News sponsored by Vodacom Transformation of Work:

By Duncan Lennox, VP & GM of Ads Privacy and Safety at Google

Billions of people around the world rely on Google products to provide relevant and trustworthy information, including ads. That’s why we have thousands of people working around the clock to safeguard the digital advertising ecosystem. Today, we are releasing our annual Ads Safety Report to share the progress we’ve made in enforcing our advertiser and publisher policies and to hold ourselves accountable in our work of maintaining a healthy ad-supported internet.

The key trend in 2023 was the impact of generative AI. This new technology introduced significant and exciting changes to the digital advertising industry, from performance optimization to image editing. Of course, generative AI also presents new challenges. We take these challenges seriously and will outline the work we are doing to address them head-on. 

Just as importantly, generative AI presents a unique opportunity to improve our enforcement efforts significantly. Our teams are embracing this transformative technology, specifically Large Language Models (LLMs), so that we can better keep people safe online. 

GEN AI BOLSTERS ENFORCEMENT 

Our safety teams have long used AI-driven machine learning systems to enforce our policies at scale. It’s how, for years, we’ve been able to detect and block billions of bad ads before a person ever sees them. But, while still highly sophisticated, these machine learning models have historically needed to be trained extensively – they often rely on hundreds of thousands, if not millions of examples of violative content.

LLMs, on the other hand, are able to rapidly review and interpret content at a high volume, while also capturing important nuances within that content. These advanced reasoning capabilities have already resulted in larger-scale and more precise enforcement decisions on some of our more complex policies. Take, for example, our policy against Unreliable Financial Claims which includes ads promoting get-rich-quick schemes. The bad actors behind these types of ads have grown more sophisticated. They  adjust their tactics and tailor ads around new financial services or products, such as investment advice or digital currencies, to scam users. 

To be sure, traditional machine learning models are trained to detect these policy violations. Yet, the fast-paced and ever-changing nature of financial trends make it, at times, harder to differentiate between legitimate and fake services and quickly scale our automated enforcement systems to combat scams. LLMs are more capable of quickly recognizing new trends in financial services, identifying the patterns of bad actors who are abusing those trends and distinguishing a legitimate business from a get-rich-quick scam. This has helped our teams become even more nimble in confronting emerging threats of all kinds.

We’ve only just begun to leverage the power of LLMs for ads safety. Gemini, launched publicly last year, is Google's most capable AI modeI. We're excited to have started bringing its sophisticated reasoning capabilities into our ads safety and enforcement efforts.

OUR WORK TO PREVENT FRAUD AND SCAMS

In 2023, scams and fraud across all online platforms were on the rise. Bad actors are constantly evolving their tactics to manipulate digital advertising in order to scam people and legitimate businesses alike. To counter these ever-shifting threats, we quickly updated policies, deployed rapid-response enforcement teams and sharpened our detection techniques.

• In November, we launched our Limited Ads Serving policy, which is designed to protect users by limiting the reach of advertisers with whom we are less familiar. Under this policy, we’ve implemented a “get-to-know-you” period for advertisers who don’t yet have an established track record of good behavior, during which impressions for their ads might be limited in certain circumstances–for example, when there is an unclear relationship between the advertiser and a brand they are referencing. Ultimately, Limited Ads Serving, which is still in its early stages, will help ensure well-intentioned advertisers are able to build up trust with users, while limiting the reach of bad actors and reducing the risk of scams and misleading ads. 

• A critical part of protecting people from online harm hinges on our ability to respond to new abuse trends quickly. Toward the end of 2023 and into 2024, we faced a targeted campaign of ads featuring the likeness of public figures to scam users, often through the use of deepfakes. When we detected this threat, we created a dedicated team to respond immediately. We pinpointed patterns in the bad actors’ behavior, trained our automated enforcement models to detect similar ads and began removing them at scale. We also updated our misrepresentation policy to better enable us to rapidly suspend the accounts of bad actors.

Overall, we blocked or removed 206.5 million advertisements for violating our misrepresentation policy, which includes many scam tactics and 273.4 million advertisements for violating our financial services policy. We also blocked or removed over 1 billion advertisements for violating our policy against abusing the ad network, which includes promoting malware. 

The fight against scam ads is an ongoing effort, as we see bad actors operating with more sophistication, at a greater scale, using new tactics such as deepfakes to deceive people. We’ll continue to dedicate extensive resources, making significant investments in detection technology and partnering with organizations like the Global Anti-Scam Alliance and Stop Scams UK to facilitate information sharing and protect consumers worldwide.

INVESTING IN ELECTION INTEGRITY

Political ads are an important part of democratic elections. Candidates and parties use ads to raise awareness, share information and engage potential voters. In a year with several major elections around the world, we want to make sure voters continue to trust the election ads they may see on our platforms. That’s why we have long-standing identity verification and transparency requirements for election advertisers, as well as restrictions on how these advertisers can target their election ads. All election ads must also include a “paid for by” disclosure and are compiled in our publicly available transparency report. In 2023, we verified more than 5,000 new election advertisers and removed more than 7.3M election ads that came from advertisers who did not complete verification.

Last year, we were the first tech company to launch a new disclosure requirement for election ads containing synthetic content. As more advertisers leverage the power and opportunity of AI, we want to make sure we continue to provide people with the greater transparency and the information they need to make informed decisions.

Additionally, we’ve continued to enforce our policies against ads that promote demonstrably false election claims that could undermine trust or participation in democratic processes. 

OVERALL 2023 NUMBERS

Our goal is to catch bad ads and suspend fraudulent accounts before they make it onto our platforms or remove them immediately once detected. AI is improving our enforcement on all these fronts. In 2023, we blocked or removed over 5.5 billion ads, slightly up from the prior year, and 12.7 million advertiser accounts, nearly double from the previous year. Similarly, we work to protect advertisers and people by removing our ads from publisher pages and sites that violate our policies, such as sexually explicit content or dangerous products. In 2023, we blocked or restricted ads from serving on more than 2.1 billion publisher pages, up slightly from 2022. We are also getting better at tackling pervasive or egregious violations. We took broader site-level enforcement action on more than 395,000 publisher sites, up markedly from 2022.

To put the impact of AI on this work into perspective: last year more than 90% of our publisher page level enforcement started with the use of machine learning models, including our latest LLMs. Of course, any advertiser or publisher can still appeal an enforcement action if they think we got it wrong. Our teams will review it and, in the cases where we find errors, use it to improve our systems. 

STAYING NIMBLE AND LOOKING AHEAD

When it comes to ads safety, a lot can change over the course of a year: the introduction of new technology such as generative AI to novel abuse trends and global conflicts. And the digital advertising space has to be nimble and ready to react. That’s why we are continuously developing new policies, strengthening our enforcement systems, deepening cross-industry collaboration and offering more control to people, publishers and advertisers. 

In 2023, for example, we launched the Ads Transparency Center, a searchable hub of all ads from verified advertisers, which helps people quickly and easily learn more about the ads they see on Search, YouTube and Display. We also updated our suitability controls to make it simpler and quicker for advertisers to exclude topics that they wish to avoid across YouTube and Display inventory. Overall, we made 31 updates to our Ads and Publisher policies. 

Though we don’t yet know what the rest of 2024 has in store for us, we are confident that our investments in policy, detection and enforcement will prepare us for any challenges ahead.