Google Ranking Factors 2026: Ten Factors Tested Against Real Data

When Cyrus and Dawn Shepard published their survey on Google ranking factors on 9 September, I got stuck at number two. More than half of the surveyed SEOs count backlinks among the three most important factors. So I checked my own site. Of my twelve German articles that ranked on page one on average in the four weeks up to 12 September, only four have a backlink according to Google’s own link data. Three of them are linked from exactly one website.

That is not a contradiction of the study. It is a question put to my own domain. seo-kreativ.de is young, my first post went live in July 2025. There has never been any active link building here. On a site like this the same factors can work differently than on a domain that has collected links for years. That is exactly what I wanted to know.

So I go through all ten factors of the study in order. For each one you get what Google itself says about it, what my Search Console shows and whether the factor holds on my site. At the end I write down what I take from it for my own work.

For scale: seo-kreativ.de has just under one hundred German posts, each with an English twin. In the quarter from 12 June to 8 September the whole domain reached 1,483,366 impressions and 3,597 clicks. That is not a large domain. Several Google updates fall into the same period and can move any of these numbers. My numbers therefore speak for my domain and for no other. I do not want to show more with them.

Mentions are the share of respondents who counted the factor among their three most important in an open question. The two columns on the right are my own data. They do not replace the study. They put one concrete domain next to it.

Cyrus and Dawn Shepard surveyed 131 SEOs for the study. Each person rated 103 factors on a scale from plus three to minus three. Then came an open question about the three most important ranking factors. The ranking of the ten most-mentioned factors above comes from those answers.

The scale rating is at least as interesting. There, search intent match was the strongest single factor, meaning how well a page fulfils the task behind a search. Zyppy explicitly stresses that this goes beyond matching keywords. The meta description, on the other hand, was rated so weakly that most respondents no longer attribute any effect to it according to Zyppy’s analysis. On site speed and the Core Web Vitals the respondents disagreed more than on most other points.

I agree with the basic message. Each of these measures plays its part, none is enough on its own. A survey reflects the average of many experiences, across domains, industries and sizes. That is exactly why I find it so useful as a starting point. Read the percentages as the assessment of many colleagues, not as a recipe for your site. Whether the order also applies to a single young domain, the survey cannot say. Only that domain’s own Search Console can.

At Google, a ranking system is a named component that takes over part of the evaluation. A ranking factor would be a single signal with a weight. Google documents the former. Whether people talk about Google ranking factors, ranking criteria or ranking signals, at the core they mean the same thing.

The central page is the guide to Google Search ranking systems. It opens with the sentence that Google’s systems look at “many factors and signals about hundreds of billions of web pages and other content”. Then follows a list of systems: BERT, deduplication, freshness, link analysis and PageRank, MUM, neural matching, passage ranking, RankBrain, the reviews system, site diversity and spam detection. According to the same page, the helpful content system, Hummingbird, Panda and Penguin have been absorbed into the core systems.

The German page So ermittelt Google das Ranking names five aspects: meaning of the query, relevance, quality, usability and context. On quality it says: “Beispielsweise wird geprüft, ob bekannte Websites auf die jeweiligen Inhalte verweisen.” In English: for example, it checks whether well-known websites link to the content. Freshness plays “eine größere Rolle”, a bigger role, for news topics. Google does not name weights here either.

The figure of over 200 ranking factors was written by Google itself. On the old “How Search Works” page, the archived version from 2013 said: “We then rank the results … using over 200 factors.” The archived version from 2016 spoke of “more than 200 unique signals”. Both pages redirect today, and the new one no longer gives a number.

1. Relevance and search intent

All sources agree here. Google names relevance as the second of five aspects, ahead of quality. In the post-trial brief of the US plaintiffs in the antitrust case, former Google engineer Eric Lehman is quoted from his testimony in court. On the value of search results, “relevance … the most important consideration”. In the study, relevance is number one, and search intent match leads the scale rating.

On my site I evaluated 3,114 combinations of page and search query for this, across 164 pages in both languages from June to September. I left out queries with more than six words, as those are mostly automatically generated queries from AI systems. When all words of the query appear in the title, the median position is 10.8. Otherwise it is 13.2. On the German pages the gap is larger, 13.1 against 17.6.

I find the counter-check more interesting. In 89 percent of the combinations the search query does not appear literally in the text even once. So my pages rank for questions they answer, not for words they repeat. A higher keyword density is not associated with better or worse positions on my site. That matches what Zyppy writes about search intent match.

At seo-kreativ.de: holds. For me it follows that most of the work happens before writing. If you miss the search intent, you can tweak technology, links and length afterwards without much moving.

2. Backlinks

Google lists link analysis as one of its ranking systems. Google’s Gary Illyes said in 2023 according to Search Engine Roundtable: “I think they are important but I think people overestimate their importance. I don’t agree that it’s the top 3 and it hasn’t been for some time.” In the study, backlinks are number two, meaning above all links from trustworthy, topically close sites.

On my site it looks like this. In the four weeks up to 12 September, twelve German articles ranked on page one on average. According to Search Console’s link data, eight of them have not a single referring website, the other four have one to three. Of all 76 German articles with at least 30 impressions, only 14 have a backlink at all. There is no visible relationship with position.

An honest caveat belongs here. The search queries that put these twelve articles on page one are small. Apart from TL;DR, DataForSEO estimates none of them at more than 210 searches per month. For five of the twelve articles it is ten or twenty, and for three the tool reports no value at all. They are update news and niche topics. For “google core update” with 720 searches per month I rank at position 13, for “google discover” with 4,400 at position 20. So without links I get to the front for small search queries. For the big ones not yet.

A cross-section proves little, though. So I additionally looked at the moment an article receives its first backlink. Only DataForSEO gives me the date for that. Search Console does not report it per target page. This could be evaluated for 54 articles. In the four weeks afterwards, 27 of them were above the rest of the domain and 27 below. After eight and after twelve weeks it was no different. The positions for the same search queries did not move either. A gain of two positions would very likely have shown up in this analysis.

I found the other direction interesting. Even before their first link, these articles were more visible than three quarters of my other posts. On my site, links come to articles that are already being found. That fits how they came about: someone finds a post through search and links to it.

The link profile itself is modest. Search Console counts 205 referring domains, 159 of them with exactly one linking page. The largest source is the WordPress directory with the entry for my plugin. I have not actively built a single link. DataForSEO additionally shows a link seller’s network of 193 domains pointing at my TL;DR pages. Only two of those domains appear in Google’s link data. More on that in Spam backlinks all of a sudden?

At seo-kreativ.de: not demonstrable. That does not mean links are unimportant. My links are single references from small websites. The effect of many strong links on a whole domain is something I cannot test on mine. It does show, though, that a young domain can reach page one for small search queries in its topic without them. It says nothing about competitive terms. On an older domain with a strong link profile it can look completely different. Before you invest in links on a young domain, take a look at your own top pages. Why I still do not ignore links is in Building backlinks for small domains.

3. Content quality and freshness

Google names quality as the third aspect, freshness has its own system. On length, Google is clear. The SEO starter guide says there is “no magical word count target, minimum or maximum”. In the document on helpful content, Google asks whether you write to a particular word count because Google supposedly prefers one. The answer follows right away in brackets: “(No, we don’t.)” The study counts originality, accuracy and freshness as quality.

Quality itself I cannot measure. What I can check is what happens after a refresh. For that I went through the revisions of all 98 German articles and counted every refresh with at least 250 words or 15 percent more text. I compared 28 days before with 28 days after, always against the rest of the domain in the same period. I removed articles about Google updates, because their demand flattens out after a short time anyway.

That left 16 refreshes of guide articles. For eleven of them, impressions rose more than in the rest of the domain. Three articles collapsed instead. One of them is my E-E-A-T guide, which I had tripled in length in April: 83 instead of 487 impressions.

Where does the gain come from? For that I looked at the individual search queries, again without the long queries from AI systems. Nine of the 16 articles had enough of them for a comparison. Eight of those nine appeared for more search queries after the refresh than the rest of the domain. On average it was around a quarter more. For the queries that already existed, the position hardly changed: slightly better for five articles and slightly worse for four. The average position of a page is misleading here, by the way. After its collapse, the E-E-A-T guide stood 27 positions better. It had simply lost the poorly placed queries.

On clicks I am careful. All 16 articles together had 56 clicks in 28 days before, 61 afterwards. One article lost 17 clicks, another gained 13. That is too little for a statement about clicks.

At seo-kreativ.de: holds. Though differently than expected. What I measured is refreshes, not quality. A refresh does not lift my existing rankings. It makes the article visible for more search queries. Little of that has reached the clicks so far. It is not a sure thing. Three of 16 articles lost. Simply inflating an article that runs well is something I no longer do since the E-E-A-T guide. If you refresh, write down beforehand what exactly is supposed to get better.

4. Authority and trust

The study means how much Google trusts a website, its authors or its brand. Google itself writes in the document on helpful content: “While E-E-A-T itself isn’t a specific ranking factor, using a mix of factors that can identify content with good E-E-A-T is useful.” And on the quality raters: “Search raters have no control over how pages rank.”

Domain Authority does not belong here, even if the name sounds like it. That is a metric from Moz. John Mueller’s tweet on Domain Authority says tersely: “We don’t use domain authority.” The API documents that became public in 2024 do contain an attribute called “siteAuthority”. How it is calculated or used is not known, according to Mike King’s analysis of the leak documents.

At seo-kreativ.de: not testable. Search Console has neither a field nor a metric for it. What the rater guidelines say about it is in my E-E-A-T guide.

5. Click and user behaviour

Zyppy understands this as how searchers select results and interact with pages. You learn nothing about it from Google’s documentation, but quite a bit from the court records. The same brief calls Navboost “one of the most powerful ranking components” at Google and refers to an exhibit from the proceedings for it. The system is said to memorise all clicks for all queries of the past 13 months. That is the plaintiffs’ account, not a finding of the court and not information from Google. The API documents that became public in 2024 contain attributes such as “goodClicks” and “badClicks”. Rand Fishkin, who published them, writes explicitly that they show neither the weight of individual elements nor which of them are used in ranking.

I cannot verify this, because clicks cannot be generated under controlled conditions. What I do see is the click-through rate itself. My strongest search query is “tl;dr”, from 10 May 2025 to 8 September 2026 with 145,718 impressions and 78 clicks at an average position of 7.2. Such lookup questions are often answered directly in the results list. How to separate that from real losses is in Impressions collapsed: how to narrow down the cause.

At seo-kreativ.de: not testable. Why bought clicks are still no shortcut is in Detecting CTR manipulation.

6. Brand signals

The study means “a brand/entity that people search for, trust, and visit”. In Search Console this shows most clearly in search queries with your own name. Among the queries Search Console displays at all, queries with “seo kreativ” or my name account for 2.6 percent of clicks and 0.04 percent of impressions. Almost all visitors come through factual questions.

At seo-kreativ.de: not yet. After a good year the site hardly has a brand and is visible anyway. For a young domain I find that normal. So a site can be found without a brand, at least mine can. Whether brand signals make a difference for me later, I can only see once they exist.

7. User satisfaction

Zyppy describes this as task completion: did the person get what they wanted after the click? That is exactly where Search Console ends. It shows impressions, clicks and positions, but not what happens on the page afterwards.

At seo-kreativ.de: not testable. What I do about it is unspectacular. I answer the question from the title in the first paragraph, and I write the rest for those who want to go deeper.

8. Technical health

In the study this point stands for crawling and indexing as the foundation of all other factors. Google is unusually concrete here. The page experience documentation says: “Core Web Vitals are used by our ranking systems.” A few sentences later, though, it also says Google shows the most relevant content “even if the page experience is sub-par”. HTTPS has been a signal since Google’s announcement of 7 August 2014, described back then as a “very lightweight signal”.

On my site this is met, equally on every page. All 213 pages in the crawl run over HTTPS, plain HTTP redirects with a 301. I checked eleven of my own pages with a performance trace. Where a Largest Contentful Paint fires, it lies between 125 and 222 milliseconds, against the threshold of 2,500. Cumulative Layout Shift is zero everywhere.

Indexing is fine as well. Of 194 articles in both languages, 192 had impressions in the four weeks up to 12 September. The remaining two, according to the URL inspection in my Search Console, were crawled by Google but not indexed. Robots.txt, fetching and indexing were allowed. That is a decision by Google about the content, not a technical error.

One line appeared in every single report: there is no field data from real users for any of these pages. Google evaluates Core Web Vitals on real users. For a domain of this size, not enough comes together for that at page level.

At seo-kreativ.de: met and therefore not testable. Because every page is equally good, no difference shows between them. Whether a slow page would rank worse on my site is something I cannot see this way. In my view, technology wins no ranking. It is the condition for there being one at all. More on that in Crawling and indexing and Optimising Core Web Vitals.

9. Topical authority

According to the Zyppy survey, this is about “how well your brand/site is associated with expertise on a particular topic”. On my domain this can at least be approximated. My German articles are grouped into eight topic clusters with four to 29 articles per cluster. If depth in a topic brings visibility, the large clusters should get more impressions per article than the small ones.

On my site it is rather the reverse. My largest cluster, 29 articles on how Google ranks, gets a median of 90 impressions per article in the four weeks up to 12 September. The smallest, four articles on web performance and AI agents, gets 298. Across all eight clusters the relationship between size and impressions per article is minus 0.40. The large clusters do have more clicks in total, simply because they have more articles. The topics also differ a lot in demand. And eight clusters is not many.

You can read my numbers the other way round too. Articles without a single backlink rank on page one in their topic. For many SEOs that is exactly what topical authority is. My cluster analysis does not capture that. It measures demand per topic more than anything.

At seo-kreativ.de: not yet in my measurement. After a good year, more depth in a topic means more articles on my site, but not more visibility per article. I will still build out the clusters, because readers keep reading there. If your domain is older than mine, it can look completely different for you. How I build them is in the hub-and-spoke model.

10. Internal links

Google says in the documentation on crawlable links: “Every page you care about should have a link from at least one other page on your site.” That is a statement about discoverability. In the study, internal links are number ten.

In August I first tried this with six internal links. My link graph had calculated them as the most effective ones on the domain, from pages with many menu links to four articles that got nothing from there yet. Because I set internal links in both languages, eight pages were affected. A few weeks later, four of them were above the rest of the domain and four below. But eight pages are simply too few for such a question. Even a real effect would have drowned in that quantity.

Luckily I already had a larger sample. In early August I brought the English versions up to date. That added 65 text links, all carried over from the German articles. As a result, 37 articles received new links from other posts, 36 of them English. 33 had enough impressions for a comparison. In the four weeks afterwards, 24 of these 33 were above the rest of the domain. The median article gained around a quarter more impressions than the rest. Before that, the same articles were unremarkable, only 13 of 32 were above average.

Editing alone was not the reason. Articles with new outgoing links and no new incoming link of their own stayed at the domain average. Otherwise unchanged articles with new incoming links rose in 13 of 19 cases. The positions for the same search queries got better for 16 of 22 articles, on average by a good one position. Nothing happened with clicks, 238 before and 241 after.

A new article does not benefit from this right away. My post on internal linking was published on 26 August and received editorial links from four other articles the next day, the English version likewise. After eleven days it stood at position 43, without a click.

At seo-kreativ.de: holds. With two caveats. The effect shows in reach and not yet in clicks. It only becomes visible in quantity. On a handful of pages there was nothing to see. 36 of the 37 articles are English. For my German pages I still lack the same comparison. If you want to test the effect on your site, take many pages at once and compare them with the rest of your domain. How I prioritise links is in Link juice and link equity.

For this I pulled two reports that did not exist a year ago: Google’s appearance filter for generative AI features and the AI report in Bing Webmaster Tools. Both are vendor figures without access to the raw data, and two separately measured properties on top. So I compare the direction and not the amount.

Google’s report provides data from 18 May 2026. In the 116 days up to 10 September it shows 191,449 impressions from AI features against 1,948,967 impressions for the whole property. That is 9.8 percent. Over the three full months the share grows: 6.2 percent in June, 10.3 in July, 13.6 in August.

The most interesting part of this report is a column that is missing. Chart, countries, devices and pages contain impressions only. No clicks, no click-through rate, no position. So the question I put to every factor in this article cannot be answered for AI features with Google’s own report. What the report can and cannot do is in Google’s GenAI reports in Search Console.

And one number I looked at twice. A single page gets 32.5 percent of my AI impressions, three pages get 46.8 percent. The strongest is not one of my SEO articles but my explainer page on the abbreviation TL;DR. Of 234 visible URLs, 125 are German, and they get 14.6 percent. Why authority does not predict AI visibility is in AI visibility depends on the topic, not the domain.

Bing shows the same pattern more clearly, because citations are counted there. How the report is structured is in AI performance in Bing Webmaster Tools. From March to August 2026, citations in Copilot answers rose from 457 to 3,974 per month, a factor of 8.7. Clicks from Bing search in the same period lay between 30 and 65 per month. In the comparison window from 18 May to 10 September there are 12,673 citations against 228 clicks. That is 56 citations per click.

One detail from Bing’s query report belongs here too. Number three among the cited queries is a sentence of twenty words about E-E-A-T in a sensitive niche. No human types like that. That is a fan-out query from the system, not a user query. Such lines also appear in my Search Console. If you recalculate ranking factors on search queries, you had better filter them out or report them separately.

What does that mean for the factors themselves? In my reading, little changes about them. Relevance stays relevance, technology stays technology. What changes is the meaning of a visibility number. Being cited does not yet give you a visitor. An impression in an AI answer is even less of one than an impression in the results list. More on that in Zero-click search, in Grounding and AI SEO and in AI Mode versus AI Overviews.

This order applies to seo-kreativ.de. On an older domain with many links and a known brand I would probably weight things differently. SEO does not work the same everywhere. Nor does it stay the way it is.

The Zyppy survey was a good occasion for me to put my own numbers next to the assessment of 131 professionals. On relevance, content and internal links they agree. The technical side is met on my site. On backlinks, brand and topical authority it looks different on my domain. That is probably less down to the study than to the age and size of my domain.

I did not find the one lever in the process. I no longer believe there is one. Many things make a little difference, and the weights shift over time. From the outside the formula cannot be seen. That is exactly what makes SEO so exciting for me.

My numbers apply to my domain and to this point in time. On your domain the same factors can work differently. That is why a look into your own Search Console is worth more than any ranking, including mine.