Alt text and AI search: how ChatGPT reads your images

For fifteen years, the reader you optimized images for was Google's crawler. Now there's a second one — the model behind ChatGPT, Perplexity, Google's AI answers, and Copilot. Your customers are already asking these tools for recommendations and "best tool for X," and the tools answer from the pages they've read. The catch is a familiar one: like Google, and like a screen reader, an AI can't actually see your images. It reads words.

So the question marketers keep asking — "does my alt text help me show up in AI search?" — has a real answer. Yes, more than you'd expect, in a specific way.

AI reads your page as text — alt text included

The systems behind AI answers send crawlers across the web — GPTBot, ClaudeBot, PerplexityBot, Google's own — and they parse your HTML. The alt attribute is text, sitting right there in the markup. When a model ingests your page, your alt text comes along with it, becoming part of what the model understands your page to be about.

For images, your alt text is often the only thing the AI can see

This matters more for images than for body copy. A language model usually doesn't look at your photo; it reads the description attached to it. Miss that description and the image is invisible — the model skips it. Include it and the model can understand the image, use it to grasp the page, describe it accurately, and surface your content when someone asks a related question. Your alt text is, quite literally, how an AI sees your image.

Which means the difference between weak and strong isn't cosmetic — it's whether the model gets a fact it can use:

The strong version hands the model something quotable. The weak one hands it a shrug.

It's called GEO, and it rewards a habit you already have

Marketers have a name for this now: generative engine optimization, or GEO — making your content easy for AI answer engines to find, understand, and cite. Good alt text is one of the cheapest GEO moves available, because it pulls triple duty. The same eight-to-fourteen words that help a screen reader user, and help Google image search, now also help an AI describe and recommend what you made. Write it once; three audiences read it.

Two honest caveats

Alt text is a contributing signal, not a magic switch — your main copy still does most of the work in whether an AI cites you at all. And some engines are starting to run vision models that can look at images directly. But even then, your alt text is the authoritative, intended description of what the image is meant to convey, and it's the reliable text signal that always gets read. It's upside with almost no downside.

The hard part was never knowing this. It's doing it across hundreds of images without losing an afternoon to the download-open-copy-paste shuffle. That's a tooling problem, not a willpower one — and the writing itself takes seconds once it's in front of you.