Originally published on The AI Prism
The Post That Hit a Nerve
In early 2025, developer Nelson Figueroa published a short, sharp complaint that resonated far beyond his own readership. His post, “AI-Generated Images Discourage Me From Reading Your Blog,” climbed to 716 points on Hacker News, a score that signals the argument touched a shared, unspoken frustration rather than a niche gripe. His thesis was personal and blunt: when he sees an AI-generated image in an independent blog, he begins to wonder whether the text was generated too. The visual becomes a tell.
Figueroa is clear about where his disappointment lands. He says he expects polished, synthetic imagery from corporate blogs but not from indie writers. “I’d rather see a shitty Microsoft Paint drawing,” he writes, “as opposed to some AI image.” The sentiment is not really about aesthetics. It is about authorship. A clumsy hand-drawn diagram signals a human was in the room. A flawless, generic render signals the opposite.
The Hacker News thread that followed ran long and divided. Some readers agreed that AI headers feel like a cheapening of the medium. Others defended image generation as a neutral tool, no different from stock photography. What is striking is how few people were neutral. The image at the top of an article now arrives with baggage, and that baggage is exactly what this publication has been handing its readers without comment.
The Uncomfortable Mirror
Here is the part this publication has to say out loud. TheAIprism runs AI-generated header images and AI-generated section dividers on every article we publish. We have done this since launch. The piece you are reading right now will almost certainly ship with a synthetic image above it, courtesy of the same pipeline that produced the one you scrolled past to get here.
So when we point at Figueroa’s argument, we are pointing at ourselves. The critique he levels at indie blogs is one we cannot deflect. We are the corporate-scale example he did not expect from individuals, except we are a small editorial team operating at a pretend scale. That tension is the reason this article exists. We wanted to interrogate the practice rather than quietly continue it under the cover of routine.
Being self-aware does not automatically make a practice right. It makes it harder to ignore. The rest of this piece tries to weigh the case for and against the images we already publish, with evidence rather than vibes, and to be honest about which side the evidence falls on when it comes to the people we most want to reach.
What the Evidence Actually Says About Trust
The intuition that AI involvement lowers trust is not just anecdotal. A 2024 study from the University of Kansas found that when AI contribution was mentioned in a news byline, readers rated both the source and the author as less credible, even when they did not understand the extent of the AI’s role. The effect held across political leanings and across levels of AI familiarity. The mere label “AI” shifted perception downward before a word of the article was judged on its merits.
That study measured text, not images. But the mechanism is adjacent. Readers build a mental model of who is speaking to them. Any signal that a machine stood in for a human erodes the assumed authenticity of the whole artifact. An AI header image is, in effect, a permanent byline note that says a machine helped make this look finished, and readers appear to read that note whether or not we intended them to.
We should be careful not to overstate. The Kansas findings are about news and credibility judgments in a controlled setting, where participants were primed to evaluate trustworthiness. Blog reading is looser, more voluntary, and more forgiving of surface choices. Still, the direction is consistent with a broader pattern across media research: people penalize content the moment they suspect it was not made by a person who cared about the specific thing they were making.
The relationship is not simple, and a German newspaper survey adds an instructive wrinkle. Researchers found that exposure to AI-driven misinformation lowered overall trust in news, yet it also raised engagement with trustworthy outlets, as readers hunted for sources they could still believe. The pattern hints that AI imagery may simultaneously repel and intrigue, pushing skeptical readers toward publishers they perceive as honest. For a small outlet, that cuts both ways: the same trend that punishes us can also reward us, if we earn the honest label.
The Credibility Gap in the Pixels
A 2025 study published in the Journal of Imaging asked participants to rate AI-generated versus human-made images on credibility using a five-point scale. The result was unambiguous. Human-created images scored a mean of 4.199; AI-generated images scored 3.527. The gap was statistically significant. Participants simply trusted the human-made pictures more, even in cases where they could not reliably tell the two apart.
The inability to distinguish is itself the problem. People cannot always identify synthetic imagery, but they report a vague unease that something is off, a smoothness or a wrongness that does not resolve. That unease does not stay inside the frame of the image. It bleeds into the surrounding text. If the header looks like it came from a template farm, the argument beneath it feels like it might too, and the reader has no obvious reason to separate the two.
Notably, the credibility penalty was slightly larger for participants without visual-professional backgrounds. In other words, everyday readers, the exact audience of an independent blog, were the most likely to downgrade AI imagery. The people we most want to reach are the people most primed to distrust what we put at the top of the page, which is the opposite of the reassurance a hero image is supposed to provide.
Banner Blindness, Now Self-Inflicted
There is a second, older reason to worry about decorative imagery, and it predates AI entirely. The term “banner blindness” was coined in 1998 after usability tests showed that web visitors consciously or subconsciously ignore anything that looks like an ad. Decades of follow-up work from the Nielsen Norman Group confirms the pattern persists: users dodge content that resembles advertising, sits near advertising, or occupies the traditional banner slot at the top of a page.
Modern AI blog headers share the visual grammar of banner ads. They are wide, polished, detached from the specific argument, and optimized to look professional rather than to communicate anything particular. A reader who has spent twenty years learning to skip the top strip of a page will skip our hero image by reflex, and with it the trust-building moment we hoped that image would provide. We are fighting a reflex we helped train.
This is the irony. We add images to make articles feel richer and more inviting. In practice, we may be adding the exact element readers have learned to mute. The image becomes wallpaper, and wallpaper does not earn attention; it quietly taxes it. Every forced header is a small withdrawal from a reader’s patience before the first sentence has had a chance to earn it back.
The SEO and Engagement Math
So why do we do it? The honest answer is that the incentives point the other way. Search engines and social platforms reward visual content. Articles with relevant images earn more shares, and image-rich pages tend to perform better in discovery surfaces where a thumbnail is the only thing a potential reader sees. A distinctive header is also a branding asset; it makes a publication look intentional rather than like a bare text feed competing with a thousand others.
There is real data behind the engagement case. Pages with at least one image regularly show higher average time-on-page and lower bounce rates than text-only equivalents, and social cards built from article imagery drive measurable click-through. For a small publication fighting for distribution, those numbers are not trivial. They are, in many cases, the difference between being read and being invisible to the people who would benefit from the reporting.
But the engagement math and the trust math pull in opposite directions. An image can win the click and then quietly undermine the read that follows. We have been optimizing for the first half of that sequence, the scroll-stop and the open, and hoping the second half, the actual reading and the returned visit, would take care of itself. The evidence suggests it does not.
The mechanics reward relevance, not authenticity. Search systems surface images through alt text, filenames, and surrounding context, so a well-labeled synthetic graphic can rank as easily as a photograph. The platform does not care who held the camera. That is precisely the trap: the optimization target is legibility to a crawler, not honesty to a human, and the two have quietly diverged in the dashboards we watch.
Why We Used AI Images Anyway
The specific choice of AI imagery, rather than photography or bespoke illustration, came down to three pressures. First, cost. Commissioning or licensing unique visuals for every article is a real expense for a small team with a fixed budget. Second, speed. A generated header takes minutes, not days, and keeps a publishing cadence intact when the news moves faster than a designer’s queue. Third, consistency. AI pipelines produce a uniform look that reads as a deliberate brand.
None of those reasons is about the reader. They are about us: our budget, our schedule, our aesthetic comfort. That is worth stating plainly, because it exposes the trade we made. We spent reader trust to buy operational convenience, and we did it quietly, article after article, without once asking whether the reader noticed or resented it. Convenience for the publisher is not the same thing as value for the person on the other side of the screen.
Scale makes the math harder to escape. A publication producing several articles a week cannot commission a unique illustration for each without a dedicated art function, and most independent outlets do not have one. The AI pipeline looked like the only way to keep visuals present at all, rather than a choice among equals. That framing deserves scrutiny, because “only option” is often a story we tell ourselves to avoid a harder one about what we are actually for.
This is also where our own coverage circles back on us. The same models that paint our headers were trained on data scraped without consent, a story we documented in “AI Companies Are Shredding Rare Books.” The imagery is not merely a trust signal; it is a product of the extraction economy we have criticized elsewhere, and running it on every page is a small contradiction we have been content to leave unexamined.
The Reading Experience We Trade Away
Step back from the metrics and consider what the reader actually experiences. They arrive at an article. Before a single sentence, they meet a glossy, generic scene: a glowing brain, a neon cityscape, two hands almost touching light. They have seen this image, or its cousin, on forty other sites this month. Their brain files it under “decorative” and moves on, and the moment a real visual could have built a bridge is already gone.
What they lose is the chance for a visual that actually helps. A genuine diagram, a screenshot of the tool under discussion, a photo of the real thing, these earn attention because they carry information the words alone cannot. We swapped those for a placeholder that carries none. In chasing the appearance of professionalism, we gave up the substance of it, and the reader is the one left to infer that the rest of the page might be equally hollow.
The cost compounds with volume. When every article opens the same way, the publication starts to feel like a content machine, which is precisely the impression Figueroa says drives him away from indie blogs. The repetition trains readers to expect sameness, and sameness is the enemy of the independent voice we claim to offer. The image meant to signal care instead signals scale, and scale is what the reader came to escape.
There is also a hierarchy of first impressions, and the header usually wins it. The headline is meant to do the work of framing, but a large image arrives faster than a line of text, setting the emotional temperature before the reader has parsed a single word. When that temperature is generic, it flattens the specificity the headline was trying to earn. We let the least informative element set the mood for the most informative one, and then wonder why the piece feels thin.
What We Are (and Aren’t) Changing
Writing this has forced a decision. We are not, today, deleting our image pipeline. The engagement and SEO case is real enough that an abrupt removal would cost reach we use to surface harder reporting, including our work on “what is actually happening to jobs,” which depends on being found by the people it affects. But we are changing the default, and the change is not cosmetic.
Going forward, we will reserve AI imagery for cases where it adds something a reader can use, and we will pair it with clearer labeling so the synthetic nature of a visual is not a hidden tell waiting to be discovered and resented. Where a real screenshot or a simple hand-drawn diagram does the job, we will use that instead. The goal is to stop treating the hero image as mandatory and start treating it as optional, earned, and honest about what it is.
We may also begin surfacing this very tension to readers directly, the way Figueroa did, because the conversation itself is part of the trust we owe. A publication that admits its own contradictions out loud is, at minimum, a publication a reader can believe is staffed by people who notice things, including their own mistakes. That belief is worth more than any thumbnail.
An Open Question, Pointed Back at Ourselves
There is no clean answer here, only a balance we have been tipping without noticing. Images help us be found; they also help readers decide we are not worth the time. The data says trust falls when machines are in the loop, and that everyday readers are the most unforgiving judges of synthetic visuals. We have read that data, and we have kept publishing the images anyway, which is the part that should make us uncomfortable.
So we will leave the question where it belongs, with us, and with you. If a header image makes you wonder whether a human wrote the words beneath it, is the extra click we bought with that image worth the doubt it plants?
References
• AI-Generated Images Discourage Me from Reading Your Blog — Nelson Figueroa (2025)
• Study Finds Readers Trust News Less When AI Is Involved — University of Kansas (2024)
• Journal of Imaging: Image Processing and Visual Attention (2025)
• Banner Blindness, Old and New Findings — Nielsen Norman Group
• AI-Generated Images Discourage Me from Reading Your Blog — Hacker News discussion
The post AI-Generated Images Are Killing Blog Reading — Including Ours appeared first on The AI Prism.
Cross-posted from theaiprism.com — Cutting Through the AI Noise 🧊
Top comments (0)