We bring together the complete study here, factor by factor, with the original scores. Because collecting data without interpreting it has limited value, each section also includes The Key SEO’s interpretation: what changes in practice, what is a meaningful signal rather than noise, and where to allocate budget in 2026.
Primary source: Cyrus Shepard and Dawn Shepard,
Google Ranking Factors Expert Survey 2026, Zyppy Signal, September 9, 2026. All numerical data in this article belong to this study.
Consult the original study.
How the study was conducted (and how it should be read)
The methodology shapes everything that follows, so it is worth establishing it before looking at the numbers.
- Sample: 131 SEO professionals, among independent consultants, agencies and internal teams of companies such as Ahrefs, Moz, HubSpot, Wix, iPullRank, SearchPilot and BuzzStream.
- Factors assessed: more than 100 candidate factors, narrowed down to 103 final factors.
- Scale: 7 points, from +3 (strongly positive effect) to -3 (strongly negative effect). Each score presented is the mean of the responses.
- Additional open question: “what are the three most important ranking factors?”, without predefined options. That’s where the top 10 comes from.
- Dataset: 13,665 data points and 2,620 minutes of reporting.
One point shapes the interpretation of the entire study:
this is not a correlation study or a controlled test.It measures expert beliefs. What it captures is the market’s operational consensus: the factors experienced professionals rely on when deciding where to invest their time and budget. This is highly valuable for prioritisation, but has limited value as evidence of causality. We return to this distinction in the limitations section.
The ten factors most frequently cited by experts in 2026
In the open question, each respondent freely named the three factors they considered most important. The percentage shows the proportion of respondents who included each factor in their top three.
| # |
Factor |
% of respondents |
| 1 |
Relevance |
57,1% |
| 2 |
Backlinks |
54,8% |
| 3 |
Content quality |
47,6% |
| 4 |
Authority and trust |
36,5% |
| 5 |
Behavioural and click signals |
29,4% |
| 6 |
Brand signals |
27,0% |
| 7 |
User satisfaction |
19,8% |
| 8 |
Website technical health |
17,5% |
| 9 |
Topical authority |
14,3% |
| 10 |
Internal links |
11,1% |
The Key SEO’s interpretation of the top ten
What matters most to us in this list is not the order, but its
structure. Seven of the ten items—relevance, quality, authority and trust, click signals, brand signals, satisfaction and topical authority—describe the same thing from different perspectives:
the extent to which an entity is recognised as the appropriate answer to a search intent. The other three—backlinks, technical health and internal links—are the mechanisms through which that recognition is transmitted and distributed.
This directly affects how we design projects. It no longer makes sense to treat technical SEO, content and link building as parallel fronts with separate budgets and reports. The correct interpretation is a chain: architecture makes the site readable; content establishes relevance and information gain; links and mentions transmit trust; and user behaviour validates or invalidates everything else. When one link fails, upstream investments do not convert.
Also notice what
does not appear in the list: no variation of traditional on-page optimisation, no meta-tag element and no speed indicator. These have not disappeared; they have moved into the category of requirements rather than competitive advantages.
1. Content relevance
Among the 103 factors tested,
Search Intent Match The definition used goes beyond the matching of keywords: it measures how well the page satisfies the type of result or task that the user seems to want.
| Factor |
Score (-3 to +3) |
| Search intent match |
+2,60 |
| Page semantic relevance |
+2,34 |
| Topical authority of the site |
+2,29 |
| Title relevance |
+2,26 |
| Passage-level relevance |
+1,99 |
| Language matching |
+1,95 |
| Geographical relevance |
+1,89 |
| Keywords in H1 |
+1,67 |
| Keywords in subheadings |
+1,27 |
| Keywords in URL |
+1,07 |
| Exact-match domain (EMD) |
+0,94 |
| Meta description |
+0,22 |
What the experts highlighted
Melissa Popp summarised the essential point: a page can have flawless technical execution, well-written headings and an exemplary URL structure, yet none of this compensates for answering a different question from the one asked. Andy Chadwick added a layer we consider decisive: when ten pages already match the same intent and cover the same ground, the differentiator becomes
information gain: what one page adds that the others do not. Shai Aharony noted the surprise of continuing to rank exact-match domains for competitive terms with a fraction of the effort devoted to the main website.
The Key SEO’s interpretation
Search intent is an architecture decision, not a copywriting decision. The +2.60 score confirms what we have argued for years: intent matching is won or lost when deciding which pages exist, not when writing the copy. A commercial page trying to answer informational intent is not fixed by adding paragraphs, but by separating the two intents into different URLs and explicitly connecting them.
Hence our emphasis on
separation between commercial content and informational content. When a service page tries to rank simultaneously for “what is X”, “how much does X cost” and “hire X”, it dilutes the semantic relevance (+2.34) of all three and satisfies none. The alternative is a cluster: a conversion-focused commercial page and informational pages covering the topic, connected with descriptive anchor text.
Passage-level relevance (+1.99) provides a technical argument for modular content. If Google evaluates passages as well as whole documents, a long article should be built as a sequence of self-contained modules, each with a subheading stating the question and a direct answer in its opening lines. This format also supports eligibility for AI-generated answers.
About EMD (+0.94): is market information, not a recommendation. An exact-match domain can accelerate short-term projects in less established niches, but it is the opposite of a brand strategy, as section 4 explains. Whoever builds a ten-year asset does not build on an EMD.
About the meta description (+0.22): the common claim that “meta descriptions are useless” is wrong. The study says they are not a ranking factor. It does not say they do not affect click-through rate. Section 5 gives organic CTR a score of +1.83. In other words, the meta description indirectly influences ranking through behaviour. It still deserves careful writing on important pages, with clear priorities rather than mass production.
2. Backlinks
The study highlights the continuing importance of links. Two backlink factors receive some of the highest scores in the survey, while spam links appear among the strongest negatives.
| Factor |
Score (-3 to +3) |
| Links from highly trusted domains |
+2,47 |
| Links from topically relevant pages |
+2,22 |
| Domain or website authority |
+1,89 |
| Page-level authority |
+1,85 |
| Links from pages with real traffic |
+1,72 |
| Unique domains linking to the URL |
+1,71 |
| Anchor text diversity |
+1,18 |
| Total number of links to the URL |
+1,05 |
| Exact-match anchors |
+0,86 |
| Nofollow backlinks |
+0,50 |
| Spam backlinks pointing to the URL |
-1,26 |
What the experts highlighted
Nitin Manchanda summarised the section’s central thesis: a hundred links with a real audience are worth more than a thousand that nobody clicks. Andrew Shotland reported a case linking backlinks to AI visibility, with a client’s ChatGPT presence quadrupling in the week after a linked mention in a Reuters article. Geoff Kenyon explained exact-match anchors: they produce real gains up to a threshold and suppress performance beyond it, with the relevant variable being the
proportion of the link profile, rather than the absolute count.
The Key SEO’s interpretation
The hierarchy of scores challenges buying links on metrics alone. Notice the gap between “links from highly trusted domains” (+2.47) and “total number of links” (+1.05). Much of the link building market still focuses on the second metric, which the expert panel itself considers almost half as relevant. In the market’s perception, one link from a trusted, topically relevant domain is worth more than two generic links.
In our view, “links from pages with real traffic” (+1.72) is the most underestimated factor in this table. It makes a normally vague qualification criterion actionable. Before accepting a placement, the question is not the publisher’s authority metric, but whether that particular page receives visitors. An orphaned page on a strong domain is a weak link. A page with an audience on an average domain is a strong link.
On domain authority (+1.89): this is a metric supplied by data providers, not a signal Google declares it uses. Its high score reflects its usefulness as a practical proxy, rather than proof that this metric exists within Google’s algorithm. We use it as a screening tool, never as the sole basis for a decision.
About anchors (+0.86 for exact match, +1.18 for diversity): diversity scores higher than exact matching. The operational implication is simple: monitor the distribution, not the count. A healthy profile is dominated by branded anchors and naked URLs, with exact matches accounting for a minority.
The -1.26 score for spam needs context. Google says it ignores most spam links. The respondents’ assessments nevertheless place them among the study’s main negatives. In practice, we retain periodic link-profile audits for projects with aggressive link-buying histories or exposure to negative SEO attacks, rather than abandoning monitoring on the basis of general public statements.
3. Content quality
This section is where the 2026 study moves furthest from previous editions. Attributes that are difficult to produce at scale occupy the top of the table, while easily replicated content sits near the bottom.
| Factor |
Score (-3 to +3) |
| Original research and first-party data |
+2,19 |
| Website-level content quality |
+2,07 |
| Content originality |
+2,02 |
| Information gain |
+1,93 |
| Content freshness |
+1,80 |
| Demonstrated experience |
+1,78 |
| Content depth |
+1,70 |
| Page answers related questions |
+1,66 |
| Effort invested in the content |
+1,61 |
| Author expertise and authority |
+1,60 |
| Prominence of the main answer |
+1,59 |
| Factual accuracy |
+1,52 |
| Alignment with expert consensus |
+1,23 |
| Image and video quality |
+1,02 |
| Page links to relevant websites |
+0,94 |
| Image alt attributes |
+0,77 |
| Content length |
+0,52 |
| AI-generated content |
-0,50 |
| Pages with affiliate links |
-0,53 |
| Stock photography |
-0,66 |
| AI-generated content at scale |
-1,82 |
What the experts highlighted
Ross Simmonds described the economic mechanism: AI has made content cheap, anyone can produce a thousand articles overnight, Google knows it, and quality therefore becomes a filter. When the cost of publishing falls to zero, the cost of trust rises. Gianluca Fiorelli identified the resulting tension: excessive consensus makes a page invisible through lack of differentiation, while too much information gain risks a penalty for excessive contrarianism. Abby Gleason offered the empirical counterpoint, reporting positive results from AI generation at scale on a site with a strong foundation of unique and user-generated content, subject to intensive human editorial review.
The Key SEO’s interpretation
The two ends of the table are the same variable. Original research and first-party data at the top (+2.19) and AI-generated content at scale at the bottom (-1.82) measure, in our view, the marginal cost of production. What scores highly cannot be replicated without privileged access to data, customers or field experience. What scores poorly can be reproduced by any competitor within hours. This is the most useful editorial budget-allocation rule we draw from the entire study.
The distinction between AI (-0.50) and AI at scale (-1.82) is the most misinterpreted finding in this section. The penalising factor is not model use, but publishing in volume without added value. In practice, we use AI at almost every editorial stage, from research to structuring and review, and maintain two rules: nothing is published without human factual verification and an original contribution, and no calendar is sized according to the model’s production capacity.
Content length at +0.52 helps put the word-count debate in perspective. This is the lowest score of all positive factors in the section, below alt tags. Briefs specifying word count as a quality requirement are optimising the table’s least relevant factor. The correct replacement combines coverage of related questions (+1.66) and depth (+1.70), measured through topical completeness rather than word count.
Website-level content quality at +2.07 is an argument against allowing an archive to accumulate unchecked. The assessment is not per page, but per site. Old, thin content without traffic is not neutral: it dilutes. The resulting action is a periodic inventory audit with an explicit decision for each URL: update, consolidate or remove. On long-established projects, this is frequently the intervention with the best return per hour invested.
About the tension pointed out by Fiorelli: We read information gain (+1.93) and alignment with consensus (+1.23) as an editorial positioning instruction, not a contradiction. The defensible position builds on established consensus and differentiates through application, using first-party data, methodology and cases. Contradicting factual consensus is a risk. Adding evidence that consensus has not yet incorporated is an advantage.
4. Brand and reputation
This is the block that best explains why technically average sites surpass technically impeccable websites.
| Factor |
Score (-3 to +3) |
| Online Reputation and Trust |
+2,01 |
| Branded search volume |
+1,98 |
| Unlinked brand mentions |
+1,30 |
| Domain history |
+1,16 |
| Website with Google Knowledge Panel |
+1,15 |
| Visible contact information |
+1,09 |
| Visible author profiles and biographies |
+1,05 |
| Domain age |
+0,98 |
| Social signals for the URL |
+0,90 |
| Social profiles linked to the site |
+0,76 |
| Investment in Google Ads |
+0,27 |
What the experts highlighted
Dan Petrovic described the mechanism as a popularity contest: once relevance to a query is established, Google does not reward content that is ten times better, and overly specialised content can even harm visibility, because the aim is to appeal to the largest possible audience and be mentioned, shared and linked to. Jan-Willem Bobbink went further, calling brand signals the strongest factor, sufficient for a technically poor site with poorly optimised content to survive as long as people search for and discuss the brand. Roger Montti summarised the thesis with a phrase circulating since 2012: branded search is the new link. Will Critchlow clarified advertising investment: no direct effect, but the ability to influence other signals.
The Key SEO’s interpretation
Branded search volume at +1.98 changes how budgets should be viewed. If searches for the company’s name act as a trust signal, investment in awareness ceases to be a branding expense and becomes
organic acquisition with a delayed effect. This changes how we assess top-of-funnel campaigns. Public relations, sponsorships and participation in industry events should be measured not only by direct traffic, but also by changes in branded search volume and subsequent organic performance.
Unlinked mentions at +1.30 score higher than some link-volume measures. An editorial mention without a link scores above “total number of links” (+1.05) in the previous section. This supports redirecting some outreach effort towards ensuring that the brand is named in the right context as an identifiable entity, rather than negotiating only for links. In projects focused on GEO and AEO, this also supports retrieval by AI systems.
Knowledge panels (+1.15), visible contact details (+1.09) and author profiles (+1.05) form a coherent group. They’re all signs of
entity disambiguation. The common denominator is allowing machines to connect the site to a verifiable person or organisation. It requires structured markup, consistent data between sources and author profiles with genuine histories, not copywriting.
Petrovic’s observation should be interpreted carefully. The idea that overly specialised content harms visibility is an observation about reach, not quality. In B2B and technical settings, the page appealing to the largest number of people is rarely the one that converts. Our interpretation concerns architecture: broad-reach content at the top to attract demand and mentions, deeper content at lower levels to qualify and convert. These objectives do not compete when they occupy different pages.
Google Ads at +0.27 addresses a recurring commercial myth. There is no direct effect. There are legitimate indirect effects through awareness and traffic that generates mentions and branded searches. This is different from buying organic rankings.
5. User signals
Following the evidence made public in the antitrust trial and the leaked API documentation, user signals show the largest increase in perceived importance compared with previous editions.
| Factor |
Score (-3 to +3) |
| User satisfaction and task completion |
+2,14 |
| Long clicks |
+1,95 |
| Organic CTR |
+1,83 |
| User-selected preferred source |
+1,48 |
| Return to the search results |
-1,04 |
What the experts highlighted
AJ Kohn kept Navboost central and offered an unusual recommendation: sites should build their own satisfaction metric, typically segmented by page type, to approximate what Google measures, because rankings have a ceiling when click signals are worse than competitors’ in the same results block. Gerry White distinguished bounce rate, which he considers a poor metric, from pogo-sticking, which he considers a clear signal. Diego Ivo raised the question that defines the next decade: when browsing becomes predominantly agentic, how will these signals be measured?
The Key SEO’s interpretation
Kohn’s recommendation is one of the study’s most actionable ideas, yet is rarely implemented. Building a satisfaction metric by page type means defining the task-completion event for each template: reading to the end and clicking the next cluster page in an article; interacting with a product page and submitting an enquiry; or clicking a contact option on a local service page. Then measure the share of organic sessions reaching that event and treat it as an SEO metric alongside rankings and impressions. This is what separates an SEO operation from a CRO operation with separate reports.
Returns to the results at -1.04 make it necessary to review the promise made in the snippet. An aggressive title that overpromises improves CTR (+1.83) and destroys the next signal. Both factors must be optimised together, never separately. Audits of titles and descriptions must therefore cross-reference CTR with on-page behaviour, and higher CTR without increased dwell time is a warning, not a gain.
Diego Ivo’s question is the one that concerns us most in medium-term planning. As more queries are resolved by agents and generated answers, human click signals become less abundant in entire segments. The likely consequence is not the disappearance of behavioural signals, but their partial replacement by citation and selection signals in AI contexts. Building a brand, first-party data and a clearly identifiable entity supports both environments. Building solely for clicks depends on a signal whose future role remains uncertain.
6. Technical SEO
Respondents generally treated technical SEO as an entry requirement. It enables other work, but does not turn average content into exceptional content.
| Factor |
Score (-3 to +3) |
| Correct canonicalisation |
+1,56 |
| HTTPS |
+1,28 |
| Crawl frequency |
+1,06 |
| Hreflang accuracy |
+1,02 |
| Structured data |
+0,96 |
| URL included in XML sitemap |
+0,80 |
| HTML and W3C validation |
+0,38 |
| URL length |
+0,08 |
| Redirect chains |
-0,93 |
| Page requires client-side JavaScript |
-1,04 |
| Duplicate content |
-1,19 |
| Blocked by robots.txt |
-2,22 |
What the experts highlighted
Marshall Simmonds defined technical SEO as a
commodity: it does not necessarily win the race, but is perfectly capable of losing it. Dana DiTomaso illustrated this with a case where misconfigured canonicals pointed to unrelated pages; correcting them produced an almost immediate improvement of around 60% in impressions and clicks. Maryanna Franco offered the framing we consider most current: assessing technical SEO as infrastructure for machine comprehension as well as retrieval.
The Key SEO’s interpretation
Notice the asymmetry of the table. The highest positive factor scores +1.56. The lowest negative is -2.22, the most extreme negative in the entire study, across all categories. Statistically, technical SEO is a domain of risk management, not advantage creation. The management implication is clear: a technical project should eliminate barriers, not maximise scores. After that, the marginal return on further technical hours falls rapidly and the budget should move to content, brand and links.
Structured data at +0.96 is the most easily misread number in the table. As a perceived direct ranking factor, its score is modest. As a condition for rich-result eligibility and a way to disambiguate entities for generative systems, its role is different. We continue to implement connected structured markup using
@id across our projects. The justification extends beyond traditional rankings to entity readability, reflected in the +1.15 score for a knowledge panel in section 4.
Client-side JavaScript dependence at -1.04 remains underestimated. In projects built with modern frameworks without server-side rendering, this is usually the most expensive problem to fix and the last to be detected, because it only appears at scale. Rendering decisions belong in the project architecture phase, not the optimisation phase.
Duplicate content at -1.19 and canonicalisation at +1.56 concern two sides of the same problem. In ecommerce with facets, filters and parameters, this is almost always the simultaneous source of wasted crawl budget and cannibalisation. It is also the clearest example that technical SEO is not separate from content strategy: deciding the facet architecture means deciding the semantic architecture.
7. Internal links and site architecture
Historically treated as less important than external links, internal links receive scores that challenge that status.
| Factor |
Score (-3 to +3) |
| Authority of internally linking pages |
+1,92 |
| Relevance of internal anchor text |
+1,66 |
| Number of internal links to the URL |
+1,49 |
| Links from global navigation |
+1,47 |
| Click depth from the homepage |
+1,06 |
| Internal anchor text diversity |
+0,91 |
| Orphaned page |
-1,79 |
What the experts highlighted
Gaston Riera mentioned discussing with colleagues the idea that a good internal link is worth as much as a good external backlink. Fili Wiese described interlinking as the only ranking lever a website controls from end to end: deciding which pages accumulate equity, which pages Google considers important and how topics connect.
The Key SEO’s interpretation
Place the two tables side by side. “Authority of internally linking pages” scores +1.92. “Domain or site authority” in the backlink section scores +1.89. In these 131 professionals’ perception, internally distributed authority is as decisive as authority received externally. The difference is that the first is free, immediate and fully controllable, while the second requires negotiation, budget and time.
Orphaned pages at -1.79 are among the study’s three strongest negatives, behind robots.txt blocking (-2.22) and AI-generated content at scale (-1.82). It is also the quietest problem, because it does not trigger errors in ordinary monitoring tools. Audits of long-established sites frequently find dozens of pages without any incoming internal link, usually left by migrations, old campaigns or publication outside the category structure. Fixing this takes hours and produces a measurable effect.
Internal anchor relevance at +1.66 has implications for interface design. Patterns such as “learn more”, “see more” and “click here” are semantically neutral and waste the signal. The alternative does not require sacrificing usability: well-written descriptive anchor text is better for both the user and the machine. This is a decision where SEO and UX converge without tension.
The strategic reading we make of this block is that of topical authority architecture. A cluster is not defined by sharing a topic, but by sharing a link structure: a central page receives links from supporting pages and links back to them, with anchors declaring their semantic relationship. Without this structure, articles about one topic are merely articles. With it, they are a topical authority asset, which section 1 showed to be worth +2.29.
8. User experience (UX)
The most controversial block of the study, with significant divergence between specialists on speed and Core Web Vitals.
| Factor |
Score (-3 to +3) |
| Usability on mobile devices |
+1,66 |
| Readability |
+1,28 |
| Website speed |
+1,23 |
| Equivalent mobile and desktop content |
+1,17 |
| Core Web Vitals |
+0,92 |
| Breadcrumbs |
+0,81 |
| Table of contents |
+0,53 |
| Popups and interstitials |
-1,31 |
| Advertising density |
-1,31 |
What the experts highlighted
Ian Lurie offered the most useful framework: UX factors matter primarily because they influence behaviour, which feeds Navboost and other system components. Ryan Jones was most direct in the opposite direction, calling speed and Core Web Vitals the most overrated elements in SEO, with a binary view of whether a page loads in time. Greg Gifford introduced context: speed matters greatly in large ecommerce, while local car-dealership pages with detailed images commonly, and even expectedly, take a few seconds.
The Key SEO’s interpretation
The difference between Jones and Gifford is not a contradiction, but dependence on context, which the study does not capture. An average of +0.92 for Core Web Vitals conceals two distributions: sectors where speed determines conversion and sectors where it does not. The operational conclusion is not to ignore CWV, but to stop treating them as a standalone SEO objective with a separate report. They are a product objective with a revenue impact, and the organic benefit comes through behaviour.
Mobile usability at +1.66, above speed and CWV, is the main result to retain from this section. With mobile-first indexing established, parity becomes a central concern: menus hiding content, blocks failing to render on small screens, unreadable tables, or information available on desktop but absent on mobile. Equivalent mobile and desktop content receives a score of +1.17.
Popups and advertising density each score -1.31, the section’s two negative factors. Both are business decisions as well as SEO concerns. This is useful evidence to bring to discussions about covering a page with a lead-capture popup moments after entry. The potential cost goes beyond annoyance to poor user engagement and returns to the search results (-1.04).
Factors that are not worth investing in
A study of 103 factors is useful for what it rates low as well as high. The following elements are perceived by respondents as having relatively little effect, yet still consume substantial time in many SEO operations:
- URL length (+0.08). Practically zero. Rewriting existing URLs because of their length is a migration risk with no return.
- Meta description (+0.22). Zero as a ranking factor. It retains value through CTR, prioritising revenue-generating pages.
- HTML and W3C validation (+0.38). Do not confuse code hygiene with ranking factor.
- Content length (+0.52). Below image alt attributes. A word-count target alone is a weak quality requirement.
- Table of contents (+0.53). Good for usability and internal navigation, irrelevant as an isolated signal.
- Nofollow backlinks (+0.50). Slightly positive, far from justifying campaigns built around them.
None of these items should be completely abandoned.
allocation: when a team spends a quarter rewriting meta descriptions and URLs, it is working at the bottom of the table while the top remains untreated.
What this study means for GEO and AEO

The survey explicitly concerns traditional organic rankings, and its authors announced a second study devoted to visibility in generative answers. Nevertheless, some signals are relevant to both environments, creating the connection we discuss here.
Several highly rated factors also support the conditions for selection and citation by AI systems. Original research and first-party data (+2.19) solve the source problem: a generative system needs something to cite that is not on a thousand other pages. Information gain (+1.93) solves redundancy. Unlinked brand mentions (+1.30) and online reputation (+2.01) build the entity, which is these systems’ retrieval unit. Passage-level relevance (+1.99) literally describes how content is fragmented and retrieved in retrieval-augmented generation architectures.
The case reported by Andrew Shotland is the most concrete example of this convergence: an editorial mention with a link in a highly trusted publication produced a fourfold increase in the client’s ChatGPT visibility in about a week. The same asset, an editorial link from a trusted source, serves both channels, with different propagation speeds.
Behavioural signals are the most relevant point of divergence. Organic CTR, long clicks and returns to results assume a human navigating search results. That assumption becomes weaker as more queries are resolved through conversational interfaces and agents. We are not recommending abandoning behavioural signals; we recommend
avoiding a strategy built entirely around them. Brand, first-party data, entity clarity and semantic structure remain relevant across both environments.
The Key SEO prioritisation matrix

A table of scores is not a work plan. To turn the study into decisions, we consider two axes: perceived impact and the degree of control the team has over the factor.
High impact and high control: act first
- Align each URL with an intent, separating commercial and informational pages.
- Improve internal linking, fix orphaned pages and use descriptive anchors.
- Remove technical barriers involving robots.txt, canonicals, duplication and JavaScript dependence.
- Audit the content inventory and decide which URLs to update, consolidate or remove.
This quadrant does not depend on third parties or an external budget and contains some of the study’s highest absolute values, including its two worst negatives. It is where every project we undertake begins.
High impact and partial control: build continuously
- Original research and first-party data production to feed content, links and AI citations simultaneously.
- Earn editorial links from trusted, topically relevant domains, checking traffic on the linking page.
- Build brand awareness, measured through branded searches and mentions.
- Define and monitor a satisfaction metric for each page type.
These initiatives have quarterly or annual horizons and compounding returns. They do not produce results in the first month and explain most of the difference after a year.
Marginal impact: maintain appropriate hygiene
- Meta descriptions, URL length, code validation and word counts.
- Core Web Vitals beyond an acceptable operating threshold.
- Social profiles and signals: useful for entity consistency, irrelevant as an isolated lever.
Study limitations (and why we mentioned them)
Presenting these figures without explaining the methodology would repeat the error that the study itself helps correct. Three notes.
First: perception is not correlation. The survey measures what 131 professionals believe, not observed algorithm behaviour. A factor may receive a high score because it is important or because it is widely discussed. These data help prioritisation; they do not constitute proof.
Second: availability bias. The sharp rise in perceived user-signal importance follows two prominent events: the antitrust trial and the API documentation leak. It is plausible that some of the increase reflects greater exposure to the subject, rather than algorithmic changes alone. This does not invalidate the survey, but requires contextual interpretation.
Third: averages conceal context. Core Web Vitals are the clearest case, with two experts describing opposite realities that are equally true for their sectors. An average of +0.92 describes neither. Applying these figures to a project must account for its sector, query type and site maturity.
Frequently Asked Questions
How many ranking factors were evaluated in this study?
The study assessed 103 final factors, selected from more than 100 candidates, through responses from 131 SEO professionals on a -3 to +3 scale, producing 13,665 data points.
What is the most important ranking factor according to experts?
Search intent match, at +2.60, received the highest score in the study. In the open question asking for the three most important factors, relevance was cited by 57.1% of respondents.
Are backlinks still important in 2026?
Yes. Links from highly trusted domains score +2.47 and links from topically relevant pages +2.22, both among the highest values. The emphasis is on qualification: domain trust, topical relevance and real traffic on the referring page.
Does AI-generated content harm rankings?
The study distinguishes two cases. AI-generated content in general scores -0.50, a slight negative. AI-generated content at scale without added value scores -1.82, one of the strongest negatives. The penalising factor is publishing in volume, not using the tool.
Which technical SEO issues received the most negative scores?
Robots.txt blocking, at -2.22, is the study’s strongest negative. It is followed in this section by duplicate content (-1.19), client-side JavaScript dependence (-1.04) and redirect chains (-0.93).
Are internal links really as valuable as external links?
The study gives the authority of internally linking pages +1.92, practically level with external domain authority at +1.89. Orphaned pages score -1.79. It is the highest-impact lever under the team’s complete control.
Source and credits
All the quantitative data presented in this article come from the
Google Ranking Factors Expert Survey 2026, conducted and published by
Cyrus Shepard and
Dawn Shepard in Zyppy Signal on September 9, 2026, based on responses from 131 SEO professionals. Observations attributed to experts paraphrase statements collected in that survey. The analysis, interpretation and recommendations labelled as The Key SEO’s interpretation are ours and should not be understood as positions taken by the survey’s authors or respondents.
Google Ranking Factors Expert Survey 2026, Zyppy Signal