The Evolution of Online Political Advertising: A Conversation with Who Targets Me's Sam Jeffers

Audio of this conversation is available via your favorite podcast service.

According to reporting in marketing and advertising trade publication The Drum, this year’s US election cycle will see record-breaking spend across all media.

Within the anticipated $12 billion in spending by politicians and interest groups, digital is expected to attract $3.46 billion, or about 28% of the total media spend in the cycle. The Drum says that represents a 156% uptick from the 2020 cycle.

And of course, the US election is just one of many around the world this year.

What can we learn about how digital advertising is being employed in politics, and how major social media and search platforms are performing? What has changed over the years? What data is available for outside researchers to study?

Today’s guest is Sam Jeffers, cofounder and executive director of Who Targets Me. Jeffers has spent several years building a suite of capabilities to make political advertising more transparent, including tools for individuals and data and support for academics, researchers and journalists. His organization also advocates for better policy from platforms, regulators and governments. (You can download the Who Targets Me browser extension to contribute data.)

What follows is a lightly edited transcript of the discussion.

Sam Jeffers:

My name is Sam Jeffers. I'm the Executive Director of Who Targets Me.

Justin Hendrix:

Sam, can you tell my Tech Policy Press listeners what Who Targets Me is, and what are its primary objectives?

Sam Jeffers:

So who Targets Me is an effort to understand how political advertising is used online. Very simply, we want as much transparency as possible so that we can see what campaigns are doing, how they're using data, how they're spending money, how they're targeting different audiences, and really try and use that to explain a really large part of modern political campaigning. A lot of the genesis for this was looking at how foreign influence campaigns were potentially using paid political advertising, but my background is actually in the campaigning itself, and I think it's just tremendously interesting and helpful for people to understand how their votes are being won, what resources are being spent and all the rest of it. That really is the goal. Can we tell this story in real time as much as possible?

Justin Hendrix:

And how many elections have you covered so far?

Sam Jeffers:

That is a very good question, and I think it's a lot. So basically, if you were to count most Western elections since 2017, that would be your number, but I would've lost count at this point. I mean, and even this year I've lost count.

Justin Hendrix:

So it's exciting to have the opportunity to talk to someone who has both big picture perspective on the role of political advertising in elections, but also such a granular perspective, so much data you've collected over time. Can you talk about how you do that? How do you get the data that you put in your databases?

Sam Jeffers:

When we started there was no transparency at all, right? So people were clearly buying large volumes of online political ads. On the big platforms there was just no data about it available whatsoever, so the only way to get going was to start crowdsourcing it. We built a browser extension, the very first version of the Who Targets Me browser extension, for the 2017 U.K. general election, which was called on very short notice. It was what we call a snap general election. So we had about six weeks from the announcement of the election to the vote itself. And because political ads had been such a big part of the 2015 election and an unexpected part of the result, and then obviously the Brexit referendum, then the Trump election as well in 2016, it felt like a fairly important thing to do, to try and reverse engineer how ads were being used.

We built this piece of software. We were quite quickly featured in the Guardian Newspaper, and overnight we almost had 15,000 people crowdsourcing this data, and essentially adding all the ads they saw into a database. That was probably one of the first sizable ad transparency efforts to happen anywhere.

By about 2018 the platforms began to release some data, so they realized there was going to be some regulatory pressure coming their way, there was a lot of media pressure coming their way, and whether preemptively or out of the goodness of their heart, it's not for me to say, but they built ad libraries. And those ad libraries began to be sources of data that we could use for this sort of stuff. So in fact, now a lot of our work is looking at ad libraries and ad APIs, pulling data from those in fairly large quantities, and then trying to add context to that information to make it more usable. I think the ad libraries themselves in some cases are fine, in other cases are quite weak, in one particular case, very poor. And what we try and do is augment that data with good stuff so that we can see who is this advertiser? Which party are they affiliated with? How much money have they been spending over time? How can we just expand that out and do that?

And then most recently the ad libraries have started to publish more information about targeting, and then that allows us to look into aspect as well. So where are those ads showing up geographically? To which demographics? What sorts of interests are they targeting? In what ways are data sources being used to reach people and target them that way? So we've started building more and more dashboards and resources to do that too.

Justin Hendrix:

I suspect that transparency that started, you say, in 2018 wasn't out of the goodness of their heart. I feel like I remember some fairly large scandals, Cambridge Analytica of course being one of the big ones. And then a lot of push by civil society and legislatures in multiple countries around the world to try to ensure this type of transparency. Huge regulatory developments across Europe especially, with the Digital Services Act. When you step back from your perch and you think about the regulatory fabric around political ad transparency, or ad transparency more generally at this point, I don't know, where do you think we're at? Are we much, much better off than we were in 2018? I mean, of course still here in the U.S. we don't have any comprehensive legislation, but has the general picture improved? Where do you think we're at?

Sam Jeffers:

I mean, I think the general picture has improved, the ad libraries, they're just better than nothing. I mean, nothing was the baseline and we've moved beyond that. The challenges we face in doing the kind of work we do is there's just a lot of inconsistency in the data. The classic problems of what is a political ad, who is a political advertiser, and what sorts of data might they publish in order to increase their transparency and trustworthiness is not very well-defined. Even in the new laws in Europe those pieces are still being written and worked out. Who gets access to that data? What's exactly in the data sets? How might we compare what an ad on Facebook and Instagram looks like versus what an ad on YouTube and Google looks like. These are treated in quite different ways by those two services and so on.

So I think the piece that's out there is we still really have the voluntary transparency is the standard we have now. We don't yet have the formal transparency and the requirements needed. And that stuff probably is coming from the EU. I guess, a concern would be that the standard isn't set in a very ambitious way and you end up with something that doesn't advance the state of the art very far at all, or allows individual services too much flexibility in the way they do things and you end up with a lack of competencies. That would be my primary concern right now, and that's some of the advocacy we're trying to do where we're trying to shout about the standard we are expecting of this regulation, and hope that it ultimately meets it.

Justin Hendrix:

So the hope is that the platforms will provide more data that you can crunch and analyze, but are you still collecting data from volunteers, folks who are using that extension and providing you with their data on a voluntary basis?

Sam Jeffers:

Yeah. And we're doing that because the standard of transparency from the platforms we don't think is quite high enough. If you take one example, YouTube, particularly has just a very narrow definition of what a political ad is, right? It's mostly candidate ads, people running at Federal and State level. It doesn't really involve local stuff. It doesn't offer ad transparency at all in lots and lots of countries around the world. It focuses on the markets where it might take a bit of that media pressure and get dragged through the mud for doing bad things. So we do try and offer now a browser extension that goes beyond the original scope, which was Facebook ads, and now includes essentially all the platforms where paid political ads are allowed to run.

And again, we think we do that because, again, Twitter's ad transparency is really sketchy. I still will call it Twitter for the rest of my days. But yeah, it's really sketchy, particularly outside the U.S. The U.K. has obviously left the European Union. That means it doesn't get any of this regulation that Europe is now providing, and therefore gets no ad transparency at all, and isn't considered a market that's worth publishing this stuff. So yeah, we've been using the browser extension really as a way of trying to exemplify what we're looking for, gather some data and so on. And yeah, we'd love listeners of this podcast to stick it on their computer and leave it there for election season.

Justin Hendrix:

We'll certainly put a link in the show notes. Speaking of Europe, speaking of the Digital Services Act, would love to delve a little bit into what you've learned under that new regime. And then, just query you a little bit on how some of the things that you're learning about political advertising relate to the concept of systemic risk in the DSA. I mean, it's still early days, I think, for precisely what systemic risk means when we think about especially politics and democracy, but what are you learning there?

Sam Jeffers:

It's a really interesting problem, right? So as you say, "systemic risk" doesn't have a clear threshold associated with it. There are different people have different interpretations of that. Some people think it's fine to have a really broad interpretation of what systemic risk is so that people can raise all sorts of different queries and questions and do it that way. Other people think, oh no, you have to know what the threshold is in order for it to... It has to have some kind of real-world impact or some kind of knock-on effect.

We were looking particularly in Germany, where we have a lot of browser extension users, the prevalence of what's quite a well-known, at least in Europe, disinformation campaign called Doppelganger, which is where Russia seemingly had been buying quite a lot of advertising, but running all sorts of other efforts as well to try and influence politics in Germany and France mainly. But we were seeing lots of ads popping up, and these ads would look a bit like newspaper cartoons. They would have a caption and an illustration or whatever, and then say a bunch of stuff about Ukraine, or a bunch of stuff about what's going on in Israel, Palestine. And they would pop up a new page, they would run €50 worth of ads, the page would get shut down by Facebook, and on they would go.

And so what we were looking for in the browser extension data was the prevalence, how often do these ads actually appear to real people? Because if you go into the ad library data and add up the numbers get very big very quickly. Suddenly you can say, okay, maybe 30 to 40 million Germans and French people were exposed to these ads, and that sounds like a lot. I mean, that's a third of the combined population of both countries or something.

When you actually look at the browser extension data, like the prevalence, the number of people who actually see this stuff, is fantastically lower, at least in our estimates. And the number of people who might see one ad is, I can't remember, it was 0.2% or something. Again, we've got a full blog post on it. I don't remember the number off the top of my head. But essentially compared to other forms of political advertising, and compared to advertising in general, the chances of you seeing one of these ads, finding the message compelling, being persuaded by it, it's hard to say, right? So the question around systemic risk is almost like we don't know enough. One data set tells us this is a very big problem with a very big number associated. A different way of looking at the data says probably not very high prevalence, not very many people exposed to it. And then we have another problem is we just have no idea what influence that has, what actual impact that has.

I suppose the argument we were making in that blog post is all of these things are perfectly valid in terms of ways of looking at it, but how you actually judge whether you've hit that systemic risk threshold, and of course then the counter arguments you'll get from platforms about whether or not that's happened will just be a open problem for quite a long time. It seems to me that in all cases you can say we did our best to mitigate these problems. The prevalence was low. Other people will say no, this is very bad. You let 3,000 pages pop up and run ads and you didn't seem to be able to stop it. The burden of proof around systemic risk is going to be a really interesting question, and how it develops over the next years is going to be the big thing.

Justin Hendrix:

I guess, that raises a bigger question about your data set and the extent to which it may help people generally study the efficacy of political advertising, whether it works, whether it does shift opinions, whether it does on occasion tip elections. There's the old Wanamaker's rule, the idea. What did he say? Something like, "I know I'm wasting half my ad budget, but I'm not sure which half." I mean, are we any closer to solving that, at least within the context of politics, or getting to a science of efficacy?

Sam Jeffers:

I'm not sure we're massively closer. Some of the transparency stuff, some of the experiments people run has definitely moved some of the science forward on what persuasion is, how it works and so on. A lot of it then seems to be based on quite shifting sands. You'll have a platform where lots of people are paying attention for a period of time, and fashion moves on, and people move to another place. Things go from text and image-based advertising being effective to everyone suddenly doing lots of video everywhere, and all that sort of stuff. So it's a funny one, just the complexity of people's media consumption and environments makes proving the impact of any single piece of communication just incredibly hard. And I think in that sense, one of the things I'm hoping work like the stuff we're doing. It puts you in a direction of saying, well, we just need to do better at sampling content, getting people to volunteer and say, "Look, this is my information environment, this is what I see. These are my opinions over time."

I would love to be in a position where we could build really good panels of users of these tools across lots of countries and be able to say okay, this is what it looks like in Hungary versus the U.S. versus Germany versus Austria versus Spain, and really do that type of work. I think we know so little at this point in terms of into the complexity of people, what they consume, what they read, what they find engaging, what they dismiss, what they ignore. That's a big and challenging project, but I think that is actually probably more helpful in the end around questions like systemic risk than access to data for example, and just trying to get hold of big platform data sets. I think we have to get better sampling stuff from the outside, and looking through people's eyes from the worm's eye view rather than the top down stuff.

Justin Hendrix:

I want to turn to the U.S. because I know you are following the U.S. election closely, and certainly we're, at the moment that I'm talking to you, almost a month out from the final day of voting. What are you learning about political ad targeting for the 2024 elections, especially for the major campaigns for president? What types of things are you observing that are novel in this cycle?

Sam Jeffers:

Interestingly, the political ad market hasn't really changed that much this cycle versus 2020 or 2022. The policies and the platforms are largely the same. The product offerings are quite similar. There's not that much new. I mean, obviously Twitter unbanned political ads, and some certainly on the right of politics seem to be using X again for advertising, but it's not that different. I suppose the thing that I'd say has changed is that the campaigns are behaving a bit differently. I mean, I think certainly for the last several cycles the Democrats just do better at digital in a very formal sense. They run a more professional outfit, they run lots of ads, they spend loads of money, they fundraise quite successfully, they switch well from fundraising from big States to targeting swing States with the messaging stuff and immobilization, all that sort of stuff.

If you were to look at the Trump campaign this time, you'd wonder if there was a campaign there at all. There's lots of fundraising, there's lots of come to my rallies. There isn't the rest of it. There's no policy stuff at all. There's nothing really in terms of mobilization and getting out the vote. They do nothing in terms of voter registration, and they spent much less money. And it's almost as if there's this kind of the charitable interpretation of that would be to say Trump knows what he's campaigning, how his campaigning works. He is such a giant planet in the solar system, all gravity, everything is attracted to him and you can't really escape it.

The other side of it is that he's just so obsessed with TV and just doing things the way he thinks he can get attention he doesn't run a professional outfit. And that is quite a big change from what people interpreted in, say, 2016 when he actually did win the election, that he was very good at Facebook, and Brad Parscale knew what he was doing and ran it. They ran a disciplined and professional thing. That just doesn't appear to be the case this time. So if he wins it will be for other reasons, I think, is probably my prediction.

Justin Hendrix:

There have been some changes in the U.S. and certainly since the last political cycle. I know there was a big judgment against Facebook, some changes to the way that the platforms allow or disallow targeting based on certain factors, sensitive factors like race, ethnicity. What are you seeing there? Are campaigns figuring out a way around that?

Sam Jeffers:

So we've just put together a new dashboard where we're aggregating all of the interest-based targeting, right? So the very classic Facebook targeting of I want to target you because you're interested in baseball or something. We've aggregated all of the reasons that advertisers are using to try and reach voters, and done that for thousands of Republican pages, thousands of Democrat pages. And what we see is just a ton of competition over Latino type stuff. And it's not that you're targeting Latinos directly, it's targeting people who watch certain TV programs, who have come from certain parts of Mexico, who have... It's so obviously race-based proxy targeting. It's faintly ridiculous to claim that the policy was a ban on this stuff. I suppose the question is how problematic that feels.

I mean, the way I see political ads is that they're a combination of money, message and targeting, and so in some senses you can say this is very good, it's pro voting, it's pro voter registration, it's putting information in front of voters in Spanish language or whatever. You can say there are legitimate reasons to do this type of targeting. The flip side is if you use that for voter suppression, if you're using that to spread disinformation, if you're using that in other ways, then that's obviously hugely problematic. I think the thing that we, again, would always like to see in terms of transparency is just clarity that people aren't using those sorts of things in a negative or problematic way, and that they're generally doing it in a pro-social, pro-democratic fashion.

Justin Hendrix:

Latino voters are a major target for both campaigns and a place where we're seeing a lot of rhetoric play out. What about election denial as a factor in political advertising? What are you seeing there?

Sam Jeffers:

It's a good question. I think the answer is, at least as far as I can see on Meta and Google, not a lot, but we could be wrong and it's not a specific focus of what we're looking at the moment.

Justin Hendrix:

What other trends are you seeing with regard to political advertising in this election cycle that you think are notable?

Sam Jeffers:

The main thing about the political ads we're seeing this time are, I mean, I'm just struck by how much the Harris campaign is outspending the Trump campaign. I mean, it has been going on for months, but it has been massive, and only very recently with a few weeks ago has the Trump campaign at least doubled its spending. And it's still spending five times less money than the Harris campaign on political ads on Facebook, for example. I mean, the gap is narrower on YouTube. It's just a massive resource gap. I think we've also seen, as you might expect, that the Harris campaign is really dominating the swing States as well in terms of spending at the moment. They're very much more geographically focused on the places where they need to win. The Trump campaign isn't doing any of that. It's almost weird. Here's seven or eight States that are going to make all of the difference in this election.

If you go back to 2016 where Hillary didn't spend enough money in Wisconsin and it cost her the election or something. If Trump loses one clear reason will be he didn't either raise enough money or spend enough money in the places where the election was actually decided. It's almost funny, like comic, to be able to see that happening in real time a month out before the election. In 2016 we had to wait for the post-mortem and what happened. Now you can say, look, that would be a reason why he might lose, and you can see it a month ahead of time. And he can see it too, right? His campaign can see it and it doesn't seem to matter.

Justin Hendrix:

I want to look ahead a little bit. There's a lot of talk about generative AIs, a lot of talk about the combination of the ability to generate synthetic media alongside targeted information about the potential viewer. There's a expectation that political advertising, as well as all advertising, may be very different in a few years time because of artificial intelligence. How do you imagine being able to follow those things or study those things? How has data collection changed for you? How do your methodologies have to shift?

Sam Jeffers:

I think, if that happens I think it will be very challenging. I mean, we already have very personalized media environments, right? Everyone's making their own choices about what they follow and which services they use and so on. But it's going to be very difficult if we get to the point of, I don't know, AI agents bringing you customized political messages on a plate, personalized video ads with candidates saying things they never actually said but fully endorsed by them, and all the rest of it. I do think there's been a lot of excessive hype about AI and elections in 2024. I think that was always misplaced. I mean, I'm quite happy to say we wrote about that well before it didn't materialize and that we were fairly confident that campaigns weren't going to use it, that it wasn't going to be something that platforms were particularly going to allow to happen.

And I think at the moment it still feels like quite a big technical gap between the way we actually consume our innovation online, which is predominantly through social media platforms, and the technical ability to actually create that level of personalization and customization. It feels a way off. It's quite expensive to do. Certainly for a campaign the risk at the moment of tools going off and saying things they shouldn't be saying, saying things that aren't true, just messing up basically, is too high for you to employ those tools on your behalf. Again, maybe at the beginning of this year I was talking to people who said, "Oh no. Look, AI is going to make all of our fundraising phone calls for us, and you're going to chat back and forward with it and that's going to happen this year. And you're going to really welcome it because someone speaking in the candidate's voice is going to ring you up and it's going to sound very personalized." And all the rest of it. None of that is really technically possible in a reliable way at the moment.

So I think the technology had to come along. I think the platforms are going to have to allow that sort of stuff to happen. Obviously, I think regulation is going to have to allow that stuff to happen as well, and so therefore it might be a bit further off than you think. Obviously one eye on that will be about how you build in some transparency and accountability to anything coming along the line as it happens rather than waiting for it to have some sort of effect, or at least worry people and scare people before you do anything about it. So fingers crossed that we don't wait until 2018 after 2016 this time around.

Justin Hendrix:

So as someone else who runs a small organization that focuses on tech accountability, how do you imagine the next few years? Things are changing, I think tech accountability movement is changing, the threats and challenges are changing, the interests of funders and even researchers in this space are changing. How are you thinking about the future of Who Targets Me?

Sam Jeffers:

It comes at an interesting time, because obviously we've had a very busy year with lots of elections and one very big election still to come, where in some senses for us it's been relatively easy for us to talk about the sorts of issues that we're interested in. We can talk about them both from the, as you say, the tech accountability perspective, but predominantly we're talking about those as just part of modern election campaign coverage. And I think it's been a positive shift this year. In a way it's been watching reporting about technology and elections move from being a curious tech story where people might explain it like, oh, maybe this is interesting, to just having normal political reporters understand how to talk about this stuff, and see it as a routine and modern part of political campaigning.

That is a good thing. That may mean that more of this just moves into journalism over time and moves out of organizations like ours, and in some senses that's actually a success on our part if it does that. I think we would like to see it move more as well into people who regulate elections and who monitor elections. So we would like to see those election commissions and international organizations who look over elections build the capability to read this data and to understand what's going on, and to use it in their own assessments of whether elections are being fairly fought, and think about how you might tweak rules to improve that in future.

So I think that's still a big gap and we would like that to happen, and I think we would like to be part of that where we can say look, we provide data and insights and so on about what's going on. And we've worked in such a variety of countries at this point with different electoral systems and media environments and all the rest of it. We've worked in countries as small as Montenegro, with half a million inhabitants and only a couple of hundred thousand voters, all the way up to Brazil and the U.S. with enormous populations and massively complicated things. I think that would be a really good outcome here is to normalize it into the space it probably should be in, which is it being about elections maybe more than it just being about technology.

Justin Hendrix:

Those seem like good ambitions, and certainly incredible track record for Who Targets Me over these last few years. Grateful for all the work that you're doing, and wonder if there are any other call-outs to our listeners, ways they can get involved other than installing that extension?

Sam Jeffers:

We have a few resources that might be of interest to people. So obviously, as you say, the browser extension is one. We have a website which is at trends.whotargets.me, which is dashboards for over 50 countries where we're tracking all of the political parties and all of their online ad spending in all of those places. We have now added some content summarization into that as well. That will translate the ads into English, it will summarize that content, you can look at weekly snapshots of what people have been saying in their ads and so on. So that's useful for the comparative politics types out there or just the let's just see what they've been saying without having to look at 3,000 of their ads, that sort of thing.

We're also in the process of building some new tooling that will allow you to run experiments on social media platforms. We're very interested in things like labeling, algorithmic ranking, these sorts of things. And because we've got technology and tooling background we're interested in seeing, okay, do these literacy interventions actually work? What happens if we change this, do that? How do people perceive this sort of content versus this sort of content? And so we're trying to run some initial panel studies and just understand what people see and how they respond to it without necessarily having to go through platforms, right? We've seen what happens when you go to platforms for permission to do certain types of studies, how difficult they make access, those sorts of things. So we're looking to find a way to make that easier for a wider range of researchers to do.

Justin Hendrix:

I assume there's a lot of data for any academics who are listening who might like to ring you up.

Sam Jeffers:

We welcome emails, yes.

Justin Hendrix:

Sam Jeffers, Who Targets Me, thank you so much for taking the time to talk to me today.

Sam Jeffers:

Thanks, Justin.