6 min read
Here is an uncomfortable question for anyone who sells anything online. If an AI shopping agent visited your website right now, on a mission to buy from you, could it actually finish the job? Not “would it find you” and not “could it pay,” but could it understand your offer, pick the right option, and complete the purchase without getting confused and giving up? A new research paper suggests that for most ordinary websites, the honest answer is no. And this is no longer a hypothetical. AI referred traffic to United States retail sites jumped roughly 805 percent year over year on Black Friday 2025, and a growing share of shoppers now start with a chatbot instead of a search bar.
This is the next layer of a shift solo owners have been hearing about for months. First your customers started asking AI for recommendations. Then the payment rails to let agents spend money started falling into place. The missing piece was whether an agent could operate a normal website at all. This week we got hard data on that, and it should reshape how you think about your own site. Here is what the research found and what to do about it.
The Study That Put Agents to the Test
On July 13, researchers Said Elnaffar and Farzad Rashidi published a preprint titled Designing Agent-Ready Websites for AI Web Agents. Their setup was clean. They built two versions of the same online store with identical products, prices, stock, and shopping tasks. One version was designed the normal way, for humans with eyes and a mouse. The other added clearer structure, obvious action cues, supporting evidence, and signals showing whether information was current. Then they turned three different browser agent models loose on five tasks across 300 total runs.
The gap was stark. The ordinary website let the agents succeed 49.3 percent of the time. The agent friendly version pushed that to 89.3 percent. Half finished attempts dropped from 43 down to 3, and the agents needed fewer steps to get things done. It is a controlled prototype and a preprint, not peer reviewed and not the final word, so hold it loosely. But a jump from a coin flip to nearly nine in ten is far too large to wave away.
The Money Rails Are Arriving Too
This research does not exist in a vacuum. Also in mid July, the Linux Foundation announced the operational launch of the x402 Foundation, an effort to standardize payments that AI agents can make directly through ordinary web requests. The list of organizations backing it is the real headline: American Express, AWS, Cloudflare, Coinbase, Google, Mastercard, Shopify, Stripe, and Visa, among dozens more. When the card networks, the processors, and the biggest commerce platforms all sit at one table, they are not debating whether agents will buy things. They are wiring up how.
Put the two stories together and the picture is clear. The infrastructure for machines to shop and pay is being built right now, and the sites that machines can actually navigate will get the orders. This is what people mean by agentic commerce, and it is coming to small sellers, not just giant retailers.
How to Make Your Site Agent Ready
The good news is that almost everything that helps an AI agent also helps a confused human, so none of this is wasted effort. The researchers grouped the fixes into three simple questions:
- Can it understand you? Are your products, prices, and page structure clear?
- Can it act? Can the agent tell what to do next and complete it without guessing?
- Can it trust the page? Is the information verifiable and clearly current?
Translate those questions into a practical checklist you can run against your most important offer page:
- Say what it is in the first line. The opening sentence should state exactly what you sell and who it is for, no clever mystery.
- Put prices and details in real text, not baked into an image a machine cannot read.
- Use one consistent name for each product or package everywhere it appears.
- Make every button describe its action, so “Book a strategy call” beats a vague “Get started.”
- Show your terms and availability plainly, including dates, stock, or delivery windows.
- Back up your claims with named evidence, and stamp time sensitive information with when it was last updated.
Then run the test that actually matters. Open an AI browser agent, point it at your page, and give it a real buying task. Do not ask “is this page good,” because that just fishes for compliments. Instead say: “Find the price, tell me exactly what is included, identify who this is for, name the proof behind the offer, and complete the next step without submitting anything. List every point where you had to guess.” Every guess the agent reports is a place a paying customer is also getting stuck. That single exercise is worth more than a month of opinions about your website.
Why a Solo Seller Should Care Before the Giants Do
It would be easy to file this under “hype for big ecommerce brands.” That would be a mistake, and here is why. Large retailers move slowly. They have committees, brand rules, and legacy systems that make even small site changes take months. You can rewrite your most important product page this afternoon. For once, being tiny is the advantage, because agent readiness rewards speed and clarity, not budget.
Consider a one person consultancy whose services page currently hides pricing behind “contact us for a quote” and buries the offer in three paragraphs of story. An AI agent sent to compare consultants under a budget will simply skip it, because it cannot verify the price or the fit. A competitor who lists a plain starting price, a clear description of who it is for, and a labeled “Book a call” button gets shortlisted instead. Same expertise, very different outcome, decided entirely by which page a machine could read.
The common worry is that this is just search engine optimization wearing a new hat. It is related, but it is a distinct layer. Being found in an AI answer gets you onto the list. Being agent ready is whether you survive the next step, when the agent tries to understand and act. You can win the first and lose the second, which is the worst of both worlds: visible but unusable.
Three Moves to Make This Month
You do not need to overhaul everything. Start here:
- This week: Run your single most valuable page through an AI browser agent with the buying task above and write down every guess it makes.
- This month: Fix the top three ambiguities the agent found, usually a hidden price, a vague button, or a missing detail.
- Ongoing: Add a “last updated” date to time sensitive pages and keep your product names identical across your whole site.
Will the Agents Add You to the Cart
The web is quietly gaining a new kind of visitor, one that does not scroll, squint, or forgive a confusing layout. It reads, decides, and either buys or moves on. The businesses that thrive in agentic commerce will not be the ones with the biggest marketing budgets. They will be the ones whose offers are so clear that a machine can act on them in seconds. That clarity has always helped your human customers too, which means there is no downside to starting now. So when the next AI agent comes shopping in your category, will your site make it easy to say yes, or easy to skip you? If you want a running guide to these shifts, that is exactly what we track at SoloAITool.
