Mitch McConnell's Photo May Not Be AI... What About Yours?
What to do (and not to do) if fake AI-generated media of you is distributed online
Many people online think that Mitch McConnell’s ‘proof-of-life’ photo is fake; however, there’s no definitive evidence that it is AI-generated (whether that means it’s not… open for debate).
This highlights how genuine evidence is becoming easier to dismiss because AI can make fabricated images, video, audio, and comments appear real. AI can be used to generate completely new content or to alter a genuine image, voice, caption, background, date, or context; someone might alter a screenshot of a comment or create a voice note with a cloned voice to make it appear that you’ve said something you haven’t. All of these scenarios can lead to numerous personal and professional risks. And AI’s capabilities in this area are changing how we decide whether to trust evidence and how we respond when accusations are made regarding that evidence.
If you’re a person who might be the subject of a deepfake or have a real video misclassified as AI, now is the time to get prepared for how you’ll handle that.
Where is the real crisis?
If you find yourself in a crisis involving AI-generated media or real media being wrongly attributed to AI, the first mistake to avoid is misclassifying the crisis. Both scenarios are a crisis of trust and authenticity and must be handled as such.
It’s a mistake to view the content as the source of the crisis because when these things grow legs, it’s because the audience has a reason to believe the incorrect narrative, and that usually extends beyond the existence of the content. A situation where a fake is mistaken for reality, or authentic material is treated as fake, tends to arise when the audience already feels a lack of transparency and is suspicious; it can also arise when people are emotionally invested.
The first questions to ask are:
Who is being harmed?
Who believes this information?
What decisions might the above make because they believe this information?
Is there any (perhaps uncomfortable) truth that explains why the audience believes this content is (or isn’t) really me?
That assessment helps you determine what type of response you need, how visible that response needs to be, and how it needs to be delivered (depending on your situation, the appropriate response will range from internal messaging and public statements to media responses).
The fact that content can be intentionally passed off as AI when it wasn’t adds another complication to this. Almost everyone knows what a deepfake is now, so if someone is caught doing or saying something that causes reputational damage and they are… not a good faith actor… they have an incentive to falsely claim that real evidence is AI. It’s been shown that falsely claiming ‘misinformation’ (“that’s fake news”) can help politicians regain support after a scandal, and as we know, what happens in politics (unfortunately) often mirrors what happens in our lives. So your audience might not take “this is AI” at face value. It’s important to really think about the context and your answer to Question #4 above to determine if there’s a reason you might be perceived as ‘strategically’ denying the veracity of a piece of content. Yes, this might be uncomfortable for you, but it’s not possible to address the source of people’s doubt if we don’t ourselves accept its existence first.
Regarding real content misattributed as AI, this could also happen not from audience allegations but as a result of you using tools that lead to an ‘AI-generated’ tag appearing on your videos (the images of my Substack screenshots often get a “contains AI-generated media” flag on TikTok, presumably because I compile them in a Canva template and download them, and I keep meaning to investigate how to turn that off). Audiences have been found to perceive headlines labelled as AI-generated as less accurate and, thus, are less likely to share them (even if they were, in fact, true and human-written). So the perception that you’ve used AI can damage trust in itself. This situation may not be as reputationally damaging as having an audience suspect that your proof-of-life photo was made in ChatGPT, but it’s worth having a statement in your back pocket regarding your stance on using generative AI and evidence of your content-making process, should you be questioned.
From a communications perspective, when a fake is believed, you must expose the fabrication and address any underlying context that made it appear believable; when the truth is doubted, you must be fully transparent.
This article was originally going to be about both scenarios, but it was getting too long, so I’ll start with the first one and write next week’s article on the Mitch McConnell-adjacent scenario.
When A Fake Is Believed To Be Real
When AI-generated material is being treated as authentic, your instinct will likely be to make the damaging content disappear as quickly as possible; however, before reporting the content and requesting removal, it’s important to preserve as much evidence as possible. Identify the following and keep a note of it:
The earliest known posts
The platforms the post is circulating on
The users’ account handles
The publication times
The captions
The comments
The reposts
The media files, if you can download them
This information will help guide your response.
The level of engagement (comments, likes, reposts, posts about the post) the post is getting will help you determine, for example, whether you need to make a public rebuttal or if you need to address it with a smaller portion of your audience; whether it needs to be addressed on multiple platforms or contained within one. Threads and semi-private Facebook groups, for example, often contain a lot of ‘drama’ that never makes it onto other platforms, even Instagram or the wider Facebook platform. Posting a response to too wide an audience or on platforms where the situation isn’t being discussed can inadvertently take a ‘fringe fake’ to a mainstream audience. A wider response will be needed when the content has traveled beyond the initial platform across multiple platforms and potentially in mainstream media. So, ask:
Which audience needs the correction?
Through which channel?
With how much detail?
Then, your tasks are to clearly establish the truth and protect anyone targeted. Complete silence is a mistake, as it makes it more likely that the fabricated content will be the first impression of the situation your audience comes across. If you don’t have an easily findable rebuttal of that false information within the platform or group where it is spreading, it will be assumed to be true.
You don’t need to make any sweeping claims before you know the facts (e.g., about who created the misinformation); just establish an authoritative source quickly so you don’t give away the narrative to those spreading and/or believing false claims. As soon as you become aware of the fake content, you can make it known that you’re aware of it, that it doesn’t reflect your values/beliefs/whatever falsehood it’s spreading, and that you’re investigating the origin of it. You can also ask your audience not to share the material further, and while it might be tempting to embed the offending content in your response to it or share it with your response attached… don’t. That’s feeding the distribution system, and your goal is containment.
Empathy is important here, too, as fakes can be very convincing, and even though you didn’t make the content, people can still feel hurt by it (and feel that they have been hurt by you). We tend to want to aggressively deny when we’re accused of doing something we haven’t done or holding a belief that we don’t have; however, strong denials can come across as excessively defensive and make you inadvertently come across as though you have something to hide.
Timing matters as much as wording.
You shouldn’t wait days or for your investigations to be finalized when the fake is spreading to people who may act on it or be harmed or even targeted as a result of it. Respond as soon as you have confirmed that the fake exists, you understand the basic claim it makes, and you’ve identified where it’s circulating. You don’t need to know who made it or prove how it was made before responding. You can make a more substantive update once you’ve verified additional facts, if necessary.
It’s understandable to want to be careful; however, delaying a response because you’re waiting for proof can create a vacuum, and we all know how nature feels about vacuums. Be accurate and avoid accusations you can’t substantiate, but don’t leave your audience with nothing but the fake to speculate on.
Here’s a sample you can modify for your situation for, for example, a text-based Instagram post or story:
“I’m aware of a fabricated [screenshot] circulating on [platform: e.g., Instagram and Threads] that [brief description of the content: e.g., falsely attributes a hateful comment to me]. I understand that this content [acknowledgement of the contents and its effect on the audience: e.g., contains extremely hurtful messaging that has negatively affected many of you]. I want to reassure you that I [credible rebuttal statement: e.g., do not hold the beliefs attributed to me in the fabricated comment], and I am deeply concerned about [the impact: e.g., the harm and distress it has caused]. I am [brief explanation of actions you’re taking: e.g., gathering evidence and investigating the origin of this screenshot, along with reporting it to Meta]. Thank you for your understanding and for refraining from sharing the fabricated content further. [Optional: Indication of future updates (only specify the date/time if you are sure you can honor that; otherwise, omit it): e.g., I will share a further update by [date/time] if there is material information to add.]
Regarding the acknowledgement of harm and distress, for example, keep in mind that acknowledging such effects is not the same thing as taking responsibility for them. You don’t have to be responsible for an outcome to acknowledge its existence, and showing empathy to the people affected by the fake content enhances your credibility because it shows that you care about the impact. A good acknowledgement doesn’t blame-shift (“I can’t believe you’d believe I’d say this!”), it recognizes the harm and distances you from it (note: that distancing part is applicable only when you genuinely didn’t do the thing).
Your next tasks are verification, reducing distribution, and redirecting back to your response.
Proving content is or isn’t AI-generated isn’t exactly a reliable science (yet), and demonstrating absolute proof that something is fake might not be possible. Verification will differ depending on your circumstances and the content of the false media. For example, you might want to demonstrate where you were at the time something supposedly happened, or ask people who were with you to verify that you didn’t do or say what the content is depicting. You may be able to show the original time-stamped version of an image, comment, or video that has been used to generate a fake.
When the harm is severe, it may be necessary to bring in qualified digital forensics and legal expertise. (Communications can’t solve a severe issue alone; it can only help share the findings appropriately in a way that is transparent and builds trust).
To help reduce distribution, report fabricated material to every relevant platform under the most accurate available category, such as impersonation, non-consensual synthetic media, harassment, privacy, fraud or coordinated inauthentic behavior. Keep confirmation numbers and any copies of reports and correspondence that you receive. A concise report identifying the false attribution, likely harm, relevant policy, and supporting evidence is more useful than a vague complaint that something is “fake”. If a journalist or high-profile creator has repeated the claim, approach them directly with an evidence-based correction request and be calm and concise about the timelines and supporting documentation when you do it.
Keep in mind that ‘takedowns’ aren’t going to stop reposts of screenshots, and a rumor can outlast the original post, so remember that the goal in reducing distribution is just that. It’s your communications surrounding the issue that will reduce your audience’s belief in the false information.
Finally, redirect back to your main statement whenever necessary. Don’t get caught up in debates with anonymous accounts and fighting people in the comments. Correct material facts once, where the relevant audience will see them, and then in future, direct people to that comprehensive update. Reporting, monitoring, and clear stakeholder communications are always more valuable than getting stuck in endless loops of reply chains that often do nothing but keep an issue current when it doesn’t need to be.
Common Mistakes To Avoid
Claiming an AI detector is definitive evidence: AI detectors aren’t reliable yet, and someone could run the thing through a different one and get a different result, which could undermine your credibility. Use AI detector outputs as one element of your wider evidence base and response, and acknowledge that they’re unreliable if you use them.
Insisting that fake content is “obviously” fake: This can come across as a bit of a ‘weak’ defense and like you’re shifting blame to your audience for not recognizing the fake content.
Repeating all the information in the fake content in your response: Going into excessive detail about the fake content itself and rebutting it line-by-line can come across as defensive and also introduce your audience to the falsehoods you’re supposed to be containing and correcting. Name the harm at an appropriate level and lead with the truth.
Getting trapped in the comments: Fighting with every commenter boosts engagement, especially if you start doing it on the fake content itself. You’ll bring even more attention to the issue. Correct once, direct people to that source of truth, then disengage (yes, I know it’s hard).
Issuing a vague statement: For example, an “apology for the confusion”. As I mentioned above, don’t get all into the nitty-gritty of it, but be specific enough about the facts (e.g., describing the type of misinformation). Being too vague can come across as you being responsible or uncertain about the underlying facts. It’s important to be empathetic and sufficiently detailed about the surrounding concern, and you can do this without accepting blame for the fabrication.
Accusing/naming other creators too soon): Misidentifying the source of the fake content can easily create a new reputational (and potentially legal) crisis. You need to be absolutely sure you know the person was responsible if you’re going to name them. And consider why you want to name them. Until you are sure you know and that you want to publicly call them out (this comes with its own risks; that is a topic for another post), stick to the facts that you can substantiate and feel comfortable with. Let your/others’ investigations establish attribution.
Build Trust First
Hopefully, you’re reading this before needing to use the advice in it. If that’s you, the best thing you can do to protect yourself against deepfakes and other AI-fabricated content designed to damage your reputation is to build trust with your audience before you need to call on it.
The work that best saves people in a crisis like this starts long before that crisis happens, by making it easy for your audience to check what’s real and giving them a record of how you communicate when something goes wrong, built over time. An audience that knows what your normal communication looks like and how you handle uncomfortable situations has more to go on than a convincing screenshot or video. I don’t even mean “be nice” here. You might be a creator who quote-comments your audience in videos and calls them out. I know several creators who do this and do it respectfully in a way that adds value to the commenter and their wider audience. If a deepfake surfaced of one of them calling someone out purely to be mean with no value whatsoever, even a person who thought the creator’s overall tactic was mean would recognize the extra-meanness of it and think, “That’s not like his usual callouts.”
You can also set up your comms channels to be unmistakable sources of genuine information. Maybe you always post important updates on your website, in your story highlights, in pinned posts… it doesn’t really matter where you put these things, just that when you have them, you put them in the same place every time so that your audience knows where to find them and they’ll know they’re genuinely yours. Link all your profiles across your platforms in a link-in-bio to make it clear which accounts are actually yours. Some creators put “this is my only account” or “I will never DM you first” in their bios when they often have their account spoofed and have fake profiles attempting to scam their followers. All of these things make it less likely that a fabricated account, screenshot, or video will become a credible source of information about you.
If you do use AI or have a particular stance on it, make discussion of that a normal part of your communication. Talk about the tools you use (or don’t) for what and why and how you disclose that use. Talking about it means people will be less surprised by an AI label or less likely to assume that an unfamiliar format must be fake. And if you have other people making or publishing content on your behalf, have a straightforward process for that and make sure that you have appropriate proofing and verification steps for anything you’re not personally creating.
Trust also comes from behavior. Correct small mistakes promptly, admit when you don’t know something, explain meaningful changes, and respond when people raise legitimate concerns. If there are patterns in how you behave, or have behaved, that make a harmful claim seem plausible, think about them now and address them before someone else tries to use them against you so there’s more distance between your stated values and any potentially negative past observable conduct.
An audience that already knows how you communicate, where to look to verify information about you, and how you behave under pressure is harder to deceive and less likely to believe that unexpectedly harmful content is genuinely from you.
Engage in a way that limits any uncomfortable truths that could cause your audience to believe that fake harmful content is really coming from you.