This year, Facebook, Google, and Twitter have all created archives of political ad spending on their platforms, making publicly available for the first time information about how much are spending, and how effective it is. However, the three companies have different standards when it comes to what constitutes a political ad and what information they’re making available.

Now, researchers at New York University have published the first analysis since these companies launched their political ad archives that compares spending across platforms. The researchers collected data on more than 884,000 political ads included in Facebook’s, Google’s and Twitter’s archives between September 9 and September 22.

While the researchers were able to distill some spending trends — Democratic Texas Senate candidate Beto O’Rourke and President Donald Trump were two of the biggest political ad spenders across all three platforms — one of the biggest takeaways from the report is just how difficult it is to track political ad spending across all three platforms. Without no congressional legislation dictating how social media companies need to label political ads, each of the three has taken a different approach to self-regulation.

Facebook and Twitter started rolling out ad transparency centers in May, while Google published its archive in August. Alongside the ad, each platform includes some information about how well the ad performed, who it reached, and how much was spent on the ad. But there are significant variations: For example, Twitter displays exactly how many impressions an ad generated, while Facebook and Google only give a range.

At the time of this study, Facebook was the only platform to also label issue ads — ads that deal with political topics but don’t endorse a specific candidate — though Twitter began enforcing a similar issue ads policy on September 30.

This means politically adjacent advertisements from for-profit companies — like solar panel companies purporting to help the environment — are showing up in Facebook’s archive, but not Twitter or Google’s.

Additionally, the three platforms vary in terms of the level of access granted to researchers. Assistant professor of computer science and engineering Damon McCoy told VentureBeat in a phone interview that he and his team started using a scraping tool to collect data from Facebook’s ad archive almost as soon as it was released in May.

However, just a few weeks after the archive went live, Facebook blocked the team’s scraping tool. So they had to wait until the company gave them access to a beta API in the beginning of October to begin collecting data again.

While anyone can search for ads in Facebook’s ad archive, McCoy said the archive appeared to stop displaying search results after the first 5,000 ads. That’s problematic for researchers, because Facebook’s microtargeting allows political groups to deploy hundreds or thousands of ads for relatively little money. For example, the researchers found at least 9,880 ads linked to President Trump’s Facebook page during the two-week period of the study. Still, McCoy and his team wrote that even in the beta API, they are prevented from paging past the first 8,000 results.

Twitter doesn’t have an API available for researchers, but McCoy said his team has been able to collect data through a custom scraper they created. Google’s political ad data is the most accessible, as it has all been made available on Google Cloud’s Big Query.

Another signiicant discrepancy between the three platforms is what information they provide about who paid for the ads. Twitter provides the name and billing information of the person who paid for the ad — but that could be an advertising company or the candidate’s PAC. Google meanwhile lists the advertiser’s Employer Identification Number (EIN), or Federal Election Candidate (FEC), and links to that page on the FEC website. That makes it easy for users to see who is funding the PAC, given that many PAC names are purposefully vague about who and what they’re supporting.

“The PACs are messy to track … there are PACs that collaborate with each other,” McCoy said. “We can’t [always] tease apart based on that who exactly was paying for this ad. This is just unfortunately somewhat of an artifact of our political financing laws.”

Though McCoy and his team were only able to track political ad spending over a period of two weeks, there are still a few findings worth noting. McCoy and his team found 117,000 political ads running on Facebook, 3,500 ads on Google, and 103 ads on Twitter during that time period. A higher number of political ads were reported on Facebook in part because Facebook puts more ads in its database and because advertisers are using microtargeting to deploy smaller, cheaper ads.

However, advertisers on Google are spending more per ad — about $1,800, compared to anywhere from $86 to $376 on Facebook. Political ads on Google also generate on average more impressions per ad (33,000-340,000) compared to 4,500 to 13,000 on Facebook and 56,000 on Twitter. However, none of these platforms specifically defines what an impression is.

McCoy and his team also found that left-leaning organizations were spending more on Facebook, while right-leaning ones were spending more on Google. We’ll see if this trend continues to hold with the U.S. midterm elections nearly three weeks away.