Google yesterday released its 2025 Ads Safety Report, documenting howGemini-powered enforcement systems blocked or removed more than 8.3 billion advertisements over the course of the year - a figure that eclipses the 5.1 billion removed in 2024 and the 5.5 billion taken down in 2023. The report, accompanied by a blog post from Keerat Sharma, VP and General Manager of Ads Privacy and Safety, marks a notable shift in how Google describes its enforcement approach: away from reactive removal after an ad has run, and toward interception before it ever reaches a user.

According to the report, over 99% of the ads blocked or removed in 2025 were stopped before they were ever seen by anyone. That pre-publication interception rate sets a new benchmark compared with prior years, when significant volumes of violating ads were caught only after they had already served.

global_2025_adssafetyreport
global_2025_adssafetyreport.pdf
751 KB
download-circle

From keyword detection to intent analysis

The technical architecture behind this shift is a departure from earlier systems. According to the 2024 Ads Safety Report, Google's older machine learning models required vast datasets to train, making them slower to adapt to new threat patterns. The 2025 report describes a different operating model: large language models (LLMs) built on Gemini that analyze hundreds of billions of signals simultaneously - including account age, behavioral patterns, and campaign structures - to assess the likely intent behind an ad before it reaches publication.

Unlike earlier keyword-based detection, which flagged content based on specific terms or phrases, the Gemini-powered models attempt to understand what an advertiser is trying to accomplish. This distinction matters in practice. A financial services firm promoting a regulated product and a fraudulent operation mimicking that firm can produce nearly identical ad copy at the surface level. What differs is the broader context: payment history, domain age, behavioral signals, and the pattern of account activity. According to Google, these contextual signals are what the current models are designed to parse.

The 2024 report had already described this transition as underway, noting more than 50 enhancements to LLMs launched that year. The 2025 report signals that the approach has matured to the point where real-time review of entire ad formats is becoming feasible. By the end of 2025, according to the announcement, the majority of Responsive Search Ads created in Google Ads were reviewed instantly, with harmful content blocked at the moment of submission. Google indicated it plans to extend this capability to additional ad formats during 2026.

24.9 million accounts suspended

Account-level enforcement showed a significant contraction compared with the previous year. In 2024, Google suspended over 39.2 million advertiser accounts - a figure that represented a 208% increase over the 12.7 million suspended in 2023. The 2025 total stands at 24.9 million suspensions, a reduction of roughly 36% year on year.

ads_safety_report_2024
ads_safety_report_2024.pdf
282 KB
download-circle

Google's explanation for this decrease centers on precision. According to the report, because Gemini better understands the intent behind an ad, enforcement can be directed more accurately at accounts that are genuinely operating maliciously, rather than casting a wide net that catches legitimate businesses alongside bad actors. The report states that incorrect advertiser suspensions fell by 80% last year. That figure has particular significance for the marketing community: in November 2025, Google separately announced that appeal resolution times had improved by 70%, with 99% of advertiser appeals resolved within 24 hours - a process that had previously taken days or weeks and left businesses with no active campaigns in the interim.

The scale of scam-related enforcement remained substantial regardless of the overall suspension reduction. According to the 2025 report, 602 million ads were removed and 4 million accounts were suspended specifically for violations most closely associated with scams - a category that encompasses fraudulent financial offers, fake goods, and impersonation of legitimate brands.

Advertiser verification as a structural layer

Google's advertiser verification program operates as a separate layer of protection alongside the AI-driven detection systems. According to the 2024 report, the program had expanded to cover more than 200 countries and territories by end of that year, with over 90% of ads seen by users on Google coming from verified advertisers on average. The 2025 report reaffirms this program as a critical mechanism, particularly because it addresses a different category of threat: bad actors attempting to enter the system before any ads are ever created, rather than those who slip through initial vetting and run violating content.

Verification creates friction at account setup. By validating the identity of an advertiser before granting access to the platform, Google aims to block fraudsters who would otherwise open accounts, run brief campaigns, and disappear before enforcement systems catch up. According to Google, signals like business impersonation and illegitimate payment details are used as early indicators of potential consumer harm at the account creation stage.

The verification program also carries consequences for repeat offenders. Advertisers suspended for serious violations are blocked from re-entering through new accounts, since their verified identity links back to the suspension. This makes circumvention more difficult than it was under systems where creating a new account required only a new email address.

Responsive Search Ads reviewed in real time

One of the more technically specific claims in the 2025 report concerns the speed of review for Responsive Search Ads(RSAs). These are the dominant ad format in Google Search campaigns - advertisers provide up to 15 headlines and 4 descriptions, and Google's systems assemble combinations dynamically based on the predicted relevance to each query. According to the announcement, by the end of 2025, the majority of RSAs created in Google Ads were reviewed instantly, meaning the system assessed compliance at the moment of submission rather than queuing the ad for asynchronous review.

The practical implication for advertisers running legitimate campaigns is reduced wait time between ad creation and serving eligibility. For enforcement purposes, it means harmful content is blocked at submission rather than after a review period during which the ad might otherwise serve. Google indicated it plans to bring this real-time review capability to more ad formats in 2026, though the announcement did not specify which formats or timelines.

User reports processed four times faster

Alongside the automated detection improvements, the 2025 report highlights a change in how Google processes feedback submitted by users who encounter ads they consider harmful or deceptive. According to the report, Gemini enabled Google's teams to act on more than four times as many user reports in 2025 compared to the previous year.

This acceleration matters for threats that evade automated detection. Sophisticated scam operations sometimes vary their content enough to avoid pattern-based systems, but still trigger complaints from users who encounter the ads. A faster pipeline from user report to enforcement action narrows the window during which those ads can run. According to Google, when a threat does slip through automated defenses, the ability to process user feedback quickly allows safety teams to address it before it reaches a wider audience - and frees those teams to focus on complex cases that require human judgment rather than routine report processing.

What the enforcement breakdown reveals

The 2025 report's category-level enforcement data shows where the volume of violations is concentrated. Among ads blocked or removed, abusing the ad network generated the largest single-category total at 1.29 billion ads. Personalization violations followed at 755 million, with legal requirements accounting for 646.7 million. Misrepresentation - the category most directly linked to scam advertising - totalled 421.5 million blocked or removed ads.

On the restricted ads side, where content is permitted under certain conditions but subject to limits on targeting, placement, or verification requirements, legal requirements led at 504.4 million. Financial services restrictions applied to 273.4 million ads. Online gambling and games accounted for 123.9 million restricted ads, with copyright at 123 million and healthcare and medicines at 119.3 million.

For publisher enforcement, the 2025 report documents action against more than 480 million web pages, with AI-driven systems contributing to the detection and enforcement of over 467 million of those pages - a detection rate exceeding 97%. Sexual content remained the leading category of publisher policy violation by volume, at 409 million pages. Dangerous and derogatory content accounted for 20.5 million pages, with shocking content at 15 million. A total of 245,000 publisher sites faced enforcement action during the year.

The 2025 total of 4.8 billion restricted ads compares with 9.1 billion in 2024. The reduction in restricted ads at the same time as the increase in blocked or removed ads suggests the models are reclassifying some content that was previously restricted into the more serious blocked-or-removed category - though Google has not explicitly stated this.

Scam defenses and the generative AI arms race

The 2025 report is notably direct about the threat posed by generative AI tools in the hands of bad actors. According to the announcement, bad actors are using generative AI to create deceptive ads at scale - producing high volumes of plausible-looking content that mimics legitimate advertising in ways that earlier detection systems were not designed to identify.

The 2024 report had addressed one specific manifestation of this: the rise of public figure impersonation ads, in which fraudsters used AI-generated imagery or audio to suggest celebrity endorsements of scam products. According to the 2024 report, Google assembled a dedicated team of over 100 experts to analyze these scams and developed countermeasures including updates to its Misrepresentation policy. That effort resulted in the permanent suspension of over 700,000 advertiser accounts and a 90% drop in reports of this category of scam ad. Google had earlier co-founded the Global Signal Exchange with the Global Anti-Scam Alliance to coordinate cross-industry scam intelligence sharing.

The 2025 report positions Gemini's real-time review capabilities as the primary response to the generative AI content threat. Because the models can assess intent rather than just content patterns, they are - according to Google - better positioned to identify AI-generated deceptive content even when it has been designed to evade detection by varying its phrasing, imagery, and structure across individual ad instances.

35 policy updates in 2025

According to the 2025 report, Google made 35 updates to its Ads and Publisher policies during the year. The 2024 report had recorded more than 30 such updates. The pace of policy revision reflects the difficulty of maintaining rules that anticipate new advertising formats, new scam techniques, and new legal requirements across more than 200 countries and territories simultaneously.

Among the policy changes with most direct operational relevance for marketers, Google clarified its ad account suspension policies in April 2025 and reorganized its dishonest behavior policy documentation in August 2025, both without changes to enforcement standards. Google also tightened its phone number policy in November 2025 to restrict numbers associated with fraudulent activity from use in call-only ads and call assets.

The report also includes localized enforcement snapshots for 14 markets: Australia, Brazil, Canada, the European Union, India, Indonesia, Japan, Mexico, New Zealand, South Africa, South Korea, Turkey, the United Kingdom, and the United States.

What this means for the marketing community

For advertisers, the 80% reduction in incorrect suspensions is the most operationally significant number in the report. Previous years saw legitimate businesses suspended alongside fraudulent operators, with no fast path to resolution. The combination of more precise enforcement and faster appeals processing - both documented in November 2025 - represents a material change in the risk profile of running campaigns on Google Ads.

The move toward real-time review of Responsive Search Ads also has practical consequences. Advertisers who previously built workflows around review queues - submitting ads ahead of a campaign launch date to allow time for review - may find those buffers less necessary as instant review becomes standard. At the same time, the shift means that policy violations are caught at the point of submission rather than during a grace period, leaving no window to catch errors before enforcement is triggered.

For publishers, the 97% AI detection rate on publisher policy violations indicates that the threshold for having content identified and acted upon has never been lower. The Google policy center improvements from April 2025 introduced clearer labeling to help publishers distinguish between issues that block ad serving entirely and those that merely restrict it - a distinction that becomes more important as detection rates increase.

The broader pattern visible across this report and the previous two editions is one of accelerating automation. The shift from keyword-based detection to LLM-based intent analysis, the move to real-time ad review, the faster processing of user reports - each of these changes reduces the amount of human intervention required in routine enforcement and concentrates human attention on cases that automated systems cannot resolve. According to Keerat Sharma's announcement, the goal is to let safety experts focus on the complex work that requires human judgment, with Gemini handling the scale that human teams cannot.

Timeline

Summary

Who: Google's Ads Privacy and Safety division, led by VP and General Manager Keerat Sharma, is responsible for the enforcement systems and the annual report documenting their outcomes.

What: The 2025 Ads Safety Report documents that Google blocked or removed more than 8.3 billion ads during 2025, restricted 4.8 billion more, suspended 24.9 million advertiser accounts, and actioned more than 480 million publisher pages. Over 99% of blocked ads were stopped before being seen by any user. Gemini-powered systems also reduced incorrect advertiser suspensions by 80% and enabled Google to act on four times as many user reports as in 2024.

When: The report covers enforcement activity during the full calendar year 2025. The announcement was published on April 16, 2026, by Keerat Sharma on the Google Ads and Commerce Blog.

Where: The enforcement actions apply globally across Google Ads, Google's publisher network, and the advertiser verification program, which covers more than 200 countries and territories. Localized snapshots are available for 14 specific markets.

Why: The report matters because it documents the practical effects of deploying Gemini-based intent analysis at scale across one of the world's largest advertising platforms. For marketers and publishers, the data points - particularly the 80% reduction in incorrect suspensions and the shift to real-time review for Responsive Search Ads - have direct operational consequences for campaign management, policy compliance, and the risk of enforcement disruption.

Share this article
The link has been copied!