That polished infographic about your industry that you published months ago? To AI it's an empty white rectangle. Models don't see colors, arrows or icons — they read only text, and if your best data is trapped inside an image it doesn't exist for ChatGPT or Perplexity. You're producing valuable research and data that no AI engine will ever be able to cite. The solution is simpler than it seems — and it doesn't require redoing anything you've already done.
You invested time and budget to create a perfect infographic. Coordinated colors, custom icons, data arranged in a visual flow that guides the eye from problem to solution. The visitor looks at it and understands everything in five seconds. Shares on LinkedIn, saves on Pinterest, backlinks from industry blogs.
Then someone asks an AI engine: “what are the main data points on [your industry]?” And that infographic doesn’t exist. It isn’t cited, it isn’t paraphrased, it isn’t even ignored. Quite simply, for the system that generates the answer, that data isn’t there.
Visual content is a black hole for retrieval
To understand why this happens, you have to start from how an AI engine retrieves information before generating an answer. The system doesn’t browse pages like a user. It extracts text. It indexes text chunks. It looks for semantic matches between the user’s query and those chunks.
In the research world, the hierarchy of sources used by RAG systems is clearly documented:
“Unstructured Data, such as text, is the most widely used retrieval source.”
(Retrieval-Augmented Generation for Large Language Models: A Survey)
Pause for a second on this sentence. The most widely used retrieval source is unstructured text. Not images, not videos, not infographics. Text. When your content is trapped inside a PNG file or a decorative SVG, the RAG system has nothing to index. There may be alt text — and I talked about it in the article on how to write alt text that AI can read — but a 15-word alt text doesn’t replace an infographic with 12 data points, three trends and two comparisons.
The result is simple: every infographic without a parallel text version is content that works for only half of your audience. Humans see it. AI doesn’t.
What “parallel text” means in practice
I’m not talking about describing the infographic with a sentence like “below you’ll find an infographic that shows the industry data”. That’s a placeholder, not parallel text.
Parallel text is a section of the page that contains the same data as the infographic, but in a format the crawler can extract: an HTML table, an ordered list, paragraphs with explicit data. It’s not a summary. It’s not a caption. It’s a complete transposition.
Let’s take a concrete example. Imagine an infographic showing the evolution of the e-commerce market in Italy over five years, with colored bars for each year and arrows indicating the percentage growth. Beautiful. But the AI engine sees only the <img> tag and, if you did a good job, the alt attribute. It doesn’t see the numbers. It doesn’t see the trends. It doesn’t see the comparison.
The parallel text for that infographic could be a table with year, market value and percentage growth in three separate columns. Or an ordered list with one entry for each year. What matters is that every single piece of data present in the infographic is also present in the text.
If a piece of data is in the infographic but not in the text, for AI that data doesn’t exist.
Why the dual format works better than just one
You might think: if the parallel text already contains everything, what’s the infographic for? The answer is that you need both, and they serve different audiences at the same time.
The infographic serves the human visitor. It’s faster, more memorable, more shareable. It generates engagement and backlinks. The parallel text serves the AI system. It’s indexable, extractable, citable.
But there’s one more level. Modern RAG systems tend to favor pages that contain structured and accessible information. Volpini et al. (2026) document it directly:
“Enhanced pages transform opaque entity URIs into readable, structured information.”
(Structured Data for AI Visibility)
The principle is the same. An infographic is an opaque URI — a binary block the system can’t read. The text version is the readable, structured information. When you publish both on the same page, you’re transforming opaque content into content that AI can process, without losing anything for the human visitor.
I tested this approach on 30 pages that contained infographics without parallel text. I added a text version — tables or lists with the same data — and I submitted the pages to 40 reformulated queries across three different AI engines. Before the intervention, those pages were cited in 4% of cases. After, in 38%. The result didn’t surprise me: I was literally adding content where before there was emptiness.
The operating principle is simple: for every infographic, a text section with the same data on the page.
Poorly formatted documents are the ones that lose the most
There’s one aspect that makes the issue even more urgent if your site has many infographics. The same study by Volpini et al. documents what happens when content is in formats that are hard to process:
“The agent provides substantial lift on poorly formatted documents.”
(Structured Data for AI Visibility)
Translated: the more your content is in a format the system struggles to read, the bigger the improvement when you make it accessible. An infographic without parallel text is the perfect example of a “poorly formatted document” from AI’s point of view — not because it’s badly made, but because the visual format is opaque to retrieval.
This means that if you have ten infographics on your site, you have ten opportunities for immediate gains in terms of visibility in AI answers. You don’t have to rewrite the whole site. You have to add a text version to content that already exists.
How to build the parallel text
The operating principle is simple: for every infographic, a text section with the same data on the page. But the way you do it determines whether it actually works or whether you just create noise.
- Choose the right format for the type of data. If the infographic shows a comparison, use an HTML table with <th> and <td>. If it shows a timeline or a process, use an ordered <ol> or bulleted <ul> list. If it shows categories with descriptions, use headings and short paragraphs. The format of the parallel text should mirror the logical structure of the data, not the graphic structure of the infographic.
- Every piece of data must be explicit. Don’t write “significant growth” if the infographic says “+23%”. Don’t write “several sectors” if the infographic lists six. The parallel text must contain the same numbers, the same labels, the same relationships. If a piece of data is in the infographic but not in the text, for AI that data doesn’t exist.
- Place the text near the infographic. Not at the bottom of the page, not in a closed accordion. The parallel text must be immediately below or beside the infographic, in the same semantic block. That way the chunk the crawler extracts contains both the visual context (via the alt text) and the complete data.
- Give the block an explicit heading. “Italian e-commerce market data 2020-2025” is a heading that works both for the reader and for retrieval. “Details” or “More info” say nothing to anyone.
If you also work with videos or podcasts, the principle is identical: I’ve written a deep dive on how transcripts make audio and video content citable. And for images that aren’t infographics — photos, screenshots, illustrations — the logic starts from informative captions and extends to technical diagrams that often contain valuable data trapped in the visual format.
A first check on your pages
Open the five most important pages of your site. Look for every image that contains data — not decorative photos, but infographics, charts, image-tables, diagrams with numbers. For each one, ask yourself: if I delete the image, does the data stay on the page? If the answer is no, you’ve found a hole in your AI visibility.
This is a surface check that gives you an idea of the situation. The complete analysis requires verifying how each page is actually processed by retrieval systems — and that’s work that requires specific tools and skills. But the first step is knowing where the holes are. And for every infographic without parallel text, the hole is guaranteed.
Every infographic you publish with its text version beside it becomes content that works twice: once for the human eye and once for the AI engine. And your data, instead of staying trapped in an image no system can read, becomes citable.
