For most of the internet’s history, getting someone to your website was the first real victory.
They searched. They clicked. They landed on your page. Then your website had to do its job.
That sequence is beginning to change.
John Lewis, one of the UK’s largest retailers, recently disclosed that 2.5% of its product searches now come from AI agents. A year ago, the number was 0.3%. The company says the growth is happening across age groups, not just among younger shoppers.
The percentage is still small. The trajectory is what matters.
Salesforce has seen the same behavior at a much larger scale. Its commerce data found a 200% year-over-year increase in shoppers beginning their purchase journey with AI-powered search. Referrals from AI conversations have also risen sharply.
A new layer is forming between businesses and their customers.
Someone wants a chiropractor for chronic headaches. A consultant who understands their industry. Software for a particular problem. A hotel that works for a family with three kids.
Increasingly, they can begin by describing exactly what they want to an AI system.
It compares.
It eliminates.
It recommends.
By the time the person reaches a business’s website, part of the decision may already have been made.
And sometimes they may never reach it at all.
The website has a second audience now
Businesses have spent decades designing websites almost exclusively for human beings.
Make the headline compelling. Make the navigation intuitive. Build trust. Answer objections. Give people a reason to act.
All of that still matters.
But your website increasingly has another reader.
Software.
AI systems need to figure out what you do, who you serve, what makes you different, whether you appear credible and whether your business deserves to be included in an answer.
That creates a strange new situation.
A business can have an attractive website that works perfectly well when a person visits it and still be almost invisible during the process that determines which websites that person considers visiting.
The first impression may happen somewhere else.
This changes what “conversion” means
Conversion optimization traditionally begins after someone arrives.
Change the headline. Improve the offer. Add testimonials. Simplify the checkout. Make the button easier to find.
Those improvements affect the percentage of visitors who become customers.
AI-mediated discovery pushes the conversion process one step backward.
Before your homepage gets the opportunity to persuade anyone, another question increasingly matters:
Can a machine understand why you belong in the consideration set?
That sounds technical. Much of it isn’t.
Consider two businesses.
One says it provides “innovative solutions designed around the unique needs of our clients.”
The other explains precisely who it helps, the problems it solves, how its approach differs, what the process involves, what it costs, where it operates and what evidence supports its claims.
A human being understands the second business better.
So does a machine.
That overlap matters.
For years, businesses have been encouraged to write for algorithms on one side and people on the other. AI search may make that distinction less useful.
Clear businesses are easier to recommend.
We have an early glimpse of how large the effect could become
Researchers recently tested conventional e-commerce websites against versions designed to make their structure and actions easier for browser-based AI agents to interpret.
The difference was substantial.
The research is early, and an experiment shouldn’t be confused with everyday consumer behavior, but the size of the difference deserves attention.
The web was built around human eyes and human hands.
Now software is beginning to navigate it on our behalf.
A checkout button a person can easily recognize may still confuse an agent. A service that seems obvious to the owner may be difficult to classify. Expertise spread across dozens of vague pages may never become a strong enough signal for a system to recommend the business confidently.
That creates a different kind of website problem.
The businesses with the deepest expertise may have the most to lose
This is especially interesting for businesses whose value has always been difficult to explain.
A specialist with twenty years of experience may know exactly why her approach works. Much of that knowledge lives in accumulated judgment: what she notices, what she ignores, the questions she asks, the distinctions she makes and the decisions that have become almost instinctive.
Her website might reduce all of that to:
“Personalized solutions tailored to your individual needs.”
The expertise exists.
The representation of it doesn’t.
Until recently, that mostly created a marketing problem. Someone visiting the site couldn’t see enough difference to understand why this person was worth choosing.
Now it can create a discovery problem too.
If the meaningful distinctions never become explicit, an AI system has very little to work with when someone asks it to compare options.
Twenty years of experience hidden behind six generic sentences can look surprisingly similar to two years of experience hidden behind six generic sentences.
The next six months
I expect businesses to begin treating machine comprehension as part of conversion strategy.
Some of this will become technical. Structured data, accessible architecture, crawlability and machine-readable information will matter.
The more interesting work will happen before any of that.
Businesses will have to become more explicit about themselves.
What exactly do you do?
Who is it for?
When are you a better choice?
How do you think about the problem differently?
What evidence supports that?
What happens when someone hires you?
What do you believe that competitors may not?
Those used to sound like branding questions.
They are becoming infrastructure.
Anthropic’s release this week of new tools intended to help retailers build shopping agents is another indication of where the web is moving. The agents themselves will improve rapidly. Businesses will then have to decide how well their existing digital presence works when software becomes an active participant in buying decisions.
The transition won’t happen all at once.
People will still Google things. They’ll still browse websites. They’ll still ask friends. They’ll still ignore recommendations and choose the company with the photograph they happen to like.
Human behavior rarely changes as neatly as technology forecasts suggest.
But adding another route to a buying decision changes the economics even before that route becomes dominant.
You don’t need 50% of customers using agents before this matters. You only need enough valuable customers doing it that being absent becomes expensive.
What I’d do this week
Forget your business name for twenty minutes.
Open several major AI assistants and behave like a customer who has never heard of you.
Describe the problem you solve.
Ask for recommendations.
Add the criteria your best customers usually care about.
Ask why one business was recommended over another.
Ask what would make the system hesitate to recommend each one.
Then look at what happened.
Did you appear?
If you did, was the description accurate?
Did the system understand what makes you different?
Could it find evidence for the claims you make?
Did a competitor communicate something more clearly than you do?
This isn’t a ranking report. Results vary by system, query, location and context.
Treat it as a new version of an old exercise: seeing your business from somewhere other than your own desk.
For a long time, the job of a website was to explain your value once somebody arrived.
Increasingly, your value needs to be understandable before anyone decides to arrive.