AI Didn’t Kill Your Website Traffic. Sameness Did.

Updated April 2026

Something has changed in the relationship between an organization’s website and the new business it used to reliably generate. Traffic for most websites has fallen off a cliff, meaning sales inquiries are nowhere near where they were, and our super-mega state-of-the-art website that felt fine eighteen months ago has started to creak like a mausoleum door.

You’d be forgiven for assuming the blame lies with some technical issue. In the old days, all we needed to do was work on our site’s page speed, add a bunch of schema.org markup voodoo, or pay around with server cache or CDNs, because technical problems have technical solutions and technical solutions have project timelines that come with that satisfying “oh yeah” moment when all the boxes turn green and our work is done.

But the issues we’re facing today aren’t, for most businesses, things that can be solved with a quick 301 redirect.What’s happening today sits at the intersection of two things that arrived at roughly the same time and are accelerating in the same direction. The first is the structural shift in how all of us use search. The second is a change in what the web itself is made of.

When search results stopped being a gateway

For most of the internet’s commercial history, a search engine worked as a referral mechanism. We typed in a query, Google et al returned a list of pages, we clicked on one we liked best, and traffic moved from the search result to the website of choice. Easy peasy. Today, that basic hand-off process has been broken. SparkToro’s analysis of search behavior found that for every 1,000 Google searches in the US, only 360 clicks reach the open web. In the EU it’s 374. The rest of those queries resolve inside Google’s own interface — through featured snippets, knowledge panels, and increasingly, AI-generated summaries that appear above everything else on the search results page.

Is this Google behaving differently than it ever has? Yes, but it’s also Google morphing into what it always intended to become: an answer engine instead of a search engine. The shift has been happening so imperceptibly that most of us didn’t even pay attention to that super-gradual downward slope of traffic, until a couple of Google Updates sharpened the blade at which point the snowball’s rolling down the hill.

AI search platforms are running a similar, parallel track. ChatGPT, Perplexity, et al handle structurally different queries in their prompt boxes compared to what people type into Google. While the average Google search query runs to 3-4 words, an AI prompt often runs to 25 or more, and the interaction is conversational rather than directional — the user is exploring a problem rather than hunting for a specific URL. Organic click-through rates in the US have dropped to 40.3%, down from 44.2% the year before, with zero-click behavior rising simultaneously. AI-referred traffic also seems to convert at higher rates than organic search traffic, which means the businesses that get cited by AI systems are being sent high-value visitors, not just mildly curious ones.

AI raises the floor, which changes everything about the ceiling

At the same time as search behavior shifts, AI-created content is changes the makeup of the web itself. Ahrefs analyzed 900,000 newly-created web pages in April 2025 and found that 74.2% of them contained AI-generated content. A separate study by Graphite examined 65,000 articles published between 2020 and 2025 and found that by November 2024, more than half of newly-published web articles were primarily AI-written, up from around 5% before ChatGPT’s launch in late 2022.

Many of us would interpret this as AI-generated content is uniformly poor, that search engines will penalize it, and writing our own copy automatically gives us an advantage. If only it were that simple. What’s actually happened is AI has raised the floor of website content quality across the web. Businesses that previously had text that was generic, error-ridden, and untouched since it was written in 2011, have used AI tools to produce something competent, grammatically correct, and superficially coherent. OK, it’s never going to win a Pulitzer, but it doesn’t have to. The website’s text is better than it’s ever been, since it wasn’t ever written with the required research, forethought, and positioning work that normally happens at the copywriting stage.

As that floor rises, the average rises with it, and average is now more crowded than ever before. A page that would have stood out three years ago for being clearly written and reasonably organized now blends into a much larger body of clearly written, reasonably organized content covering the same ground in more or less the same way. The same tools are available to every business in every category, and most of them are producing content that sounds like everyone else using those same tools. At scale, competence became indistinguishable from sameness. We’ve optimized for the mean, but the mean isn’t where success lies.

We’re still trying to solve the vocabulary problem

Underpinning both of these shifts is something that predates AI by years and has become more expensive to get wrong: the gap between how a business describes itself and how a buyer describes the problem they’re trying to solve.

The website copy of most businesses is written from the inside out, because that’s how the organization thinks about its own products and services — its internal language, the features it’s most proud of. But buyers approach the same subject from the opposite direction. They’re describing a situation, a frustration, a question they’re trying to answer. The research on how buyers actually begin the purchasing process consistently shows that around two-thirds of B2B buying journeys start with a broad, problem-focused search rather than a brand name. Nine times out of ten a buyer isn’t looking for us specifically. They’re looking for their problem, and trying to find out if anyone out there understands it well enough to be worth contacting.

A website that fails this test may not be “broken” from a technical perspective. It may load quickly, pass Core Web Vitals tests, have clean schema markup, and render perfectly on any mobile device we can throw at it. The issue now is that it fails commercially, because the words on it don’t match the words in the buyer’s head.

What search and AI systems actually reward

Graphite’s study also found that, despite AI articles being published in enormous volumes, the pages that tend to rank at the top of search results still lean toward human-written or substantially human-edited content. The volume advantage that AI offers in content production doesn’t always translate to a visibility advantage in search results. Search engines and AI platforms are trying to determine which content most credibly addresses a given question. A page earns credibility because of things like authority signals, backlinks, domain reputation, and an established expertise in the subject matter at hand, but it’s also about the specificity and originality of what’s being said. A page that reflects actual experience or direct knowledge of the problem it’s addressing reads differently to an LLM than a page cobbled together from the same publicly available information as every other page on the same topic.

In other words, the marketing value of genuine insight — thinking that could only have come from direct engagement with a problem, a market, or a buyer — has increased at the same rate as the supply of competent but generic content, even if they’re both moving in opposite directions. That means producing content that works takes longer, requires more from the people creating it, and demands that the organization actually has something to say about its subject beyond what a well-prompted language model would say about it. If you’re like most businesses, you’ll find reasons not to bother doing this. But the same is true of your competitors, which is probably the more useful and opportunity-driven way to think about it.

The website as one property among many

We also need to stop thinking of our sites as the zenith of our customer content publication channels Research on how B2B buyers evaluate purchasing options shows they’re drawing from multiple sources simultaneously. Our websites is merely one node in that network, not the hub everything else passes through. This matters because a website that says one thing while the organization’s LinkedIn presence says something slightly different, or where the tone and vocabulary shift noticeably between the site copy and the content the business publishes elsewhere, creates a coherence problem buyers notice without necessarily being able to give it a name. The impression formed is of an organization that hasn’t thought carefully about what it is and what it does.

The content problem visible on the website is usually the same content problem present across every other customer-facing property. The website is where it tends to become visible first, because it’s where traffic gets tracked and where the absence of new inquiries eventually forces the question. But the issue is upstream of any single page, which means any fix that’s limited to the site is more likely to treat the symptom instead of the cause.