How we investigated Los Angeles street takeovers

The Open-Source Intelligence (OSINT) Reporting Lab at the USC Annenberg School began investigating the prevalence of street takeover content on social media in February 2025. Over the course of six months, we archived and analyzed 400 videos, mostly from Instagram, and watched over 93 hours of footage on the livestream service Kick. Both sources of user-generated content resulted in two spreadsheets: a social media dataset and a livestream dataset.

Though our investigation was largely powered by publicly available online content, we also applied traditional investigative reporting methods, including interviews, public records requests and field reporting. We ultimately spoke with over 20 individuals, including law enforcement officers, county and state lawmakers, street safety advocates, community members, business owners, street takeover participants and family members of people who were injured.

What resulted was the most comprehensive reporting yet on street takeovers in Los Angeles.

How we built our datasets 

The two spreadsheets we worked out of gave us two distinct views of the street takeover landscape. The social media dataset featured short clips of takeovers in the past five years, and the livestream dataset gave us a more recent portrait of takeovers and allowed us to analyze the events from start to finish.

Keep up with LAist.

If you're enjoying this article, you'll love our daily newsletter, The LA Report. Each weekday, catch up on the 5 most pressing stories to start your morning in 3 minutes or less.

Subscribe

Our project started with the goal of finding videos posted on social media depicting street takeovers in the greater L.A. area. We found these videos through targeted queries on search engines and social media platforms. Though we found relevant content on multiple platforms, we mainly focused on Instagram and YouTube and used a snowball sampling approach — starting with one post and quickly finding others by looking at the posts of other users who liked or commented on it — to grow the sample of accounts we reviewed.

Over the course of several months of online scouring, our team found hundreds of posts. We limited our social media dataset to 400 videos, which we verified and catalogued for further analysis.

We developed a second dataset by monitoring a Kick livestreamer named Conner Remaly, who regularly attends and livestreams takeovers in the greater Los Angeles area. In November 2024, one of Remaly’s viral livestreams showed two 19-year-old men hit and pinned under a Dodge Charger. From late May to mid-July, we watched 17 of his livestreams, totaling over 93 hours and featuring 167 takeovers across Los Angeles County.

How we located street takeovers 

Since our project was focused on takeovers in the greater L.A. area, we identified the location of takeovers in both datasets. This process, known as “geolocation” by open-source intelligence (OSINT) practitioners, involves the precise determination of where a photo or video was taken using distinct landmarks in the content (e.g., signage, buildings, topography) and cross-referencing with authoritative sources such as satellite, street view or geotagged imagery.

For our takeover content, we focused on street signs, businesses and residential homes to determine where a street takeover took place. This was an easy task for some videos because the signs for the intersecting streets were clearly visible. In other cases, geolocation was more difficult because the takeover videos were filmed during nighttime, often had obstructive features such as smoke and fireworks or were low-resolution. We sometimes had to alter the brightness and contrast of the content to enhance certain features.

Street takeover video posted on Instagram (left) was geolocated to the intersection of Susana Road and Victoria Street in north Long Beach using Google Street View imagery (right).

(

2025 Google

)

The regular appearance of repeat locations, however, also helped our team quickly identify some locations regardless of the video quality. In the example above, although we were analyzing a poor-quality video, our team was already very familiar with the intersection of Susana Road and Victoria Street, a top takeover spot in north Long Beach.

As part of the geolocation process, we captured the exact latitude and longitude of the intersections at the intersecting streets.

How we fact checked 

Online content isn’t always what it claims to be. Images and videos can be manipulated with photo-editing tools or generated with artificial intelligence. As part of our authentication process, we aimed to find people who appeared to have been injured in the videos and livestream footage we reviewed. This included “Dillon” — a pseudonym used in place of the real name — and Efrain Rodriguez, whose injuries were among the most violent that we saw in our dataset. For weeks, we diligently searched online to identify them and reach out for interviews.

We found Dillon as part of our regular monitoring of takeover content on social media. One of the many videos depicting his injury came across our Instagram feed and featured a comment where he identified himself. This breakthrough eventually led us to his contact information on social media. He spoke with us and asked that we not use his real name due to concerns the takeover community would be hostile if they knew he spoke with reporters. We were able to verify the details of his injury during the interview.

We found Rodriguez, whose injury was livestreamed by Remaly on Kick, through court records. One of our reporters viewed the felony complaint filed against the arrested driver at the Orange County Superior Court in Fullerton. The complaint, though short, listed Rodriguez’ partial name and gave some information on the suffered injury: “bone fracture.” With this information, along with clues in previous reporting, we found a GoFundMe page set up by Rodriguez' sister, who was seeking $50,000 for his medical care. Rodriguez declined to speak with us after multiple attempts to contact him.

As part of our work to speak with takeover participants, we also sought out interviews in the comments section of street takeover videos on Instagram. Although most responses were dismissive of our work, sometimes referring to us as “feds,” we were able to find and interview one participant through this outreach.

OSINT Reporting Lab put out requests in the comments section of street takeover videos on Instagram.

(

Instagram

)

How we did our data analysis

In addition to geolocating the takeovers in our datasets, we took a close look at what was featured in the content. Our team indexed each takeover to identify instances of arson, assault, drug use, fighting among spectators, gun shots, looting, people getting injured by vehicles, vehicle destruction and other unique features.

In our social media dataset, 41 of the videos featured at least one person getting injured by a vehicle. Several videos in our research showed spectators getting knocked out by vehicles. Our livestream dataset, for example, showed 27 businesses looted during 19 takeovers between June 15 and July 11. It’s also clear in the data that fireworks, laser pointers, strobe lights and the inhalation of nitrous oxide are regular features of takeovers.

We organized the datasets into two spreadsheets, which our team spent several months putting together. Each incident in our datasets required multiple data points such as the post URL, account username, date of post, post caption, latitude and longitude, street intersection and observed features. Our livestream dataset collected additional information, including the duration of the takeover, the responsible law enforcement agency and looting details.

Though the shortest takeover in our livestream dataset lasted just 20 seconds, the longest takeover lasted 69 minutes. On average, takeovers in our dataset lasted 12 minutes.

With our livestream dataset, we found that the top four intersections for takeovers from May 31 to July 12 accounted for 20% of the total 167 observed takeovers. Of the total takeovers, 74% took place in the county’s second district, represented by Supervisor Holly Mitchell. The district extends from Koreatown to the South Bay.

The vast majority (75%) of takeovers in our livestream dataset took place in areas under the jurisdiction of the L.A. County Sheriff Department’s Carson, Century and Compton stations; the Los Angeles Police Department’s 77th Street and Southeast divisions; and the Long Beach Police Department’s North Division.

What we verified in-person 

As part of our comprehensive investigation into street takeovers, we knew it was important to observe some in real life.

Using Remaly's Kick livestreams of takeovers, one night in July, part of our team geolocated and observed takeovers in real time from Cerritos to South Los Angeles, and back and forth between Carson, Compton, Gardena and Lynwood.

Between locations, we saw participants recklessly drive around us, racing to the next spot. We witnessed the chaos of the takeovers, saw the lackluster response by the Sheriff’s Department and spoke with concerned residents in the areas.

With additional reporting by Raima Amjad.