I Built a Website Tool That Isn't Really on the Website Owner's Side

I realized something uncomfortable about Curio recently.

I'm not entirely sure it's in the best interest of the people I'm asking to pay for it.

At least, not in the way we've traditionally defined their interests.

Curio puts an assistant on a website that reads the site's own content and answers visitors' questions from it. It cites the pages it used. If the answer isn't in the material, it says so. If it gets something wrong, the site owner can correct it. If a support question reaches the point where the assistant can't responsibly help, it hands the conversation to a human.

All of that sounds useful.

But consider what happens when it works perfectly.

A visitor arrives at a website looking for one specific thing. Maybe the answer is buried halfway down a 2,000-word article. Maybe it's somewhere in the documentation. Maybe the visitor doesn't even know which page contains it.

The traditional website would like that person to look around.

Click something.

Scroll.

Open another page.

Spend some time.

See an ad.

Maybe see three ads.

Perhaps wander into something else while they're there.

Curio would rather just answer the question.

And that's a problem.

We built an economy around making people look

For most of the web's history, the interests of a visitor and the interests of a website owner have been treated as though they're naturally aligned.

They aren't always.

The visitor wants the thing they came for.

The publisher may need the journey required to find it.

Pageviews, session duration, pages per session, ad impressions, newsletter conversions, product discovery — these are all perfectly understandable things for a website owner to care about. In many cases they're what pays for the website to exist.

But they create a strange incentive.

If I can give you the answer in ten seconds, but making you hunt for it keeps you around for three minutes, which experience should I build?

For decades we've quietly rewarded the second one.

Search engines are now discovering the other side of this problem at enormous scale. Recent experiments on Google's AI-generated search answers have found that giving people synthesized answers can substantially reduce clicks to the publishers whose material helped produce those answers. One 2026 field experiment found AI Overviews reduced outbound organic clicks by 39.8% when they appeared; separate research using Wikipedia estimated significant traffic reductions from exposure to AI-generated summaries.

Publishers are understandably alarmed. The European Commission is currently examining this very tension, including whether publishers should be able to opt out of Google's AI-generated search features without sacrificing their search rankings.

There is an economic bargain underneath the old web: websites make information, search engines send people to it, and some fraction of those people create enough economic value to fund the information.

AI can break that bargain by answering without referring. Recent research has even given the phenomenon essentially that name.

But Curio creates a slightly different version of the problem.

Because Curio isn't standing between the visitor and the website.

The visitor is already there.

What if the website itself stopped making you look?

Imagine you've landed on a company's site because you want to know whether its product works with something you already use.

There are 70 pages on the site.

You could search the navigation.

You could find the documentation.

You could open the integration page.

You could skim it.

Or you could ask:

Does this work with X?

And Hugo comes back with:

Yes.

A short explanation follows, along with the exact source so you can check the answer yourself.

Your information problem is over.

You may leave.

If I own that website and my KPI is pages per session, Hugo has just failed spectacularly.

If I'm the person who asked the question, Hugo did exactly what I wanted.

That distinction has been bothering me.

Because I think I've accidentally built Curio to be loyal to the second person.

Hugo works for the person asking the question

Curio is installed and paid for by a website owner.

But once Hugo appears in the corner of the page, his obligation is surprisingly simple:

Help the person talking to him.

Not keep them talking.

Not maximize engagement.

Not find an excuse to send them through another funnel.

Not invent an answer because an answer is better for conversion than "I don't know."

Just help.

That principle explains a surprising amount of Curio's architecture.

Hugo cites his sources so the visitor doesn't have to trust him blindly. If the site's material doesn't support an answer, he's supposed to admit it rather than improvise one. Conversations aren't kept by Curio by default. And the support assistant is explicitly supposed to recognize the boundary of what it can handle and hand the visitor to a person rather than bluff its way through.

Those choices aren't particularly good at maximizing engagement.

They're good at minimizing bullshit.

And I increasingly think those are different design goals.

This might actually be in the owner's interest

There is another way of looking at it.

Maybe we've been measuring the wrong thing.

Suppose one visitor spends six minutes on your website, opens four pages, fails to find the answer, and leaves frustrated.

Another spends 45 seconds there, asks one question, gets the right answer with a citation, and leaves satisfied.

Which was the better visit?

Most analytics dashboards would celebrate the first one.

I think the second one was better.

And over a long enough period, I suspect the owner benefits from that too.

A customer who can understand your product is more valuable than one who has been successfully detained by your navigation. A person who gets an honest "I don't know" may trust you more than someone who gets a confident hallucination. A support question resolved instantly costs less than a ticket. A difficult question handed to the right human is better than an automated answer that creates a larger problem.

And when someone asks something that is buried somewhere on your site, the fact that you spent the time to publish the answer finally matters.

Curio can surface the obscure page.

The old documentation.

The paragraph nobody reaches.

The FAQ item seven levels deep.

The little piece of institutional knowledge that somebody bothered to write three years ago and everybody forgot existed.

The visitor doesn't have to know where it is.

They just have to ask.

That's not increasing engagement.

It's increasing usefulness.

There is a difference between extracting attention and earning trust

A lot of software built for website owners is ultimately designed to extract something from the visitor.

An email address.

A conversion.

Another pageview.

Another minute.

Another impression.

Another opportunity to sell.

None of those things are inherently bad. Businesses have to sell things. Publishers have to make money.

But somewhere along the way we began treating attention itself as evidence that we had served someone well.

It isn't.

Sometimes a visitor spends ten minutes on your website because it's wonderful.

Sometimes they spend ten minutes because they can't find the damn shipping policy.

Those ten minutes look remarkably similar in an analytics report.

Curio's premise forces me to confront that.

If Hugo gets good enough, he should reduce some kinds of engagement.

He should eliminate pointless searching.

He should eliminate unnecessary clicks.

He should eliminate questions that never needed a human being.

He should eliminate the three minutes between I wonder if they do this and yes, they do.

And if the visitor has what they need after that, Hugo should be perfectly content to watch them leave.

Maybe that's the deal

So no, I'm not sure Curio is always acting in the website owner's immediate interest.

If the goal is to maximize the amount of attention extracted from every visitor, it may occasionally do the opposite.

That's okay.

Because I don't think the healthiest relationship between a website and its visitors is one where the website wins every interaction.

I think the site's job is to be useful.

Curio just makes that bargain unusually explicit.

The owner supplies the knowledge.

Hugo makes it accessible.

The visitor asks the question.

Hugo gives them the best answer the owner's own material can support, shows them where it came from, and knows when to get out of the way.

Maybe the owner gets another pageview.

Maybe they get a customer.

Maybe they get a support escalation.

Maybe the visitor simply says, that's what I needed, closes the tab, and goes on with their day.

For a long time, we've treated that last outcome as a loss.

I'm starting to think it might be the clearest evidence that the website worked.