AI
The Alt Text That Decides Whether an AI Agent Can Buy From You
A new study puts a number on something accessibility advocates have said for years. When websites are built so screen readers can use them, AI shopping agents work better too. AudioEye, a digital accessibility company, tested six leading AI models across 1,560 shopping tasks on major retail sites, and found that agents completed purchases far more reliably on pages with strong accessibility foundations. The least accessible site in the study saw agents finish only 31 percent of tasks, while its accessible version reached 96 percent.
The gap comes from how agents read the web. People see buttons and pictures. Agents parse the accessibility tree, a structured map of the page built for assistive technology. When a product image carries a clear text description in the code, the agent knows exactly what it is looking at. When that description is missing, the agent starts guessing, and guesses cost extra steps. AudioEye found agents burned 43 percent more tokens on the least accessible version of a site, which makes every agent driven transaction more expensive to run.
The methodology earned credibility. AudioEye ran the study both with its own accessibility fixes loaded and with them removed, across retail, travel, and direct to consumer sites. The performance difference held in both directions, which suggests the effect comes from the page structure itself rather than any vendor trick. Six models, ten attempts per task, and the pattern stayed consistent.
Industry voices are connecting the dots. AudioEye's chief executive told TechCrunch that agents rely on the accessibility tree rather than the visual page, and that readiness for AI commerce starts with clean semantic code. The company is now building an accessibility product aimed at agents themselves, designed to keep automated buyers moving through checkout. Meta's Muse launch this month, which lets AI assistants shop on a user's behalf, gives the finding immediate commercial weight. Every major platform now wants agents spending money, and broken pages keep that revenue from flowing.
The business case is now straightforward. Accessibility improvements that once looked like compliance overhead now read as revenue infrastructure for the agent economy. Retailers that clean up their page structure get shoppers who arrive as software, completing purchases around the clock with full carts and clear intent. The checkout page is becoming an API for machines, and the stores with the clearest code win the most agent traffic.
For site owners, the takeaway is simple. Write the alt text, label the buttons, and structure the page for the readers who need it most. The agents will follow the same path, wallets open. The web's oldest good habit just became its newest growth lever.
Quick answers
What is this story about?
A new study puts a number on something accessibility advocates have said for years. When websites are built so screen readers can use them, AI shopping agents work better too. AudioEye, a digital accessibility company, tested six leading AI models across 1,560 shopping tasks on major retail sites, and found that agents completed purchases far more reliably on pages with strong accessibility foundations. The least accessible site in the study saw agents finish only 31 percent of tasks, while its accessible version reached 96 percent.
Why does this story matter?
For site owners, the takeaway is simple. Write the alt text, label the buttons, and structure the page for the readers who need it most. The agents will follow the same path, wallets open. The web's oldest good habit just became its newest growth lever.
Sources
- PR Newswire via Morningstar on AudioEye study on AI agent task completion
- Agile Brand Guide on AudioEye study breakdown
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.