AI Images on Personal Blogs Make Readers Doubt the Writing

You open a personal blog post that a friend sent you. Before the first sentence there is a wide, glossy illustration: a figure at a desk bathed in teal light, a suspiciously smooth coffee cup, hands that are almost right. You close the tab without reading a word. Nothing in the text has changed. So why should a picture at the top of a page change what you believe about the sentences underneath it?

That reaction — and the reasoning behind it — is the subject of a short post by an independent blogger who describes a "growing hatred" for AI-generated images on blogs, and says plainly what the picture does to him: it makes him wonder whether the writing is machine-made too. He asks people who run personal blogs to stop using them. He would rather look at a bad drawing made in Microsoft Paint.

The images in question come from text-to-image models: software that takes a written description — "a developer at a desk, warm lighting, cinematic" — and returns a finished picture in a few seconds, at effectively no cost. They belong to the same family of tools as the large language models that produce fluent paragraphs on request, and both are commonly grouped under the label generative AI. Until recently, a custom illustration on a personal blog meant someone had drawn it, paid for it, or spent an afternoon in an image editor. Now it means someone typed a sentence.

The picture is the only part you can inspect

Here is the awkward position a reader is in. The thing they actually care about — did a human being think these thoughts and write these words — is the thing they cannot check. Machine-written prose can be grammatical, organised, superficially reasonable. There is no watermark on a paragraph. The writer knows exactly how the post was made; the reader knows nothing and has no way to find out. Economists call this an information asymmetry: one side of an exchange holds facts the other side needs and cannot obtain.

When people cannot verify the thing they care about, they fall back on the things they can observe. A visible production choice becomes a proxy for an invisible one. And on a blog post, the picture at the top is the most inspectable choice there is. You do not have to read it. You can see in half a second whether someone made it, chose it, or generated it.

One signal, applied to the whole page

So the image gets read as evidence about the text. If the author reached for an automated tool for the part you can see, the reasoning goes, what did they reach for on the part you can't? The blogger states this directly — the image makes him wonder if the writing is AI-generated "to some extent." One choice, made about a decorative element, spreads outward and colours the reader's judgement of the entire work.

This is a leap, and it is worth being honest that it is a leap. But it is not an irrational one. It is what people do at every point of sale where quality is hidden: they judge the restaurant by the state of the menu, the contractor by the state of the van.

Why an indie blog is held to a standard a corporate one is not

The same illustration on a company's product page provokes none of this. The blogger says so himself: he expects it from corporate blogs, just not from independent ones. That asymmetry is the crux of the whole thing, and it follows from what a personal blog is for.

A corporate blog exists to rank in search results and explain a product. Nobody opens one expecting an individual sensibility; the content was already the output of a process involving a marketing team, an SEO brief and a review cycle. There is no human voice being promised, so none can be broken.

A personal blog promises exactly the opposite, and it is the only thing it has to offer. Its value proposition is that the words came from one specific person with one specific point of view — the thing the blogger points to when he says that whatever else you might criticise about his site, you know for a fact you are getting the thoughts of a real human being. That is the product. Automation does not merely dilute it; it negates it.

Which is why the crude Microsoft Paint drawing wins. Judged as an image, it is worse in every measurable way — clumsy lines, wrong proportions, no polish. But it carries something the generated illustration cannot: proof that a person sat there and made it. The reader is not grading the picture. They are reading it as a marker of human effort, and on that scale the bad drawing scores higher than the good one. What looks like snobbery about tools is really a reader trying to answer the only question that matters to them — is anyone home — using the only evidence on offer.

The suspicion is not evidence

There is a gap in this reasoning that nobody has closed. No one has demonstrated that writers who generate their images also generate their sentences. It is entirely possible to labour for a week over a piece of prose and then spend ten seconds producing a header image because you cannot draw and do not want an empty space at the top of the page. The inference runs on the visual alone.

But an inference does not have to be correct to be costly. The reader who bounces at the illustration never reaches the paragraph that would have proved them wrong. For anyone writing under their own name, that is the practical shape of the problem: the signal you send about your images is being read as a signal about your mind, by people you will never hear from, before they have read a word you wrote.