The rapid rise of AI-generated content on social media presents a practical challenge for election officials: how to quickly determine whether something is authentic and wade through the potential deluge of information at scale. This dual threat can impede or dilute an effective response to ensure the right information stays in front of voters. When a suspicious screenshot or media file circulates, officials need tools to evaluate the threat so they can decide whether to monitor, escalate, or issue a public response.
Digital Footprints or Signals
There are two key signals or footprints that are making their way through broad industry adoption, providing officials with capabilities to verify authentic or inauthentic content. It’s important to note that these tools are still maturing, much like generative AI commercial tools, but nonetheless, they still provide a boost to more traditional detective work. The first is watermarks, and the second is provenance.
- Watermarks. A watermark is a signal hidden inside the image or video itself, a bit like the security thread in paper currency. You cannot see it, but the right software can, if it knows what to look for. Several major AI companies now offer public tools that check for their own watermark. You upload the image or video, and the tool tells you whether that company’s system made it.
- Keep in mind: most tools only recognize their own maker’s watermark. If a fake was made with a different tool, or if the platform stripped the watermark when the file was uploaded, the check comes back empty. Empty does not mean real. It just means that the tool did not find its own mark.
- OpenAI’s verification tool checks whether an image was generated by OpenAI systems and was launched in May 2026. Images from OpenAI systems now carry Google’s SynthID watermark, the same standard several other companies have adopted, so one shared signal can appear across images from different makers.
- Google SynthID is a standalone detection portal and is still in limited early access, with a public waitlist. As of their public event on May 19, 2026, Google also began building SynthID checks into Google Search and Chrome, which should be easier to use than the portal.
- Google reverse image search (now part of Google Lens) is not a watermark check. It shows you whether an image, or pieces of it, has appeared online before, which helps most with recycled or out-of-context real photos and face-swaps. It is less useful against fully AI-generated images, which reuse nothing.
- Provenance. This is the concept of tracing the origin or source of where something came from. It’s similar to the physical chain of custody; it just applies to digital artifacts like photos or videos. One of the predominant open standards is Content Credentials from Coalition for Content Provenance and Authenticity (C2PA), a coalition that includes well-known tech companies and more. The premise is simple: Advancing content transparency through metadata and cryptographic signatures in the image or video itself. So, if anyone alters the image after the moment of signing, there’s a disruption in the chain to aid detection.
- There are online tools, such as contentauthenticity.org, that can help dig through content credentials embedded in digital artifacts.
Worth a Note of Caution
These tools help, but do not lean on them too hard. Metadata is easy to strip, and many social platforms remove it automatically when a file is uploaded. Watermarks and signed credentials are harder to remove, but a determined actor can still weaken or erase them, for example by screenshotting, re-saving, or regenerating an image. And it is worth repeating: a clean result is not proof that something is real. Most authentic photos carry no watermark or credential at all. Treat these checks as one input, not a final verdict.
Your Strongest Defense
Your strongest defense is establishing yourself as a trusted voice and being the place people check regularly, coupled with proactive communication about the great security work you’re already doing. This is often referred to as pre-bunking, which educates voters before they encounter something suspicious and helps them recognize that something is off. Then, if they do suspect something, they know where to go for legitimate information or updates.
Things You Can Do Now To Prepare:
- Confirm you have a .gov site and official accounts that are named for the office, and note them on your documentation so it’s readily found.
- Routinely post educational content to build rapport with the community and help spread accurate information over time.
- Directly tell voters that they may encounter questionable information and where to look for accurate and up-to-date information. It’s important to do this before something happens, not during.
- Pre-draft responses and holding statements so you can respond more quickly if an incident does occur.
- Run a tabletop with an AI scenario, and consider adding AI-generated content of you or your office to test your procedures and education.
- If you need assistance here, the Exchange is available to help.
The Resource Library section of the newsletter spotlights election security resources. All highlighted resources are available online in the Resource Library.