A campaign banner says “30% off until 12 November.” OCR returns “80% off until 17 November.” A single recognition error can turn a polished visual into a false offer. A reliable workflow separates extraction, factual review, and accessibility description.
Three distinct outputs
Extracted text is a draft. Double-check numbers, signs, addresses, product IDs, and small print.
Editorial verification compares the draft with the original artwork and approved campaign terms. OCR cannot know whether the offer is current.
Alt text conveys the image’s purpose to people who cannot see it. It should not reproduce every design element or replace the offer details in the page body.
For a banner showing a blue bottle and “30% off until 12 November,” useful alt text might be “Blue thermal bottle; 30% discount offer until 12 November.” If the offer appears as ordinary text beside the image, keep the alt focused on the bottle and its role to avoid repetition.
Before publishing
- Open the original at full size and extract the text.
- Verify each number against the approved source, not the model output.
- Check the mobile crop.
- Write alt text for the image’s function: product, chart, infographic, or decoration.
- Have another person check the final asset and link; retain the approved version.
Yasnora AI has image and document analysis workflows. Mobile OCR is in beta, and access depends on the account and workflow. This guide does not claim that the product automatically generates alt text; an editor can write it from verified information. See our earlier image quality checklist. Yasnora AI is part of Wicsora. Explore the image workflows.
FAQ
Can I publish OCR output without review? No, especially when it contains prices or dates.
Should alt text repeat every word on the banner? Usually not. Convey meaning and information missing from nearby text.
What about decorative images? If they add no information, use an empty alt attribute.
Does OCR verify the offer is current? No.