Can AI search engines read text inside images on my site?

AI search engines answer from text they can fetch. A crawler requests your page, stores what comes back, and the engine later pulls passages from that store into an answer. If your price list is a PNG, the crawler stores a file name and maybe some alt text. It does not store the numbers in the picture.

Three facts drive this:

• No operator documents image text extraction. OpenAI's crawler page describes what GPTBot, OAI-SearchBot and ChatGPT-User are for. Anthropic, Perplexity and Meta publish similar pages. As of October 2026, none of them says the crawler reads text inside images for search.
• Most AI crawlers read raw HTML only. Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but "don't execute them. They can't read client-side rendered content." If they don't render scripts, there's no reason to assume they run text recognition on images.
• Engines quote sentences. The one positive on-page signal in the evidence is that a specific fact, stated plainly on a fetchable page, gives the engine a sentence it can quote. A fact locked in an image is not a sentence.

So the practical answer is simple. Anything you want AI answers to state about you should exist as text in the HTML your server sends, before any JavaScript runs. That applies most to prices, hours, addresses, plan limits and one-line product definitions. Our guide to making SaaS pricing pages readable to AI covers the most common case.

Not every crawler behaves the same way. The split that matters is between crawlers that render pages and crawlers that don't.

The evidence base reports independent re-testing in June 2026 that found none of the non-Google crawlers above render JavaScript. Googlebot and AppleBot are the exceptions.

Google is the one place to be careful. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link in AI Overviews or AI Mode, and that there are no extra technical requirements. That tells you text in the indexed page is what counts. It does not tell you Google's AI features quote text from inside your images, so don't rely on it.

Images and JavaScript often fail together. A carousel that loads banner images through a script is doubly hidden from a non-rendering crawler. If your site is built on a JavaScript framework, check whether a headless CMS hides your content too.

Be clear about the shape of this claim. It rests on absence: no operator documents reading image text for retrieval. That is not proof that no engine ever does it. It is a reason not to bet a price or a policy on it.

What the evidence does support:

• Access comes first. A critical survey of 45 GEO studies (arXiv, July 2026) found that already-retrieved content can change how it is cited, but no technique shows a stable effect on getting discovered. The levers that hold up are access levers, like making content fetchable as text.
• Rendering is the strongest finding. Content that appears only after JavaScript runs may be invisible to AI crawlers. Image-only text sits in the same bucket: present for humans, absent for the crawler.

What the evidence does not support:

• Alt text as a citation lever. Alt text is an HTML attribute, so a raw fetch returns it. Nothing in the evidence shows engines weight it or cite from it. Write it to describe the image for people using screen readers, and put the fact itself in body text.
• Any number for how often image text gets missed. Nobody has measured it. We won't invent one.

Some formats look like images but behave like text, and some look like text but aren't.

• Inline SVG with real text elements. The text is in the markup, so a raw HTML fetch returns it. An SVG saved as a separate image file with text converted to outlines does not carry readable text.
• CSS background images. Text baked into a background image is pixels, with no alt attribute at all.
• Canvas and charts. A chart drawn with a script on a canvas element has no text in the HTML. Add a short sentence stating the key number next to it.
• Scanned PDFs. A PDF that is a scan is an image in a wrapper. See whether AI engines can read PDFs for the details.
• Suppression tags. Check that the text you add isn't inside an element marked data-nosnippet, which Google and Bing document as keeping content out of AI summaries. Read how hiding text from AI summaries works before using it on purpose.

One thing not to do: add a text version that's hidden from people and shown only to AI crawlers. That is cloaking. Hidden text aimed at AI systems is the same family of tactics covered in our guide to prompt injection and hidden text, and it is a risk, not a fix.

Fix this page by page, starting with the facts AI answers most often get wrong about you.

1. List your load-bearing facts. Prices, plan limits, hours, locations, founding year, and the one-line description of what you sell.
2. Find where each one lives. If a fact appears only in a banner, infographic, menu photo or screenshot, mark it.
3. Fetch the raw HTML. View source, or run the free JavaScript rendering check, and search for each fact. If it's missing from the raw response, a non-rendering crawler can't see it.
4. Restate each fact as visible text. Keep the image if it helps people. Add a plain sentence or table cell beside it, rendered on the server.
5. Confirm crawlers can reach the page. Text only helps if the crawler gets a 200. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

None of this guarantees a citation. It removes a block you control, which is the part of AI visibility the evidence actually supports.