In the last 4 years, it has been trendy for journalists, academics, and some policymakers to criticize personalized digital ads. Recent vehement attacks against digital ads include:
Surveillance Capitalism: “[H]ow global tech companies such as Google and Facebook persuaded us to give up our privacy for the sake of convenience; how personal information (“data”) gathered by these companies has been used by others not only to predict our behavior but also to influence and modify it; and how this has had disastrous consequences for democracy and freedom.” When asked to name specific harms, authors like Zuboff (who for years lived off-grid in rural Maine, and then lost all her physical possessions to a fire) give a complete wordcel rant about “privacy; dystopia; control; monopoly; manipulation; intrusion; exploitation; democracy; misinformation; fear; freedom; power; rebellion; slavery; resistance.” (Guardian, 2019)
Selling your information, destroying your privacy: The critique, often by media companies like the NYT and CBS, that large companies like Google and Facebook “sell your data”. (EFF, 2020)
Causing political polarization, eroding democracy: This critique of digital ads may seem outrageous, given that US newspapers have been funded by ads for 130 years or longer, but serious journalists and academics have recently argued for it. (Wired, 2019)
A bill to stop personalized ads: Congresswomen Anna G. Eshoo (D-CA), Jan Schakowsky (D-IL) and Senator Cory Booker (D-NJ), introduced the Banning Surveillance Advertising Act, a bill that would dramatically alter the ways tech companies like Google and Facebook can leverage personal data for online ads. (Adweek, 2022)
I believe these critics fundamentally don’t understand the nature of modern computing, markets, advertising, e-commerce, and tradeoffs between value to consumers and small businesses versus security, privacy, competition, economic freedom, and economic growth.
So in this essay, I will dissect these critiques, by:
- Starting with 3 philosophical views on ads
- Explaining how computing systems and ad systems work at a high level
- Discussing the benefits of personalized ads
- Addressing the complicated tradeoffs in tech policy
- Scrutinizing the main critiques or personalized ads
3 philosophical views on ads
Ben Thompson at Stratechery has noted that most arguments about personalized ads break down into philosophical views about personalized ads. In his essay “Philosophy and Power; Advertising, Targeting, and Tracking; The Real Winners”, he writes:
One of the particularly challenging aspects of this debate is that “privacy” is a great brand name: who doesn’t want privacy? That means, though, that everyone wants to claim privacy as the justification for their preferred outcomes, which is a problem because many of these folks have different goals. There are, for example, three distinct questions about ads themselves:
Is advertising good or bad?
I start here both because I think it is one of the most fundamental questions in this debate, even as it is one of the most overlooked. My sense both from observation and from some of the feedback I get is that a lot of “pro-privacy” folks are just fundamentally opposed to advertising. This constituency will both support just about any privacy initiative, and also be disappointed in the eventual outcome.
Are targeted ads good or bad?
Ads have always had some degree of targeting; even your local billboard was chosen because of its location. Obviously, though, the specificity can vary greatly. One of the factors that drove the explosion in cable channels, for example, was the realization that focusing on a specific type of programming could attract a specific type of audience, making advertising on that channel more attractive to advertisers that wanted to reach that type of audience. Magazines took this concept further than anyone, combining content-specificity with subscriber-level demographic data to attract advertisers.
Still, everyone who read the same magazine saw the same ads; the level of specificity was at the publication level. The web, though, blew this all up: one of the things that made Google so powerful is that the search engine understood the web at the level of the individual article; search results don’t link to the home page of a website, but to a specific post (this is a classic example of how changing the point of integration in a value chain modularizes complements). The question posed here, then, is less about whether targeting as a concept is good or bad — people that dislike even publication-level targeting probably don’t like ads period — and more whether the increase in specificity is good or bad.
Is tracking good or bad?
This is where the zero transaction costs that I discuss in the context of Aggregation Theory come in: it is effectively free on a marginal cost basis to keep track of every page that a user visits, building a far richer understanding of their interests than was every possible when data was only available at the level of the publication, and expensive to collect and reference to boot. That, by extension, meant it was possible to show individual ads to individual users based on that understanding, and here’s the kicker: not only were those ads more effective, but they were also cheaper, both because inventory on the web is effectively infinite (thanks to the zero marginal cost of both creating and distributing content), and also because zero transactional costs applied to the buying and selling of ads as well.
To put it another way, this question is about the shift from content-based targeting to behavior-based targeting.
Once we understand that most critics just object to ads overall, or even targeted ads (as opposed to the waste of millions of people getting the same ad), the question is less one of privacy and more about business models. Do we want to force millions of media companies and app developers to solely rely on payments and subscriptions, and hence eliminate ads or targeted ads, or allow them? Note that, in studies I’ve seen, eliminating tracking probably removes ~30-50% of the value of personalized ads, and removing targeting another 20-30%. Hence if we allow ads, but not personalized ones (no targeting and tracking), you decimate roughly 50-80% of the revenue of media companies and app developers who depend on personalized ads. The small businesses who run ads will have to pay 2x to 5x as much to get to customers. You basically put everything behind subscription paywalls and destroy e-commerce. I will argue below that this has disastrous consequences for the global poor and middle classes, small businesses, and the free internet.
There is also a harder and more legitimate and nuanced discussion around tradeoffs between tracking and privacy. I will address it below in the tradeoffs section.
How computing systems and ad systems work at a high level
It’s important to have a baseline understanding of how the internet and ad systems work before discussing drastic regulatory changes. I’ve been dismayed by how little tech journalists, lawyers, and regulators know about basic concepts, and even how little how software engineers in other fields know about ad systems, so let’s put some basics out here. I will start with simplified concepts, and get more technical along the way, so bear with me. The specifics may differ by company, but at a high level, this is how the major personalized advertising companies work (Google, Facebook, Snap, Pinterest, Twitter, Baidu, Alibaba, Amazon, Apple, etc).
The computing setup: Billions of people have desktops, laptops, and phones, which we will call “clients.” Large content companies, from YouTube to Facebook to TikTok to Apple News and Snap, have content like text, images, video, stories, etc, that they send to clients via websites on web browsers, or directly through mobile apps. The content is stored on and served from millions of servers that live in data centers around the world, owned by Google, Amazon, Microsoft, Meta, etc. Complicated recommender systems pick which content to serve, and in-between content, have ad slots for which to show ad impressions.
The ad setup: Advertisers create the ads they want on the 2-3 dozen platforms that exist (e.g. Google, Bing, Facebook, Tiktok, Snap, Twitter, TradeDesk, etc). Platforms have surfaces on where they can deliver ads (e.g. YouTube, IG Feed, Bing search results, etc), and this is called “inventory.” Advertisers want their ads only shown to users with a higher intent to discover and buy their products – e.g. they don’t want feminine products shown to men, or hunting products shown to non-hunters. The ad platforms don’t sell user data – it’s pretty locked down, but they have a sense of who likes what based on what ads they clicked on or their organic interests. The advertisers may pass their data of prior customers to the ad platforms (e.g. email addresses or people who previously bought feminine products or hunting gear), but it’s a one-way transfer. So to be clear, Google, Bing, Snap, and Facebook don’t sell your data. They sell advertising slots. Advertisers bid for those slots based on an action they care about (ad impressions, clicks, or sales/conversions), customer information they may pass along, or just demographic targeting (persons with Y characteristics in Z location). When the advertisers choose what types of people and demographics they want, they are targeting their ideal customer. Given that running digital ads is expensive (and higher ad costs ultimately get paid by consumers), they want efficient campaigns to keep prices low (and save consumers money).
The machine learning ads ranking recommender setup: After advertisers do the targeting above (the demographic and customer limits), this information is sent to massive ML ranking systems. These systems may use millions of signals, compressed into tens of thousands of numerical tensors called “features” or “embeddings”, which are then fed into an assembly line of hundreds of machine learning models. All this operates in higher-dimensional (non-human-understandable) vector space to maximize an “objective function”, which is the human-defined main goal of the system. Most of these systems have an objective function of serving users ads for specific, high-quality products or services they could be interested in, while also simultaneously doing the best job for advertisers (give them the best results on the action they want, e.g. views/clicks/conversions, given their budget). This is a modern engineering marvel, as these are the most advanced AI systems in the world, generating hundreds of billions of dollars in revenue, and trillions in commercial value. For the most part, they are complete black boxes. For more technical details on how ranking works, see these posts from Facebook and Snapchat (for a more technical example of one type of online ads ranking system, see this 2020 paper from Alibaba).
The benefits of personalized ads
OK, so it’s easy for critics outside the digital ad industry to paint a dire picture of it and criticize personalized ads. But what are the benefits?
1) Personalized ads fund the free internet
Ads pay for much of the internet’s content and apps, which strongly benefit poor and middle-class internet users around the world.
Let me be specific about which apps are funded by personalized ads: Search (Google and Bing), Gmail, Google Suite (Docs, Sheets, Slides), Facebook, Instagram, Whatsapp, Messenger, Snapchat, YouTube, Tiktok, Safari, Chrome, Firefox, etc. The world’s most-used software is funded by personalized ads, and these apps are hence free. If regulators ban personalized ads, or even tracking (reducing the value of ads from 35-80%), this software will have to transition to being paid. Imagine paying between $10-20 per month, for each piece of software above (just like you may for Spotify, Netflix, HBO, or the ad-free YouTube premium). That would easily add thousands of dollars of extra annual spending for the average American household, and basically exclude poorer households in the US, and much of the global poor and middle class. So if you want 2 internets – once for the rich and privileged, and a vastly degraded one for the poor and middle class – banning or hindering personalized ads will get you there.
Digital ads are a $580bn market in 2022 – if you cut that back by 35-80%, the free software that they fund will become paid. Note that making this paid doesn’t make up for the revenue loss from personalized ads – some users are just priced out and can’t pay, and many small businesses go under. Everyone is poorer. Eric Seufert has a great piece with a similar point: “If personalized advertising is banned, who bears the cost?”
2) Personalized ads level the playing field between small businesses and large corporations
During a 2021 survey carried out among small businesses in the United States, 45% of respondents stated that they paid for digital advertising. The average spending amounted to $534 USD monthly, and 93% were planning to keep or increase it over the following 12 months. When regulators cut down on “third party” data flows from small businesses to the platforms, the main winners are the 3 closed platforms, Google, Apple, and Amazon. That’s because they have large amounts of “first party” data on buying habits, as do large corporations like Walmart, Target, Costco, Home Depot, etc. The main losers are small businesses who have to pay more for digital ads, and the platforms that serve them, like Shopify and Facebook. Likewise, when Apple put out its ATT policy to stop tracking (basically only allowing Apple to track and have personalized ads, but not others), it was a blatant anti-competitive move that hurt the advertising industry working with small businesses, like Shopify. I love Tim Cook. But he’s a shark – he can fly the privacy flag while using his platform power to replicate the data practices of his competitors and crush them while he’s at it.
3) Personalized ads power the independent state and local media that support democracy
39% of US newspaper revenue comes from personalized, digital ads. If 80% of that goes away, the trend of dying newspapers will get worse. Roughly 20% have shut down since 2004, roughly 1800 papers, and this decline will be even steeper. Since small and local papers disproportionately cover local news and act as a check against bad government, communities lose transparency and accountability. Research shows that taxes go up and voter participation goes down after newspapers die. That’s right – hurting personalized ads actually harms democracy at the local and state level.
4) Personalized ads allow political newcomers to challenge incumbents, allowing for dynamism and change
Banning the use of personalized ads effectively secures the elections for the candidates with the greatest financial support from corporations and super PACs who can bankroll expensive marketing campaigns. Personalized ads, and especially those on social media, help the underdog and newcomers. In 2018, a record number of women, Muslims, and people of color got elected into the 116th Congress. Many of these previously unknown politicians could win the office despite their limited financial resources thanks to their effective use of social media and personalized ads. Hence it seems particularly troublesome that incumbent politicians want to ban personalized ads – they basically seem to want to hobble their own competition.
5) Personalized ads do all of the above at a low cost
So what are the costs of personalized ads? First, web users may spend a fraction of a second on their feeds or pages seeing ads, or may have to watch 3-5 seconds of video ads between their overall video time. That seems low. Second, users have to be ok with the businesses they shop at sharing their information (a one-way flow) with platforms. Is this the businesses’ data on their customers, or the users’ private data (it’s an open policy question with big consequences)? Recall that platforms don’t share their users’ data – they just make advertising slots open. Some users may object – maybe they want vastly more expensive products by stopping their favorite brands (e.g. Nike, Disney, Starbucks, and all the smaller companies) from targeting them with ads. They can get opt-outs from the brands, but then digital marketing costs go up. And guess who pays for this? The users. There’s a market failure here because brands can’t charge the opt-outs more, though some are smart and give discounts and cash to users who give more information (e.g. the Uniqlo example below). Third, some people worry about sensitive data being shared, like their health information, or details about their sexuality or political views. However, that is already regulated by the GDPR data regulators, which narrowly carves out sensitive categories of data, and so is mostly phased out by ad platforms (in my view, it’s the best part of GDPR).
Addressing the complicated tradeoffs in tech policy
The privacy fundamentalists are often the loudest voice in the tech policy debates, but that’s because most policymakers don’t understand what’s involved. The complete list of tradeoffs is between privacy, competition, national security, data control & security, low-cost goods for consumers, small business success, free speech and assembly, economic liberty, and funding the free internet.
Let’s go through these one-by-one, and show what an ads privacy fundamentalist viewpoint would mean for stopping tracking for ads, or even stopping targeting (hence ending personalized ads).
Privacy vs competition: When the EU passed extreme laws on data privacy, and other countries followed, this benefited the large tech companies who are the only ones with the staffing and technical expertise to follow through (esp. those with first-party data like Apple and Google). Some countries give exceptions to all large companies except a handful of US companies (like discussions with the DMA and DSA now). This leads to the question, do lawmakers really care about privacy, or protecting weaker domestic tech companies from global competition?
Privacy vs national security: We can make systems very private and do full e2e encryption, but that would make them off-limits to national police and intelligence (the messaging debates for Whatsapp, Signal, etc). It’s the same for ads transparency and how much we value advertiser privacy vs national security. Also, how do we feel about a Chinese company, TikTok, getting a lot of freedom to collect data from US and EU users and storing it on servers the CCP can access?
Privacy vs data control & security: Making user’s data portable increases control, but reduces privacy and security as now it’s more widely available for hacking and data breaches. Does this sound implausible? Facebook learned the hard way in the Cambridge Analytica scandal when it allowed academic researchers to get access to some data that they then abused and sold to a bad actor private company to target for political ads.
Privacy vs low-cost goods: As I showed above, if tracking and targeted ads go away, then businesses will have to spend more on digital marketing. Goods costs will go up.
Privacy vs small business success: Ultimately small businesses don’t have the marketing budgets of their larger counterparts, or their tech-savvy. When privacy and data sharing get difficult, they will sell less and go under, while large companies will take their business.
Privacy vs free speech and assembly: We take for granted in the US that there are free speech and assembly rights to target your speech and assembly to narrow audiences. So if you’re a Democrat, you may want to run political ads to mostly Democrats and some independents, and not waste money targeting Republicans. If we stop ad tracking and targeting, political parties lose this, which means powerful incumbent politicians with large war chests will do better.
Privacy vs economic liberty: Who owns a users’ data when they visit a commercial website? Does the business own it, or the user? Can the business set terms forcing the users to consent so the business can share it? Or can you force a business to take anonymous shoppers? Would you want to force a local retailer or ice cream store to take a hooded, masked-up, anonymous shopper?
Privacy vs funding the free internet: As I mentioned above, if you get very strict on ads privacy and stop tracking and targeting, then the free apps they fund will have to be paid. So who would you rather have to bear the cost: the generally richer privacy fundamentalists who can pay for subscriptions, or poorer and middle-class consumers who prefer free apps? (link)
Scrutinizing the main critiques or personalized ads
With the deeper context and facts above, let’s return to analyze the main criticism of personalized ads.
The Surveillance Capitalism critique. This misunderstands the nature of how computers work (the main proponents of this don’t seem to understand computer engineering basics). All cloud-enabled tech performs surveillance, not just ads systems, due to the nature of I/O devices like cameras, sensors, and phones on the edge sending data back to cloud databases. Anytime you use any piece of tech with a microchip and an internet connection, data is being gathered on you and passed to someone else’s servers, for a range of purposes. We live in a world of constant surveillance because we all get a huge benefit from our iPhones, Alexas, Ring Cameras, Gmail, Instagram, etc. Stopping ads will not make this go away (though some new privacy-preserving ML techniques may mitigate the harm). The harder question is what types of surveillance worry us most? Is it Kellogg collecting your data to target you and selling you more cereal via Snapchat? Or is it the US / Chinese / Russian governments reading your personal files, emails, social media, etc?
The “Selling your information, destroying your privacy” critique. Platforms like Google, Apple, Amazon, Microsoft, and Meta almost never sell your information. When they have ad systems, they sell advertising slots where advertisers can bid on segments (without knowing who will see the ads). In fact, the biggest abusers of selling information are the mainstream media like the New York Times, which often both track and sell their users’ data to many people, while flagellating the big platforms that don’t do that. (link) (link)
The critique that the ads business model causes political polarization, eroding democracy. The alternative usually suggested are subscriptions, but studies like this one on Paywall and Content Polarization suggest that subscriptions are worse, and that “newsrooms got more polarized after the paywall adoption: journalists who wrote more left-leaning news articles are more likely to get new byline assignments.” Other studies go further and “call into question the predominating assumption in previous research that social media is a major driver of polarization in society.” (link) (link)
There are legitimate issues with personalized ads, but what this essay frames for you is how valuable personalized ads are – they fund the free internet and apps for the global poor and middle class, support small businesses, and generally help new politicians and democracy.
My biggest concern with personalized ads is that the systems aren’t good enough. Instead of being shown ads for shoes I may like or political candidates that I would seriously consider voting for or donating to, I may get shown irrelevant ads that irritate me and waste my time. So my counterintuitive assertion is that ads are not personalized enough – we have to sit through bad ads instead of the ideal end-state where all the ads are as interesting and entertaining as the main show, as the Superbowl ads in the US sometimes are, but also personalized to the interest of each and every viewer.