Stop Publishing Content. Start Publishing Proof.

Here's an uncomfortable number: 74% of new web pages now contain AI-generated content. That's not a forecast. That's an Ahrefs study of 900,000 pages published in a single month. Businesses are publishing more than ever. The internet has never had more blog posts, more how-to guides, more "ultimate" anything.

And nearly all of it is invisible.

Google search traffic to publishers fell 33% globally in the year to November 2025. SparkToro's latest study found that 68% of Google searches now end without a single click to any website. AI Overviews appear on nearly half of all queries. The content treadmill that worked five years ago doesn't just produce diminishing returns. For most businesses, it produces nothing.

But something interesting is happening at the same time. Some businesses are gaining traffic. Some are getting cited by AI. Some are becoming more visible, not less. The difference isn't that they're publishing more. It's that they're publishing something fundamentally different.

They're publishing proof.

The Websites Winning Google Look Nothing Like You'd Expect

Cyrus Shepard, founder of SEO research firm Zyppy, analysed over 400 websites to identify what separates the winners from the losers in 2026. He measured five features against year-over-year traffic changes using Spearman correlation coefficients. The results cut against everything the content marketing industry has been selling.

The five features that predict whether a site gains or loses Google traffic:

The most striking finding: sites with four or five of these features win at rates of 68-70%. Sites with none win at 13.5%. That's not a marginal difference. That's a structural divide.

Shepard's third feature, proprietary assets, is the one most businesses overlook entirely. A proprietary asset is something your site has that no other site can replicate: original data, unique tools, primary research, aggregated reviews, customer-generated content, pricing benchmarks. It's information that exists because your business exists.

This is the opposite of what most businesses publish. Most businesses publish articles explaining things Google already knows. They write "What Is [Industry Term]" guides that AI can generate in seconds. They create content about their industry instead of content from their industry.

The distinction matters more than ever. We've written before about why Google stopped rewarding good SEO and started rewarding good businesses. Shepard's data puts precise numbers on that shift.

Why Evidence Works: The Psychology Behind the Data

If the data shows that proprietary assets win, the next question is: why? Why would Google's algorithm and AI models both converge on rewarding original evidence over polished articles?

Rory Sutherland, Vice Chairman of Ogilvy UK, has spent decades studying what he calls costly signaling theory. The core idea: people infer quality and trustworthiness based on the perceived cost or effort invested in something. A FedEx letter is assumed important because who would spend eight dollars to send something trivial? A wedding invitation on embossed card stock signals different importance than an email, even though the information is identical.

Original research is the costly signal of the content world.

When a plumbing company publishes their actual average response times across 500 callouts, that's expensive to fake. They had to do the work. They had to collect the data. They're exposing themselves to scrutiny. That vulnerability is precisely what makes it credible. As Sutherland argues, "the meaning and significance attached to something is in direct proportion to the expense with which it is communicated."

Robert Cialdini's authority principle explains the same phenomenon from a different angle. People defer to demonstrated expertise, not claimed expertise. There's a measurable difference between "We're the best plumbers in Adelaide" and "Our average callout time across 500 jobs last quarter was 47 minutes, with 94% of issues resolved on the first visit." The first is a claim. The second is evidence. One triggers scepticism. The other triggers trust.

This maps directly to what Lily Ray, one of the leading voices in SEO, calls the Experience component of Google's E-E-A-T framework. Google added "Experience" to its quality guidelines in December 2022, signalling that first-hand knowledge matters as much as credentials. Phrases like "in our experience," "after auditing 50 client accounts," and "what we observed across 200 campaigns" aren't just stylistic choices. They are extractable signals that both Google's algorithms and AI models use to assess content quality.

The convergence is remarkable. Behavioural economics, persuasion psychology, and search engine design all independently arrived at the same conclusion: evidence outperforms assertion.

AI Can Write Articles. It Can't Publish Facts.

Here's where the competitive advantage becomes structural.

AI can write a perfectly competent 2,000-word article about plumbing, accounting, physiotherapy, or any other service industry. It can explain concepts clearly, use the right keywords, and structure the content logically. What it cannot do is publish original data from your business. It can't report your average customer satisfaction score. It can't share the results of your latest pricing analysis. It can't describe what you observed across your client base last quarter.

This is why AI-generated content tends to make everyone sound identical. It draws from the same training data, producing the same conclusions in the same tone. The only way to stand apart is to contribute something the model hasn't seen before.

And this matters for AI visibility too, not just traditional search. Research into how LLMs decide which sources to cite shows that original data is one of the most reliable triggers for AI citations. When ChatGPT, Perplexity, or Google's AI Overview needs a specific statistic, it has to cite the original source. If that source is your business, you become the reference point for your industry.

The numbers back this up. Only 38% of pages cited in AI Overviews also rank in the traditional top 10 results. That means Google's AI is actively looking beyond the usual ranking factors to find sources with unique, citable information. Being "well-optimised" is no longer enough. Having something worth citing is what gets you into the answer.

Byron Sharp's marketing science framework helps explain the broader strategic implication. Sharp identifies physical availability as one of the two fundamental drivers of brand growth: the degree to which your brand is easy to find and buy when someone enters the market. In Sharp's framework, Google Search is physical availability, not advertising. It's the digital equivalent of shelf space.

When your business publishes original data that gets cited across AI Overviews, industry articles, and competitor analyses, you're building physical availability at scale. You're not just ranking for a keyword. You're becoming a reference point that shows up wherever your industry gets discussed. That's the difference between being a source and being invisible.

What "Evidence" Actually Looks Like for a Small Business

"Publish original research" sounds like advice for companies with research departments and six-figure content budgets. But every business sits on data that nobody else has. You just need to recognise it and package it.

Here's what evidence-based content looks like across different industries:

The pattern is always the same: take something you already measure, already observe, or already know from doing the work, and make it public. You're not creating new research infrastructure. You're surfacing the expertise you already have.

A physiotherapist who's treated 3,000 knee injuries knows things about recovery timelines that no amount of desk research can replicate. A pest controller who's inspected 500 homes in a specific suburb knows which streets have termite problems. An electrician who's quoted on 200 solar installations knows the real payback periods, not the manufacturer's marketing numbers.

That knowledge is your proprietary asset. It's expensive to acquire (you had to do the work), impossible to fake (the specifics would unravel under scrutiny), and genuinely useful to the people searching for it. It ticks every box that Shepard's research identified, that Sutherland's costly signaling theory predicts, and that Google's E-E-A-T framework explicitly rewards.

The Compounding Effect

Evidence-based content doesn't just perform better. It compounds.

A generic "5 Tips" article competes with thousands of identical articles and degrades as AI produces more of them. A piece of original research gets cited by other sites, referenced by AI models, linked in industry discussions, and shared by people who find it genuinely useful. Each citation builds authority. Each reference builds the kind of trust that makes buyers choose you over competitors.

This is the flywheel that most SMEs never build because they're too busy publishing their fourth "Ultimate Guide." The businesses winning visibility in 2026 aren't the ones publishing the most. They're the ones publishing what nobody else can.

What This Means for Your Business

The internet is drowning in content. It is starving for proof.

Further Reading

• 5 Data-Backed Features of Websites Winning Google in 2026 - Cyrus Shepard's original research on 400+ sites
• In 2026, Less Than One Third of Google Searches Still Send a Click - SparkToro's zero-click search data
• The E-E-A-T Method - Lily Ray's framework for building first-hand experience signals
• Rory Sutherland on Costly Signaling and Trust Placebos - The psychology behind why evidence builds trust
• How LLMs Decide Whom to Cite - Research on AI citation patterns and source selection