Organic Search Winners Share 5 Traits

Google’s March 2026 core algorithm update concluded on April 8. The search giant doesn’t provide recovery guidelines for businesses whose rankings have decreased. We’re left with search-engine optimizers to create tactics that align with the winners to help losing sites maintain organic visibility.

A just-published study by SEO pro Cyrus Shepard of Zyppy Signal is an example. He analyzed organic search traffic of 400 winning and losing websites over the past 12 months and classified them by business model, content types, creator profiles, and other definable traits. From there, he identified five characteristics of winning sites.

Here are Cyrus’s five features of sites that consistently maintain prominent organic rankings on Google.

Proprietary assets

Of the 400 analyzed sites, 92.9% of the winners own proprietary assets that are difficult to replicate, such as datasets, products, images, or studies.

For example, a fashion ecommerce site may use its user data to report trends in colors or seasonality. A site with extensive product reviews could repurpose them into shopping guides.

Completes a task

According to the study, 83.7% of winning websites help searchers do something: buy, download, or search.

Winning sites tend to help users accomplish whatever they’re looking for. Losing sites may offer meaningful info on topics, but the searcher must go elsewhere to complete the task.

The solution may be a unique product or an interactive tool. For example, a tutorial site could offer interactive tools, quizzes, and workbooks to help students practice math.

Niche expertise

Expertise within a niche was a trait of 75.9% of the winners.

Winning sites tend to focus on a topic in which they have deep knowledge and experience.

Those sites become go-to authorities for specialized subjects. Hyper-specific travel blogs, for example, often outrank global travel brands.

Unique product or service

A unique product or service is a trait of 70.2% of sites that consistently rank well across core updates. Cyrus’s study found that informational sites (news publishers and affiliate sites) lost the most traffic and that offering a product may be the answer.

For example, a recipe site can sell a subscription meal plan, a book, or access to a private cooking community.

Strong brand

A strong brand, a destination site, was a trait of 32.6% of organic search winners. Cyrus found a high correlation between winning in organic search and having a strong profile of branded search terms.

The more searchers query a business’s name, the more that site is a destination and a strong signal to Google. Treat your brand search metrics as a key performance indicator, in other words.

I’ll add one feature for 2026 that Cyrus doesn’t address: sites that rank prominently in organic search offer something that AI cannot easily replicate.

What Search Engines Trust Now: Authority, Freshness & First-Party Signals via @sejournal, @cshel

Search has not become more chaotic. It has become more continuous.

If the last two years have felt like a blur of updates, volatility, and shifting guidance, you’re not imagining it. What’s changed is not just what search engines value. It’s how those values are evaluated.

The traditional model (the model we’re accustomed to) – periodic updates, relatively stable ranking signals, and long feedback loops – has been replaced by something faster and less discrete. Search engines are now heavily influenced by/running on AI systems that continuously test, interpret, and refine results, so what looks like constant algorithm change is actually ongoing model adjustment.

It’s this shift that has redefined what search engines trust.

The Algorithm Isn’t Static Anymore

For years, SEO operated on a predictable rhythm: core updates arrived, the rankings shifted, and then the industry analyzed the damage, identified patterns, and adapted.

That model assumed a relatively stable system punctuated by updates, but that assumption no longer holds.

Modern search systems incorporate multiple layers of AI-driven evaluation, including ranking systems, retrieval mechanisms, and answer-generation layers. These systems do not wait for quarterly updates. They iterate constantly, adjusting weighting, refining interpretation, and recalibrating outputs in near real time.

What we’re left with is a shorter signal half-life. What worked six months ago may still matter, but it is being re-evaluated continuously rather than periodically.

This is why it feels like we’re in a persistent state of chaos. The system is never settled; it’s always learning.

From Ranking To Evaluation

Traditional SEO focused on ranking documents. Pages competed as whole units, evaluated on signals like links, relevance, and technical accessibility. That model still exists, but it is no longer the full picture.

AI-driven search introduces a second layer: retrieval and synthesis. Instead of simply ranking pages, systems increasingly extract and recombine information from multiple sources to produce answers. This changes the competitive unit, pages still rank but fragments are what get used.

In practical terms, your content is no longer evaluated solely as a document or single URL. It is evaluated as an entire collection of potential answers. Each section, paragraph, and list becomes a candidate for inclusion in AI-generated responses.

Why does this distinction matter? Because it shifts the role of trust. Search engines are not just deciding which page deserves to rank; they are deciding which source is trustworthy enough to be a resource.

Redefining “Trust” In Search

Trust used to feel like a score – it was a combination of authority signals, content quality, and technical hygiene that resulted in stable rankings.

Today, trust behaves more like a probability – it is continuously evaluated, recalculated, and reinforced based on new data. It is not assigned once and retained. It is earned repeatedly.

How is trust determined? There are three factors that dominate the evaluation: authority, freshness, and first-party signals. Each plays a distinct role in how AI-driven systems determine what to surface.

Authority: The Entry Point

Authority has always mattered, no question, but what has changed is where it sits in the process. In an AI-driven system, authority functions as a filter. It determines whether your content is even considered. Not all sources get equal treatment because not all sources are considered authoritative. There is a systems bias toward entities they recognize – brands, authors, and domains that have demonstrated consistent expertise and visibility across the web.

A certain quantity of backlinks is no longer a reliable proxy for authority. Entity-level authoritative presence requires more proof than just links. The search engines build an understanding of who you are (and your authority) based on:

• Mentions across other authoritative sites.
• Consistent authorship and topical focus.
• Brand recognition within a subject area.
• Inclusion in structured knowledge systems.

These signals create what can be thought of as “entity gravity.” The stronger your presence, the more likely your content is to be included in the candidate set for retrieval.

The key distinction is that authority does not guarantee visibility, it guarantees eligibility. Without it, your content may be well-written, well-structured, and technically sound – and still be ignored.

Authority Comes Before Structure

There is a common misconception that better formatting or clearer writing alone can improve visibility in AI-driven search. Sorry, but it cannot, at least not in isolation.

Authority determines whether your content is selected. Structure determines whether it can be used. So, if your brand lacks recognition, your content may never be retrieved. If your content lacks structure, it may be retrieved but never cited. Both layers are required for this to work well.

This is why entity-building efforts, like PR, partnerships, thought leadership, and brand presence, have become inseparable from SEO. They influence not just rankings, but inclusion.

Freshness: The Signal Of Ongoing Relevance

Freshness has also evolved, or maybe it’s more accurate to say that it’s diverged.

In the past, all types of content benefited from freshness, and that fresh factor was often tied to recency. Newer content could reliably receive a temporary boost, especially for time-sensitive queries.

Today, that old kind of freshness only benefits time-sensitive publishers like news outlets. For everyone else, freshness is less about when something was published and more about whether it is being maintained.

When we’re looking at how freshness is evaluated for non-news publishers (i.e., everyone else), we see that AI-driven systems prioritize sources that demonstrate ongoing relevance. This includes:

• Regularly updated content.
• Clear timestamps and revision history.
• Reinforcement of key topics over time.
• Alignment with current information and context.

Outdated content introduces risk. If a system cannot determine whether information is still accurate (especially at grounding), it is less likely to include it in a synthesized answer.
Freshness, in this sense, becomes a trust reinforcement loop. Updating content signals continued expertise. It reduces uncertainty. It increases the likelihood of inclusion.

Please do not confuse this with rewriting everything constantly. It means maintain the content that matters.

First-Party Signals: The Ground Truth

The third big shift is the dramatically increasing importance of first-party signals. AI systems are designed to synthesize information, but they still depend on source material. The quality of that material directly affects the quality of the output. As a result, systems favor content that represents original, verifiable input rather than recycled summaries.

First-party signals include:

• Original research and data.
• Proprietary insights and analysis.
• Direct product or service information.
• First-hand experience and expertise.

These signals reduce ambiguity. They provide a clear source of truth. They are easier to attribute and harder to replicate.

This is one of the reasons the “content at scale” model has struggled in recent years. Large volumes of derivative content offer little new information. They increase noise without increasing value.

AI systems are not looking for more content; they are looking for better inputs. If your content does not add something unique, it is unlikely to be selected.

The Hidden Layer: Usability

So we know that authority gets you considered, freshness keeps you relevant, and first-party signals establish credibility. But none of that matters if your content cannot be used, and this is where many sites fail.

A page can rank well and still have no presence in AI-generated answers. When that happens, it is rarely a ranking issue. It is an extractability issue.

AI systems do not read pages the way humans do. They do not navigate, interpret, and synthesize in a leisurely, exploratory way. They retrieve what is easy to extract and move on.

Content that performs well in this environment tends to share a few characteristics:

• Clear, descriptive headings.
• Logical hierarchy (H1, H2, H3).
• One primary idea per paragraph.
• Direct, declarative statements.
• Lists and tables where appropriate.
• Key points introduced early, not buried.

This is not about writing style. It is about reducing friction.

If a system has to reinterpret your content to isolate the answer, it is less likely to use it. If it can lift a sentence or a list directly, it is more likely to include it. In this sense, structure is not cosmetic. It is functional.

Why “Good SEO” Isn’t Always Enough

Many teams are encountering a frustrating pattern: They rank well, traffic is stable, but they are absent from AI-generated answers.

The first instinct is to look for ranking issues. Then, when that doesn’t fix the problem, move on to re-optimizing keywords, building more links, or publishing more content. These are solutions that do not address the real problem.

Ranking determines whether you are visible in search results. Retrieval determines whether you are used in answers. Those are not the same system. A page can perform well in traditional SEO metrics and still fail to provide clean, extractable segments for AI systems. When that happens, competitors with clearer structure or stronger authority are more likely to be cited, even if they rank lower.

This is not a contradiction, rather it is a shift in evaluation.

Practical Implications

The implications for SEO are straightforward, even if the execution is not.

First, please stop treating updates as isolated events. They are outputs of a continuous system. Optimizing for long-term direction is more effective than reacting to short-term volatility.

Second, invest in authority at the entity level. Build recognition beyond your own site. Where and how you are mentioned matters as much as what you publish.

Third, maintain your content. Freshness is not a one-time signal. It is an ongoing demonstration of relevance.

Fourth, prioritize first-party value. Original insights, data, and expertise are more durable than derivative content.

Finally, structure for usability. Make your content easy to extract, not just easy to read.

Trust Is Now Dynamic

Search engines no longer assign trust once and move on. They evaluate it continuously, so you need to continuously monitor and maintain your trust signals.

Authority determines whether you are considered. Freshness determines whether you remain relevant. First-party signals determine whether you are credible. Structure determines whether you are usable.

All four are required.

If your content cannot be selected, extracted, and trusted quickly, it does not matter how well it ranks. That is the shift, and it is not going away.

More Resources:

Featured Image: beast01/Shutterstock

Your AI Visibility Strategy Doesn’t Work Outside English via @sejournal, @DuaneForrester

This series has been written in English, tested in English, and grounded in research conducted primarily in English. Every framework discussed here (vector index hygiene, cutoff-aware content calendaring, community signals, machine-readable content APIs) was conceived by an English-speaking practitioner, stress-tested against English-language queries, and validated against benchmarks that, as this article will show, are themselves English-weighted by design. That is not a disclaimer, but it is the central problem this article is about.

The AI visibility discourse at large carries the same limitation. One 2024 study analyzing AI evaluation datasets found that over 75% of major LLM benchmarks are designed for English tasks first, with non-English testing treated as an afterthought. The strategies built on top of those benchmarks inherit the same bias.

Enterprise brands are not the villains in this story. Translation-first search content strategies produced imperfect results globally, but markets had learned to live with the nuanced failures. Traditional search indexed what existed, ranked it imperfectly, and the degradation was quiet enough that no one filed a complaint. LLMs raise the bar in a way search never did, and the reason is structural, which is what the rest of this article examines.

The Platform Map

Before optimizing AI visibility in any market, a brand needs to answer a question the English-centric visibility discourse rarely asks: Which AI system are your target customers actually using? The answer varies more dramatically by region than most global marketing teams have accounted for.

In China, a market of 1.4 billion people, ChatGPT and Gemini are not accessible. The AI visibility contest happens entirely within a separate ecosystem. Baidu’s ERNIE Bot crossed 200 million monthly active users in January 2026, and Baidu holds the leading position in AI search market share, according to Quest Mobile. But Baidu is no longer operating in a vacuum. ByteDance’s Doubao surpassed 100 million daily active users by end of 2025, and Alibaba’s Qwen exceeded 100 million monthly active users in the same period. A brand’s English-optimized content architecture is not underperforming in this ecosystem. It simply does not exist there.

South Korea tells a different version of the same story. Naver captured 62.86% of the South Korean search market in 2025 (more than double Google’s share) and since March 2025 has been deploying AI Briefing, a generative search module powered by its proprietary HyperCLOVA X model, with plans for up to 20% of all Korean searches to surface AI-generated answers by end of 2025. Naver is also a closed ecosystem where results route to internal Naver properties, not necessarily the open web. Western brands whose structured data and llms.txt implementation was designed for open-web crawlers are operating with architecture that was never built to reach Naver’s retrieval layer. China and Korea alone account for well over a billion AI-active users on platforms a standard global visibility strategy does not touch.

The Map Is Far Bigger Than We’re Drawing

Those two markets are the ones that get cited because their scale is impossible to ignore. But the platforms being built outside the English-dominant orbit extend considerably further, and the breadth of what has launched in the last two years deserves attention on its own terms.

Europe

• France – Mistral AI’s Le Chat was the No. 1 free app in France after its February 2025 launch; the French military awarded Mistral a deployment contract through 2030, and France committed €109 billion in AI infrastructure investment at the 2025 AI Action Summit.
• Germany – Aleph Alpha trains in five languages with EU regulatory compliance by design, backed by Bosch and SAP.
• Italy – Velvet AI (Almawave/Sapienza Università di Roma) is built specifically for Italian language and cultural context, designed for EU AI Act compliance from inception.
• European Union – The OpenEuroLLM initiative, launched in 2025, is developing a family of open LLMs covering all 24 official EU languages.
• Switzerland – Apertus (EPFL/ETH Zurich/Swiss National Supercomputing Centre, September 2025) supports over 1,000 languages with 40% non-English training data, including Swiss German and Romansh.

Middle East

• UAE/Abu Dhabi – Falcon (Technology Innovation Institute) ranges from 7B to 180B parameters; Falcon Arabic, launched May 2025, outperforms models up to 10 times its size on Arabic benchmarks.
• Saudi Arabia – HUMAIN, backed by the sovereign wealth fund, is framed as a full-stack national AI ecosystem.
• South and Southeast Asia
• India – Bhashini (Ministry of Electronics and IT) has produced over 350 AI-powered language models; BharatGen, launched June 2025, is India’s first government-funded multimodal LLM.
• Singapore / Southeast Asia – SEA-LION (AI Singapore) supports 11 Southeast Asian languages; Malaysia, Thailand, and Vietnam have deployed MaLLaM, OpenThaiGPT, and GreenMind-Medium-14B-R1, respectively.

Latin America

• 12-country consortium – Latam-GPT launched September 2025, led by Chile’s CENIA with over 30 regional institutions, trained on court decisions, library records, and school textbooks, with an initial Indigenous language tool for Rapa Nui.

Africa/Eastern Europe

• Sub-Saharan Africa – Lelapa AI’s InkubaLM supports Swahili, Yoruba, IsiXhosa, Hausa, and IsiZulu; Nigeria launched a national multilingual LLM in 2024.
• Russia/Ukraine – GigaChat (Sberbank) is the dominant domestically deployed Russian AI assistant; Ukraine announced a national LLM in December 2025, built with Kyivstar and trained on Ukrainian historical and library data.

This list is not really meant to be exhaustive, but it is meant to be disorienting.

Every entry above represents a retrieval ecosystem, a cultural signal hierarchy, and a community proof-point structure that a North American-optimized AI visibility strategy does not reach. But the more important observation is about which direction these models were built in.

The old content strategy model was centrifugal: the brand sits at the center, creates content, translates it, and pushes it outward into markets. Traditional search accommodated this because crawlers are indifferent to cultural authenticity: they index what is there. The imperfect results were tolerated because most markets had no better alternative.

These regional models were built in the opposite direction. A government mandate, a national corpus, a specific cultural identity, a language’s syntactic logic, that is the origin point. The model was trained on what that place knows about itself. A brand’s translated content arrives as a foreign object with no parametric presence, carrying the syntactic and cultural signatures of its origin language. Translation does not retrofit cultural fit into a model that was built without you in it.

And this does not stop at the English/non-English boundary. Even within English, regional identity shapes what a model treats as native. Irish English carries vocabulary – craic, gas, giving out, that exists nowhere else. Australian idiom, Singaporean English, Nigerian Pidgin all have distinct fingerprints. A U.S. brand’s content may read as subtly foreign to a model trained predominantly on British or Irish corpora. The direction of the problem is the same regardless of whether the language is technically shared. So often these aren’t just words. They’re compressed cultural signals. A literal translation gives you the category, but often strips out aspects like intensity, intent, emotional tone, social expectation, or shared history.

The Embedding Quality Gap

The reason translation does not solve this is not just strategic. It’s structural, and it lives in the embedding layer.

Retrieval in AI systems depends on semantic similarity calculations. Content is encoded as a vector, queries are encoded as vectors, and the system identifies matches by measuring distance in that vector space. The accuracy of those matches depends entirely on how well the embedding model represents the language in question. Embedding models are not language-neutral. (I think of this as a kind of cultural parametric distance, or a language vector bias issue.)

The most rigorous current evidence comes from the Massive Multilingual Text Embedding Benchmark (MMTEB), published at ICLR 2025. Even across more than 250 languages and 500 evaluation tasks, the benchmark’s own task distribution is skewed toward high-resource languages. The benchmarks practitioners use to evaluate whether their embedding architecture works in other languages are themselves English-weighted. A leaderboard score that looks reassuring may be measuring performance on a test that does not represent the language actually in use.

The structural cause is well documented: the Llama 3.1 model series, positioned at release as state-of-the-art in multilingual performance, was trained on 15 trillion tokens, of which only 8% was declared non-English, and this is not just a Llama-specific problem. It reflects the composition of the large-scale web corpora used to train most foundation models, where English content is overrepresented at every stage: crawl filtering, quality scoring, and final dataset construction. Research comparing English and Italian information retrieval performance, published May 2025, found that while multilingual embedding models bridge the general-domain gap between the two languages reasonably well, performance consistency decreases substantially in specialized domains; precisely the domains enterprise brands operate in.

The embedding gap does not produce obvious errors. It produces quietly degraded retrieval and content that should surface does not, without any visible failure signal. The dashboards stay green. The gap only becomes visible when someone tests in the actual market language.

When Translation Isn’t Enough

Below the embedding layer sits a problem that is harder to instrument: Cultural context shapes what a model treats as relevant in the first place. Research published in 2024 by Cornell University researchers found that when five GPT models were asked questions from a widely used global cultural values survey, responses consistently aligned with the values of English-speaking and Protestant European countries. The models were not asked to translate anything; they were asked to reason, and their default frame of reference was shaped by the cultural composition of their training data.

Consider a brand headquartered outside France, but operating in France. Their content, even if professionally translated, was likely written by non-French-speaking teams with non-French-market authority signals: the institutional citations, the comparison frameworks, the professional register. Mistral was built on French corpora, with French institutional relationships and French media partnerships as its baseline for what counts as authoritative. A Canadian brand’s French content, for example, is tolerated by a French-speaking human reader. Whether it clears the threshold for a model trained on native French content as its definition of relevance is a different question entirely.

The community signals argument from the previous article in this series applies here with a regional dimension. The platforms that drive AI retrieval through community consensus differ by market. In China, Xiaohongshu now processes approximately 600 million daily searches (nearly half of Baidu’s query volume) with over 80% of users searching before purchasing and 90% saying social results directly influence their decisions. The community signals that matter for AI visibility in China are not the ones a strategy built around English-language review platforms is generating.

A brand may have excellent English-language retrieval infrastructure, strong community signals in Western markets, and a well-architected machine-readable content layer, and still be effectively invisible in Korea, structurally disadvantaged in Japan, and culturally misaligned in Brazil. This is not a failure of execution as much as a failure of assumption about which direction the optimization flows.

What Enterprise Teams Should Do

An honest note before the framework: The documented, auditable evidence base for enterprise-level non-English AI visibility strategies does not yet exist in a form that holds up to scrutiny. Work is being done, but a citable case study requires a defined baseline, a measurable intervention, a controlled timeframe, and independently validated results. A practitioner’s assertion that their work applies to your situation is not that. The absence of rigorous case data is a reason to build with intellectual honesty about what is validated versus directional, not a reason to wait. With that in mind, here’s what you can do today:

Audit AI visibility per language and per market, not globally. Query performance in English tells you nothing about performance in Japanese, and performance with global AI platforms tells you nothing about performance inside Naver’s AI Briefing. The audit needs to happen at the market level, using queries constructed in the local language by native speakers, not translated from English.

Map the AI platforms that matter in each target market before optimizing. The list in the previous section is a starting point, not a permanent reference, as this landscape shifts quarterly. Optimization work (structured data, content APIs, entity signals) needs to be built toward the platforms that actually serve each market.

Build localized content, not translated content. The four-layer machine-readable architecture discussed in this series applies in every language. But a translated version of an English content API is not a localized one. Entity relationships, cultural authority signals, and community proof points all need to be rebuilt for local context. The optimization direction is inward from the market, not outward from the brand.

Accept that English-English is not a single market either. The same structural logic applies within English. A US brand’s content may carry American syntactic and cultural signatures that read as subtly foreign to models trained on predominantly British, Irish, or Australian corpora. Regional English is not a rounding error. It is evidence of the same underlying principle operating on a smaller scale.

Accept that a single global AI visibility strategy is insufficient. The frameworks developed in English, including the ones in this series, are a starting point for one slice of the global market. Extending them globally requires treating each major market as a distinct optimization problem: different platforms, different embedding architectures, different cultural retrieval logic, and a different direction of trust.

There is real work to be done. If we step back and look at the big picture again, it’s clear that markets that were once willing to live with the nuanced failures of translation-first content strategies are increasingly operating on platforms built to serve them natively, and that gap is widening. You know I like to name things when the industry hasn’t gotten there yet so here it is: this is the Language Vector Bias problem. And the brands that start closing it now are not catching up to a solved problem. They are getting ahead of the most consequential visibility gap we aren’t really talking about.

More Resources:

This post was originally published on Duane Forrester Decodes.

Featured Image: Billion Photos/Shutterstock; Paulo Bobita/Search Engine Journal