Skip to content
Book a free consult
Interface sounds
By Kishan Thankey 7 min read UXAI ProductStrategy

The Agent-Ready Website: What Breaks When Your Visitor Is Software

A second kind of visitor is reading your site: one that does not scroll, does not wait for your scripts, and never emails to ask what something costs. It takes a record and leaves. Here is what it can and cannot see, why most sites fail at the same step, and the fixes that pay off whether or not the agents ever arrive.

One product page read by two visitors: a person who scrolls, waits and leaves with an impression, and an agent that extracts a record and leaves with most of its fields marked unknown.
Contents

Your website was designed for someone who scrolls. Who hovers over a spec to see more, waits half a second for the price to fade in, opens the returns PDF if it matters enough, and emails you when something is unclear. Every good instinct in modern web design assumes that visitor.

A second kind of visitor is now reading the same pages. It does not scroll, does not hover, does not wait for your scripts to finish, and will never send you a question. It makes one pass, fills in the fields it can, and leaves with whatever it got. If the fields are empty, you were not rejected. You were unreadable.

What actually changed, and how much

The plumbing arrived faster than the behaviour. Stripe and OpenAI published the Agentic Commerce Protocol in September 2025, an open Apache-licensed standard for checkout flows between buyers, agents, and businesses, launched alongside ChatGPT Instant Checkout. Google announced the Universal Commerce Protocol on 11 January 2026, co-developed with Shopify, Etsy, Wayfair, Target, and Walmart. Two rival standards for the same idea, both open, both shipped inside twelve months.

The behaviour is further out, and the honest version of this argument depends on saying so. Bain’s December 2025 forecast puts agentic commerce at 300 to 500 billion dollars in the US by 2030, which is 15 to 25 percent of e-commerce, and notes that most consumers are still uncomfortable letting AI complete a purchase end to end. What is already common is the step before the purchase: Bain puts 30 to 45 percent of US consumers using generative AI for product research today.

That is the useful shape of it. Software is already doing the shortlisting, and shortlisting is where you get eliminated.

What a person forgives and software cannot

A person who cannot find your price scrolls, then guesses, then asks. An extractor records price: unknown and moves to a competitor who stated it. The gap between those two behaviours is where almost all of the damage happens, and it is invisible in your analytics because nothing was clicked.

Five ways a page is built and what an extractor gets from each: price inside the hero image, a greyed-out button for stock, a policy PDF, specifications that load on scroll, and a four-step booking flow, each producing an unknown field.

None of those five is a design failure by human standards. A price set in the hero image looks better than a price in a paragraph. Loading the spec table on scroll makes the page faster. A PDF policy is what your lawyer sent you. Every one of them is a reasonable decision that happens to be invisible to a reader who cannot run your JavaScript.

The record an agent leaves with

Tick what is true of your site

A person leaves your page with an impression. Software leaves with a list of fields it either resolved or did not. Tick the statements that describe your site today and watch the record on the right lose entries.

Extracted record

unverified, read from prose

What it is
Trail running shoe, road to light trail, unisex
Price
unknown
Availability
unknown
Delivery
unknown
Returns
unknown
How to buy
Direct product URL, cart accepts an agent
Best for
Road runners moving to light trail
Contact
help@brand.example, 9 to 5 ET

4 of 8 fields resolved. An unknown is not a rejection, it is a blank the agent fills from someone else.

Every one of these is fixable without a redesign, and each fix helps a hurried person as much as it helps software. That is the honest case for doing this work now, while agent traffic is still small: none of it is wasted if the agents never show up.

The fork

A visitor lands on your product page. Which one is it?

A person

Scrolls, waits for the script, hovers for the spec, and emails you when the price is unclear. Forgiving, and slow.

An agent

One pass, no scrolling, no waiting, no follow-up question. It takes what is stated and leaves with the rest blank.

You do not get to know which one arrived. The page has to work for the reader that asks you nothing.

Everyone loses at the same step

Watch the sequence an agent has to complete to be any use to the person who asked it something. Find the page. Resolve the facts. Compare against the alternatives. Find a route in. Act.

The five steps an agent takes: find, resolve, compare, route in, act. Most sites pass step one and stop at step two, which is where the remaining three are lost.

Most sites pass step one comfortably. Search engines have been finding pages for twenty five years, and the same work that makes you citable in an AI answer gets you found here too. Step two is where it ends for almost everyone, and steps three through five are never reached, not on their merits but because there was nothing to compare.

That is why “we rank well” is no longer reassurance. Being found and being usable are now separate problems, and only one of them shows up in your reports.

Being found and being usable stopped being the same problem. Only one of them shows up in your analytics.

The fixes are unglamorous, and they pay twice

Nothing on this list requires a redesign, a rebuild, or a new platform.

State the price in text, not only in an image or a script-rendered element. Say stock and lead time in words, because a disabled button is not a fact. Put the returns, cancellation, and warranty terms on a page instead of in a PDF. Move key specs out of tabs and accordions that only populate on interaction. Give every bookable or buyable thing a direct URL. Then mark it all up with structured data so the facts are labelled rather than inferred.

Here is the part that makes this an easy decision rather than a bet. Every one of those changes helps a human in a hurry, on a phone, on a bad connection, or using a screen reader. A page that states its price plainly converts better with people too. If agent traffic never materializes for your business, you are left with a faster, clearer, more accessible site, which is the same thing you would have wanted anyway. That asymmetry is rare enough to act on.

The same logic runs through the end of forms: the interface stops being the product, and the facts underneath it start being the product.

What this does not mean

Four things worth refusing, because the vendor pitch around this is already ahead of the evidence.

Do not rebuild for agents. Agent-completed transactions are a small share of trade today and Bain’s own forecast is five years out. Anyone proposing a platform migration for this is using a trend to sell you a project.

Do not generate hundreds of thin pages to feed the machines. That is the old content-farm play in new clothes, and it degrades the site for both audiences.

Do not hand over pricing authority. If you do open a checkout to agents, the terms stay yours: the ACP specification explicitly lets a merchant accept or decline on a per-agent, per-transaction basis, and that control is the point. An agent should read the rules you published, never set its own.

And do not confuse readable with persuasive. Making your facts extractable gets you into the comparison. It does not win the comparison. If your price and terms are simply worse, clarity will surface that faster, which is uncomfortable and also useful.

What to do Monday

Open your most important product or service page and read it as software would: ignore anything that needs a click, a scroll, or a script. Write down what you can still learn. If price, availability, or how to buy is missing from that list, you have your first three tickets, and they are content edits, not a project.

Then measure it rather than assuming. The probe method for AI search visibility works the same way here: ask an assistant to compare you against two named competitors and see which facts about you it gets right. For a worked version of the whole thing on a storefront, the agentic commerce case study walks through the layers end to end.


Want to know what an agent can actually read on your site? Book a free consult and we will pull your key pages apart the way an extractor does, and hand you the list of fixes ranked by what they cost and what they unlock.

Frequently asked questions

What does it mean for a website to be agent-ready?

It means the facts a buyer needs are stated in text a machine can read without running your scripts, clicking through a flow, or opening a PDF. Price, availability, delivery, terms, who the thing is for, and a direct route to buy or book. An agent-ready site is not a redesign, it is the same site with its facts moved out of images, PDFs, and interactions and into plain, marked-up content.

Are AI agents actually buying things yet, or is this hype?

Both, and the split matters. The plumbing is real and recent: Stripe and OpenAI published the Agentic Commerce Protocol in September 2025 alongside ChatGPT Instant Checkout, and Google announced the Universal Commerce Protocol on 11 January 2026 with Shopify, Etsy, Wayfair, Target, and Walmart. Actual agent-completed purchases are still a small share of trade, and Bain's December 2025 forecast puts agentic commerce at 15 to 25 percent of US e-commerce only by 2030. Treat it as a real trend on a slow clock, not a reason to panic.

Do I need to rebuild my website for AI agents?

No, and any vendor selling you a rebuild for this is selling you a rebuild. Almost everything that blocks an agent is a content and markup problem inside your existing site: a price rendered by script, a policy locked in a PDF, stock signalled only by a greyed-out button. Fix those in place. The design stays.

Will this hurt the experience for human visitors?

It usually improves it. Every fix on the list also helps a person in a hurry, on a bad connection, or using a screen reader: facts stated plainly, terms readable without a download, a price you can see before checkout. That is the strongest reason to do this work now, because none of it is wasted if agent traffic never grows.

How is this different from SEO or answer engine optimization?

Answer engine optimization is about being quoted when an assistant describes your market. Agent-readiness is about being usable when software tries to act: compare, shortlist, book, or buy. The underlying discipline is the same, which is stating facts plainly and marking them up, but the failure modes differ. You can be quoted often and still be impossible to transact with.

Found this useful?

Share this with your network on LinkedIn, it helps more than you think.

Enjoyed this read? Get the next one in your inbox.

When we publish something worth your time, you will be first to know. No spam, unsubscribe anytime.

Keep reading

Three doors: two grand ones labeled fine-tune and RAG drawing all the attention, and a plain glowing third door labeled better context that most teams should open first.
AI ProductStrategy

Fine-Tuning vs. RAG: What Your Product Actually Needs (Usually Neither First)

The most-asked technical question in AI product work has a decision tree for an answer, not a winner. Facts that change want RAG. Voice and format want fine-tuning, rarely. And most teams' real gap is a third, cheaper thing nobody argues about on the internet.

Read article
An AI search box answering a question with a paragraph that cites one highlighted company, while a greyed list of ten blue links fades below the fold.
StrategyAI Product

Answer Engine Optimization: How Customers Find You When AI Answers First

Your next customer is asking ChatGPT, not scrolling Google. The answer cites two or three sources, and either you are one of them or you are invisible. A scorecard shows whether an answer engine would cite you today, and the fixes are more honest than any SEO trick.

Read article
A receipt for one task done twice: the agent's token line items totaling eight cents, next to the person's twenty minutes totaling fifteen dollars.
StrategyAI Product

What an AI Agent Actually Costs to Run: The Math Nobody Shows You

Every agent pitch quotes the build price and goes quiet on the running cost. The running cost is tokens times price times volume, and a live calculator shows why it is almost always the smallest number in the room, next to the human minutes it replaces.

Read article

Have software that should be smarter?

Let’s map a free AI-transformation roadmap for your product.