiment score 95 (A), while neutral-sentiment government content scores 80 (A). This is consistent with advertiser behavior: few brands worry about appearing next to IRS.gov or GOV.UK.

The Sentiment Effect

Sentiment is the single most powerful lever in GARM scoring (after the category itself). Here's how the sentiment distribution breaks down within each risk tier:

Several patterns emerge:

B-grade sites are overwhelmingly positive. 77.9% of medium-risk content has good sentiment — these are gambling sites, religious organizations, political news, and military content that present their topics positively. The 22.1% with neutral sentiment are informational or editorial in tone.

D-grade is exclusively neutral. All 73 very-high-risk sites have neutral sentiment. These are Crime and Sensitive Topics sites where neutral reporting meets high-risk categories, producing scores of 30 (just above floor).

F-grade sites are mostly positive. This seems counterintuitive until you remember that 99% of F-grade sites are Adult content, where the floor applies regardless of sentiment. The remaining F-grade sites are Crime and Sensitive Topics sites with negative sentiment.

What Brand-Safe Looks Like: The A-Grade Web

The 29,487 brand-safe websites break down into familiar categories:

This is the "safe web" from an advertiser's perspective — and it accounts for 93.8% of all scored sites. The distribution mirrors the broader web's category breakdown, confirming that the commercial internet is overwhelmingly brand-safe territory.

GARM in Practice: What Advertisers See

To illustrate how GARM scoring works in practice, here's how different types of websites would score:

The table illustrates GARM's core principle: risk is determined by category, but suitability is modulated by tone. The same topic can span three grade levels depending on whether the sentiment is positive, neutral, or negative.

The 11 GARM Categories: Coverage and Gaps

The GARM framework defines 11 risk categories. Our mapping covers 7 of them:

The four unmapped categories (Online Piracy, Obscenity, Drugs, Terrorism) don't have direct equivalents in LLMSE's 58-category taxonomy. These would require content-level text analysis rather than category-level mapping — a potential future enhancement.

What This Means for Advertisers

The web is safer than you think

93.8% of websites in our dataset score A for brand safety. The persistent narrative that the internet is a minefield for advertisers is overstated. The commercial web — the part that accepts advertising — is overwhelmingly composed of business, technology, education, health, and entertainment content that carries zero GARM risk.

Sentiment is the real differentiator

Within risk categories, sentiment creates a 35-point swing. A gambling site with positive sentiment (75, B) and a gambling site with negative sentiment (40, C) are the same topic but completely different advertising contexts. Advertisers who use category blocklists alone are leaving brand-safe inventory on the table.

The medium-risk tier is where the nuance lives

981 domains (3.1%) sit in the B-grade zone — religion, gambling, politics, military, disaster coverage. These are legitimate content categories where brand suitability is genuinely subjective. A betting company welcomes gambling context; a children's toy brand doesn't. GARM's contribution is giving both advertisers the same vocabulary to express their preferences.

Floor content is a small, known quantity

1.8% of sites hit the GARM floor (all adult content). This is a small, well-defined category that programmatic ad platforms already filter effectively. The brand safety challenge isn't the floor — it's the gray zone above it.

How to Check Any Website's GARM Score

LLMSE offers a free GARM brand safety analysis for any URL. Enter a URL and get back the GARM category mapping, risk level, score, grade, and any applicable issues.

You can also use our comprehensive audit to check GARM alongside SEO, E-E-A-T, AEO, readability, and accessibility — all in one scan. For programmatic access, our REST API provides GARM scoring at GET /api/v1/garm?url=.

Methodology

This report analyzed 31,424 websites scored for GARM brand safety as of February 24, 2026. GARM scores were generated using LLMSE's GARM analyzer (v1.5.22), which maps LLMSE's 58 website categories to the GARM Brand Suitability Framework's 11 risk categories.

Scoring uses a base-score-plus-sentiment-delta model: each mapped category has a base risk score (floor=0, high=30, medium=60, low=80), adjusted by content sentiment (Good: +15, Neutral: +0, Bad: -20), and clamped to 0-100. Categories with no GARM mapping score 100. GARM grading thresholds are A: 80+, B: 60-79, C: 40-59, D: 20-39, F: 0-19.

Limitations: Four GARM categories (Online Piracy, Obscenity, Drugs/Tobacco/Alcohol, Terrorism) are not represented in our category taxonomy and cannot be detected. The analysis is based on category-level mapping, not page-level content analysis — individual pages within a category may vary in their actual brand safety risk.

This analysis was conducted using LLMSE, which has classified over 1.4 million websites across SEO, EEAT, WCAG accessibility, readability, and GARM brand safety dimensions. All data reflects the database as of February 2026. To analyze your own site, visit llmse.ai/classify.