This seems like it would absolutely wreck the experience for people using screen readers.
It only works as a decoy when you give it to the LLM as an image. As html it appears like normal human friendly text, which is what screen readers use to interpret the text.
How? AFAIK screen readers don’t do OCR.