Start with the caption, because the caption is where the trick lives. An AI-generated image of Jesus rendered as a crab, worshiped by other crabs, was posted to Facebook with the line “Made it with my own hands!” According to Renée DiResta and Josh A. Goldstein’s study in the Harvard Kennedy School Misinformation Review, that post drew 209,000 reactions and more than 4,000 comments. The image is the bait. The caption is the hook. The comment thread is the water.
What follows is a close read of the post as a designed object — image, caption, and comment surface together — using the documented record and keeping speculation labeled as such.
The artifact, precisely
The crab-Jesus image belongs to a family of AI-generated pictures that DiResta (Stanford Internet Observatory) and Goldstein (Georgetown’s Center for Security and Emerging Technology) catalogued in a study of 125 Facebook Pages, each of which had posted at least 50 AI-generated images. The researchers sorted those Pages into spam, scam, and “other creator” categories. As of April 2024, the Pages in the sample had a mean follower count of 146,681 and a median of 81,000.
The crab-Jesus post is the study’s most quotable exhibit, and it earns the attention. The image is not subtle. It is a composite of sacred iconography and seafood, rendered in the hyper-saturated, slightly plastic register that diffusion models produce when prompted toward awe. The caption, though, is the part that turns a strange picture into a machine.
“Made it with my own hands!” is a sincerity test
Read the caption against the image and the contradiction is total. No hands made this. The sentence is not a description; it is a prompt for a specific kind of reply. It invites congratulations, praise, and the small social transaction of acknowledging someone’s craft. The image supplies the spectacle; the caption supplies the role the viewer is meant to play.
This is the first design decision, and it is not an accident. The caption converts a surreal image into a sincerity test, and the comment thread is where that test gets scored. A close read that stops at the pixels — the crab legs, the halo, the lighting — misses the mechanism entirely. The post is a three-part object, and the caption is load-bearing.
Double legibility is the payload
The crab-Jesus image is legible on two registers at once. To a media-literate viewer, it reads as absurdist collage, a joke about the aesthetics of synthetic devotion. To another viewer, it can read as sincere devotional or folk-art content — strange, yes, but not obviously fake. The researchers found that comments on these images often suggested users did not recognize them as synthetic, including users who congratulated an AI-generated child for an AI-generated painting.
That double legibility is a feature of the format, not a failure of the audience. The image does not need everyone to believe it. It needs enough people to react to it, and the ambiguity widens the pool of reactors. The media-literate viewer who comments “this is AI” and the viewer who comments “beautiful” are both feeding the same metric.
The funnel behind the picture
The study describes spam Pages using clickbait tactics to direct users to off-platform content farms and low-quality domains. The image is the hook, the caption is the ask, and the comment section is the conversion surface. 404 Media’s reporting on the phenomenon described hundreds of AI-generated spam pages posting dozens of times a day and being rewarded by Facebook’s recommendation algorithm.
The researchers also documented that at times the Facebook Feed recommended unlabeled AI-generated images to users who neither followed the Pages posting them nor realized the images were synthetic. They suspect this happens because the algorithmic Feed promotes content likely to generate engagement, and they note Facebook has increased the share of “unconnected posts” in users’ Feeds over recent years. That is a description of correlation and researcher suspicion, not a measured causal finding, and it should be read that way.
Two audiences, one metric
Sort the comments on a post like crab Jesus and you find two kinds of respondents. One group plays along, offering the praise the caption requested. The other group calls it out, sometimes with infographics warning others about scams. Both are engagement. The platform does not distinguish between a comment that says “amen” and a comment that says “this is fake” when it counts interactions.
This is where the criticism gets interesting and where it gets easy to condescend. The useful move is not to mock the believers or applaud the debunkers. It is to describe the co-production: the post supplies a role, and the audience completes it, in either direction. Slop is not a property of the image alone. It is a relationship between a template and a crowd.
The template, not the subject
Here is a hypothetical example, clearly labeled as such: a Page posts a photorealistic image of a cake with the caption “This is my first cake! I will be glad for your marks.” That is the same template as crab Jesus with a different emotional register — pride instead of awe. The study found exactly this pattern, with AI-generated cake photos and the phrase “This is my first cake. I will be glad for your marks” appearing across multiple Pages, and posts with that phrase gathering more than a million collective interactions since December 2023.
The reusable asset is the template: a photoreal image, a first-person claim of authorship, and a caption that solicits a specific reaction. The subject is interchangeable. Crab Jesus is memorable because the subject is bizarre, but the machinery would work with a cabin, a wood carving, or a child’s drawing. The study’s common themes included AI-generated houses and cabins (43 Pages), children (28 Pages), wood carvings (19 Pages), and Jesus (13 Pages).
What the record does and does not establish
The documented facts are specific: the crab-Jesus post, its caption, its 209,000 reactions and 4,000-plus comments, the 125 Pages studied, the follower counts, the off-platform funnel, and the pattern of commenters not recognizing synthetic origin. A 2025 conference abstract on Albanian-language Facebook posts found a related pattern: AI-generated images paired with captions designed to provoke interaction spread widely by exploiting emotions such as anger, fear, pride, and empathy, not because of their credibility.
What the record does not establish is who made the first Shrimp Jesus image, when, or how many total views the family of posts received beyond the documented figures. It does not establish that Meta’s enforcement reporting addresses this specific phenomenon; the transparency report reviewed for this piece covers policy areas like spam and fake accounts but does not mention Shrimp Jesus. And it does not establish that the recommendation algorithm caused the engagement rather than correlated with it. Those distinctions matter if the goal is criticism rather than vibes.
The craft lesson
For anyone writing about platform-mediated visual culture, the crab-Jesus post is a useful discipline. The artifact is not the image. It is the post: image, caption, and comment surface together. The meaning of slop is co-produced by its viewers, and the critic’s job is to describe that co-production without condescending to either audience.
Read the caption. Read the comments. Notice which role the post is asking you to play, and notice when you are playing it. That is the whole anatomy, and it is more interesting than the crab.
FAQ
What is Shrimp Jesus?
It is a family of AI-generated images in which Jesus is merged with a crustacean, posted to Facebook by Pages that researchers classified as spam, scam, or other creators. The most-cited example is an image of Jesus as a crab worshiped by other crabs, captioned “Made it with my own hands!”
Did people believe the images were real?
Comments on these images often suggested users did not recognize them as synthetic, according to DiResta and Goldstein. The study gives the example of users congratulating an AI-generated child for an AI-generated painting. A subset of commenters did identify the images as fake and warned others.
Why do Pages post this content?
The study describes spam Pages using clickbait to drive users to off-platform content farms and low-quality domains, and scam Pages attempting to sell non-existent products or obtain personal details. The image is the attention hook; the caption solicits the reaction that feeds distribution.
Does the Facebook algorithm cause this?
The researchers documented that the Feed at times recommended unlabeled AI-generated images to users who did not follow the posting Pages, and they suspect this reflects the Feed’s promotion of engagement-likely content. That is a suspicion grounded in observation, not a measured causal finding.
How should a critic write about an image like this?
Treat the post as the unit: image, caption, and comment thread together. The caption tells you what reaction the post is asking for, and the comments show how different audiences complete the transaction. Describing that co-production is more useful than judging the picture in isolation.