Can People Tell If Your Website Was Built With AI?

What a visitor is actually deciding

Watch anyone use a site they have never seen before. They land, they scan the headline, they scroll once, and they reach a verdict. The verdict is about risk: is there a real business here, or am I about to waste twenty minutes and possibly some money. Nothing in that process requires a theory about how the page was assembled.

Which means a page can be obviously machine-built and still pass, as long as it is specific about what the business does and who it is for. And a page can be hand coded over three weeks and fail, if it says nothing concrete. The tooling is invisible. The vagueness is not.

Detection and judgment are two different things

Detection is a conscious claim: an AI made this. Judgment is a feeling: something is off here, I am not sure I trust it. Almost all of your traffic produces the second one and never arrives at the first.

The distinction matters because people spend their effort on the wrong problem. They try to disguise the tooling, swapping one font for another, when the thing actually costing them money is an absence of evidence. A visitor who cannot work out what you sell, to whom, for how much, leaves. They will not have formed an opinion about your build process on the way out.

I have also seen the reverse mistake, which is assuming nobody notices anything. Something does register. It is just not the thing most people are guarding against.

Who can actually tell, and whether they matter

There is a real population of people who can look at a page and name the pattern. Roughly in order of how often they show up:

• Other builders and designers. They have seen the same default layout a few hundred times and they clock it instantly.
• People who spend a lot of time online. Not professionals, just heavy readers. They cannot explain it, but they recognize the shape.
• Your competitors. They will notice, and some of them will say so.
• Journalists and partners doing due diligence. A small group, but the one where it costs you most.

Notice who is missing from that list. The person with a problem, a budget, and a browser tab usually has no frame of reference for your site beyond the three other sites they looked at this morning. They are comparing you to those, not to a mental catalogue of AI output.

That gap is closing, though. Every month more people have seen more of the same pattern, and recognition gets cheaper with exposure. The tells that read as neutral in 2024 read as a shrug now. I would plan for that trend to continue rather than hope it stalls.

The signals people read without naming them

When someone says a site feels fake, they are usually reacting to one of a short list of things. These are not design crimes in themselves. They are absences, and absence is what reads as risk.

Symmetry with no information in it. Three equal boxes, three equal headings, three equal sentences. The layout promises a comparison and delivers filler. Real content is lumpy, because real businesses have one thing they are great at and two things they are fine at.

Copy that would fit any company. If you could paste your headline onto a competitor's site and nobody would notice, the headline is doing no work. This is the single most common failure I see, and it has nothing to do with grammar or polish. Specificity is the fix, and I wrote about the mechanics of that in how to write website copy that converts.

Nobody in any of the pictures. Abstract shapes and glowing blobs are the cheap default. A single real photograph of a real person doing the real work outperforms an entire gallery of them.

A polish gradient. A gorgeous hero section, then a pricing page with placeholder tiers, an empty blog, and a contact form that goes nowhere. The mismatch is louder than either half. People read it as abandoned.

Missing boring evidence. A city, a real price, a named human, a dated post, a phone number that someone answers. The boring details are what separate a business from a landing page about a business.

AI detectors exist. Your buyer is not using one.

There are tools that claim to tell you whether a page was AI-generated, and some of them work reasonably well on prose. They do not evaluate layout, and nobody runs one before buying a $400 service from a small company. They are a curiosity, not a gate.

The detector research is still useful for a different reason: it tells you how much company you have. Graphite sampled 55,400 English-language article URLs from Common Crawl and found that AI-written and human-written articles now sit at roughly half each, with the AI share crossing human for the first time at 50.9% in Q4 2025 and settling at 49.9% in Q1 2026, according to their published analysis. Their own false-positive checks on articles published before ChatGPT existed came in at 1.4% to 1.8% depending on the detector.

Two things follow. The first is that detectors flag some human writing as machine writing, so a clean score was never going to be proof of anything. The second is bigger: at that volume, machine involvement has stopped being a distinguishing accusation. It is the water. Nobody is going to beat you by pointing out that you used the same category of tool as half the internet.

What volume has done to the question

A few years ago, the fact that AI touched your site was interesting. Now it is background. The question that replaced it is harsher and more useful: does this page know anything?

That is the test readers apply, even if they would never phrase it that way. A page that knows something names a specific customer, a specific objection, a number it stands behind, a tradeoff it is willing to admit. A page that knows nothing hedges, generalizes, and sounds pleasant. Models are very good at pleasant. They cannot supply the knowing part, because you are the only one who has it.

This is also why the sameness problem is an opening rather than a sentence. I went through the mechanism in why every AI website looks the same, and the short version is that the bar for standing out dropped exactly as the average rose.

How to find out instead of guessing

You cannot audit your own first impression. You have seen the page too many times. So stop guessing and run a cheap test on actual humans.

Show the page to five people who do not know what you do. Give them ten seconds, then take it away and ask two questions: what does this company do, and who is it for. If they cannot answer both in one sentence each, you have your answer, and it is not about AI.

Then ask a third question, but only after the first two: did anything about it feel off, and if so, what. People are unreliable at diagnosing why, and very reliable at reporting that they felt it. Take the feeling as data and ignore their proposed fix.

Five people is enough. The same two complaints will show up three times each, and those are the ones worth your weekend. If the complaints are about the look rather than the content, the specific tells and their fixes are a separate piece of work.

Should you tell people you used AI?

Almost nobody asks. In practice you get to choose, and either choice is defensible.

Volunteering it has a small upside with a technical audience, who tend to respect the honesty and sometimes want to know how. It has a small downside with buyers who have been burned by cheap work and hear the word as a warning. Neither effect is large enough to decide the matter for you.

What disclosure definitely does not do is rescue a thin page. Saying you built it with Claude does not make a vague headline less vague. And hiding it does not save a page that has nothing to say. If you want to be open about it, be open about it in a way that costs you nothing: a line in the footer, a short note on your about page, an honest answer when someone asks. This site carries a badge saying how it was built, which has never once come up as an objection.

So: can people tell? A handful can, and the number grows every year. But that is the wrong thing to optimize. Nobody has ever bought anything because a website was hand coded, and plenty of people have walked away from pages that felt like nothing was behind them. Spend the effort on being specific and on showing evidence that a real operation exists. Do that and the question of which tool typed the HTML stops mattering, because it was never the question your visitor was asking.

Related: How to Audit Your Own Site for the AI Look · How to Write AI Copy That Doesn't Sound Like AI · How to Build a Landing Page With Claude, Step by Step

Common questions

Can people tell if a website was made by AI?

Most visitors cannot and do not try. Other builders, designers, and heavy internet users often recognize the default layouts and generic phrasing, but ordinary buyers form a judgment about whether a real business exists rather than about which tool built the page. Generic design reads as untrustworthy, not specifically as machine made.

Are there tools that detect AI-generated websites?

Several exist and they mostly analyze written text rather than design or layout. Published detector testing shows false positive rates of roughly one to two percent, meaning some genuine human writing gets flagged as machine written. No customer runs a detector before buying from a small business, so treat these tools as a curiosity rather than a risk.

Does it hurt your business if visitors think your site was AI-generated?

The damage comes from what the generic look implies rather than from the AI itself. A page that could belong to any company suggests nobody is home, which costs you trust at exactly the moment a visitor is deciding whether to spend money. Specific copy and visible proof of a real operation repair that, regardless of how the page was built.

Should I disclose that I used AI to build my website?

It is your call and almost nobody asks. Technical audiences often respect the honesty, while some buyers hear the word as a warning about cheap work. A short note in the footer or on your about page is enough if you want to be open. Disclosure will not fix a vague page, and concealment will not save one.

How can I check what visitors actually think of my site?

Show the page to five people who do not know your business, give them about ten seconds, then ask what the company does and who it is for. If they cannot answer both in one sentence, the content is the problem. Ask afterward whether anything felt off, and treat the feeling as data while ignoring their suggested fix.