I recently saw reports that Meta deleted AI “deepfake” videos that were spreading false information about a presidential election.
While seeing this, I thought. If, by adding
@ritualfnd , we could go beyond just “deleting fakes” and reduce the confusion itself at the very moment when election-season deepfakes first start to spread, that would be a pretty cool change.
Every election, we come to see similar scenes.
This deepfake’s unknown video spreads like fire first, and checking and corrections arrive much later.
Rather than asking “Can we remove this right away?”, if we ask “How can we make the first few hours play out differently?” and reduce the confusion from deepfakes, what would happen?
So I imagined an experiment that adds Ritual and slows the spreading speed itself.
The key is three steps...
First, catch the surge first.
narrativeWatch does not judge right or wrong; it only looks at “speed.”
If shares grow exponentially in a short time compared to usual, it raises a flag right away.
A map appears of where and which link groups are shooting up.
Second, show context first.
factPack attaches one small card to the link.
In three short lines: whether there is a first source, when the last update was, and whether verification or rebuttal has appeared.
Users see this card before they play the video.
At this point, “throw a screenshot and run” becomes difficult.
Third, lower the speed for a short time.
Only until verification is attached, we slightly reduce autoplay and recommendation boost.
This is not a ban or deletion.
It is a cooldown that buys time so checking can catch up.
The moment a trusted source is attached, it is lifted right away.
To make this not just words, we verify with A/B.
Plan A uses normal labels like usual, and Plan B turns on the three steps above.
We send the same timeline to both groups and compare the following.
-Half-life: the time until the reshare speed bends down to half
-Time for correction to arrive: when a verify/rebuttal card first appears on a user’s screen
-Share-with-context rate: sharing “card video,” not “video only”
-Context-free re-upload rate: within 24–48 hours, the portion that goes back up without the card
We do not get greedy with the goals.
Shorten the half-life by 40–60%, make the correction card arrive within 30 minutes, make the share-with-context above 25%, and make context-free re-uploads below half.
This is not an exaggeration; it is simply “giving the first one hour back to people.”
Execution is simple.
Group the same video family by URL, hash, and caption similarity, and make the card follow everywhere (oEmbed/OG).
Publish thresholds and cooldown rules with version and timestamps, so later we can explain with records why we intervened.
Even if only one of media, watchdogs, or platforms moves together, the effect appears; if all three join, it becomes faster.
We cannot make deepfakes disappear overnight.
But we can slow the spread and attach facts faster.
If we apply
@ritualnet three steps-surge detection, context card, and cooldown-the first hour of election season changes.
The important thing is transparency.
Because a receipt of when and by what rule we intervened remains, we can speak with experiment results instead of a “censorship” debate.
The goal is simple: change the golden hour of confusion into the golden hour of confirmation.
@Jez_Cryptoz @joshsimenhoff