DiResta / Goldstein study · Evidence case
Religious pictures can become part of an attention economy as well as devotional expression. The evidence must distinguish those uses.
Finding: A published study includes religious imagery within a selected sample of AI-image pages; it is not a census of religious art.
The study examined 125 Facebook pages posting substantial quantities of generated imagery. It distinguished spam, scam and other creator categories; its examples include religious composites. The sample was selected for the study, not drawn to represent all religious users.
A religious picture may be an acknowledged illustration, an act of devotion, a joke, an engagement tactic or part of a deceptive claim. Appearance alone cannot settle the creator’s purpose. The surrounding caption, account history, destination links and documented conduct provide different pieces of evidence.
An interaction count is not a count of sincere belief. Comments may express recognition, disagreement, humor or automated activity. Even a large number cannot establish how many viewers thought a picture recorded a real event.
Suppose a synthetic picture is captioned as a pastor endorsing a product. The decisive question is whether the pastor actually endorsed it, not whether the portrait looks realistic. A labeled fictional illustration makes a different claim. This example illustrates the distinction and is not an allegation about a particular pastor.
A selected set of Facebook pages can reveal patterns of activity within that set. It cannot establish the prevalence of the activity across the entire platform unless the sampling method supports that inference. The distinction is fundamental to this case because vivid examples can make a phenomenon feel universal even when the research asks a narrower question.
The study is useful evidence that generated religious imagery can appear within audience-building activity. It should not be turned into a claim that all religious AI images are scams or that everyone engaging with them is deceived. Those conclusions would erase differences the record needs to preserve.
A reader citing the research should identify the unit studied: pages, posts, images or interactions. These are not interchangeable. A page can publish many images, and a single person can make several interactions. Counting one unit and describing another can produce a misleading impression of scale.
A picture may appeal to devotion, curiosity, humor or surprise. An account may seek attention while some viewers respond sincerely. The creator's strategy and the audience's experience need not be identical. A useful analysis avoids assuming that one explanation excludes every other one.
The surrounding conduct matters. Captions, posting patterns, destination links and requests for money can provide evidence about what an account is doing. An image's religious appearance alone cannot establish motive. Likewise, an attractive or moving image does not establish that its distribution is honest.
This is why the case should distinguish illustration from deception. An openly synthetic devotional image makes a different claim from a fabricated photograph of a miracle or a false endorsement by a religious leader. The moral and factual questions change with the claim made to the audience.
A person may comment because an image is moving, absurd, offensive or useful for discussion. A share may criticize the post rather than endorse it. Some activity may not correspond to a unique human participant. These possibilities mean that engagement is not a direct measure of religious belief or credulity.
Qualitative responses can still be informative when attributed carefully. A comment may illustrate one interpretation, but the article should not promote it into a population finding. The selection process matters: choosing the most striking responses can exaggerate the apparent intensity or uniformity of an audience.
For religious researchers, a stronger study would distinguish recognition of synthetic content, emotional response, acceptance of the caption and subsequent action. Someone can know an image is invented and still find it meaningful. Someone can believe a false caption without adopting any new religious view. Those differences should guide the questions asked.
A realistic image can be mistaken for a record of an event, particularly when a caption supplies a date, place or named person. Responsible presentation tells the audience what kind of evidence it is—or that it is not documentary evidence at all. The status should be visible where the image is encountered.
A hypothetical church newsletter might use an imagined scene to illustrate hospitality. That can be legitimate if readers understand it as illustration. Presenting the same image as a photograph of the congregation's outreach would create a false claim. The pixels may be identical; the caption and context change the truthfulness of the publication.
AI Faith Monitor applies the same principle to its own visual work. Images should support explanation without fabricating an event, endorsement or eyewitness record. Visual appeal is valuable, but it cannot come at the cost of making the audience misunderstand what happened. This is an editorial responsibility, not merely an aesthetic preference.
Audience growth can create incentives to publish material that provokes quick reactions. Religious imagery may be powerful in that environment because it connects with identity, hope and belonging. That is an analytical explanation of why the subject deserves scrutiny; the motives of a particular account still require evidence.
A publication should also examine its own incentives. A sensational headline about religious deception may reproduce the same attention-seeking pattern it criticizes. The corrective is to make the claim precise, explain the sample and avoid using an extraordinary example as shorthand for an entire community.
For churches, the practical question is whether communications reward accuracy and care or merely reach. A large interaction count does not establish that a post served the congregation well. Leaders can ask whether it helped people understand, invited honest participation and represented those depicted fairly.
A sincere intention to encourage faith does not make an invented event true. Christian truthfulness applies to images, captions and implied endorsements as well as to prose. A community can value imagination while distinguishing it from testimony about what occurred.
Care for neighbors also includes people whose likeness, reputation or vulnerability may be used to attract attention. A realistic depiction of a leader beside words they did not say can mislead even when the message sounds benevolent. The publication should verify the statement rather than assume that a positive purpose makes the attribution acceptable.
Christians may disagree about which sacred images are appropriate. That theological disagreement should not obscure a shared factual question: does the presentation claim to document something, and is that claim supported? Separating the questions allows a more useful conversation than treating every objection as either hostility to art or gullibility about technology.
Truthfulness applies to captions and implied claims as well as images. A church can use illustration while clearly distinguishing it from documentary evidence. Love of neighbor also requires care for people whose likeness or religious trust could be exploited. Christians may disagree about acceptable sacred art while agreeing that an invented endorsement should not be passed off as real.
Begin by recording the exact claim, account, accessible date and source URL. Look for an earlier version and for independent evidence of the purported event or endorsement. Preserve the difference between the earliest version located and a proven original. Do not make authenticity depend solely on visual oddities.
Then identify what remains unknown. An account's purpose, the image's production method and the audience's belief may require different evidence. A responsible report can establish a false event claim without pretending to know every viewer's response. The enduring lesson of this case is that religious imagery belongs in media verification because of what it can be made to imply—not because all synthetic religious art is inherently deceptive.
The study supports findings about its sample. No current platform-wide audit, independent identification of image generators or representative sentiment measure is claimed here.
Religious pictures can become part of an attention economy as well as devotional expression. The evidence must distinguish those uses.
Source: misinforeview.hks.harvard.edu — Essay Summary; Findings; Methods
Explicit religious imagery; creator intention and audience belief require separate evidence. Meaning source: Source: misinforeview.hks.harvard.edu
No representative public-reaction sample is attached. Source authors and project participants are not a population sample.
No systematic media sample is attached to this case.
Truthfulness applies to captions and implied claims as well as images. A church can use illustration while clearly distinguishing it from documentary evidence. Love of neighbor also requires care for people whose likeness or religious trust could be exploited. Christians may disagree about acceptable sacred art while agreeing that an invented endorsement should not be passed off as real.
A published study includes religious imagery within a selected sample of AI-image pages; it is not a census of religious art. Original source text inspected; human review pending
2026-09-25: Added a sourced case with explicit claim boundaries and narrative analysis.