# Google AI Ranking Factors: What 130+ Experts Scored Highest, and the Order to Work In > Zyppy's September 2026 survey of 130+ search experts ranks 13 Google AI ranking factors. Crawl access (+2.20), query-answer match (+2.15) and brand memory in the LLM (+2.08) lead; structured data scores +0.80 and llms.txt +0.05. Here is what each factor means and the order to work in. | Field | Value | |-------|-------| | Published | 2026-09-18 | | Category | AIO | | Author | Aro Ogata | | Tags | AI Overviews, AI Mode, ranking factors, GEO, AIO, expert survey, query fan-out, entity | | URL | https://linksurge.jp/blog/en/google-ai-ranking-factors-expert-survey-2026/ | --- ## Key Takeaways: What Wins Is Being Reachable, Relevant and Known On September 16, 2026, Zyppy (Cyrus Shepard) published "Google AI Ranking Factors," a survey of more than 130 search experts who rated how much each of 13 factors affects whether a page or source is surfaced, selected or cited in Google AI Overviews and AI Mode, on a 7-point scale from +3 to −3. After working through all 13 scores and the expert comments, here is what stands out. 1. **The top three are access, query match and brand memory** — AI crawl access and snippet eligibility scored +2.20, query-answer match +2.15, and brand or entity presence in the LLM's memory +2.08. Being readable, relevant and already known beat every AI-specific tactic 2. **Rankings still matter, and fan-out rankings edged out the main query** — Organic fan-out coverage rankings scored +1.91, organic search ranking +1.89. As one expert put it, pages get cited because they rank, either for the visible query or for the fan-out queries underneath 3. **Facts have to be specific and unique** — Citable, specific facts scored +2.07 and unique first-party information +1.85. Page eleven saying what ten pages already said gets nothing. Cross-web consensus (+1.81) still matters, so you need both originality and corroboration 4. **Structured data scored +0.80 and llms.txt +0.05** — Not zero, but an order of magnitude below the leaders. Almost every expert agreed you do not need to "chunk" content for Google. Ordinary structure, headings, tables and lists, is what counts 5. **The experts themselves say they are guessing** — "Lots of these should really be 'don't know'" and "This will likely age poorly" appear in the write-up. These are expert judgments as of September 2026, not measurements Let's break down the full ranking, the top factors, how to read the bottom ones, and the order to work in. --- ## 1. The Survey: 130+ Experts, 13 Factors, a 7-Point Scale Honestly, the value of this survey lies less in the decimals than in who answered and what they were asked. The contributors include people from Ahrefs, Moz, seoClarity, Wix and HubSpot, and names like Lily Ray, Aleyda Solis, Barry Schwartz and Kevin Indig. It is a who's who of English-language search. The method had three steps. - For each factor, respondents were asked how much it affects whether a page or source is surfaced, selected or cited in Google AI Overviews or AI Mode - They rated each factor on a 7-point scale, from +3 (strongly positive) to −3 (strongly negative) - They added open-ended comments So the published numbers are the mean of expert opinions. They are not Google's internal weights and not log data. With that caveat in place, the ordering becomes informative. ![Google AI Ranking Factors: How 130+ Experts Scored 13 Signals](/blog/images/google-ai-ranking-factors-expert-survey-2026/en-01.png) | Rank | Factor | Mean score | |:--|:--|--:| | 1 | AI crawl access & snippet eligibility | +2.20 | | 2 | Query-answer match | +2.15 | | 3 | Brand / entity in LLM memory | +2.08 | | 4 | Citable / specific facts | +2.07 | | 5 | Organic fan-out coverage rankings | +1.91 | | 6 | Organic search ranking | +1.89 | | 7 | Unique / first-party information | +1.85 | | 8 | Cross-web consensus & corroboration | +1.81 | | 9 | Source / publisher reputation | +1.78 | | 10 | Extractable content structure | +1.69 | | 11 | Answer prominence | +1.65 | | 12 | Structured data | +0.80 | | 13 | llms.txt file | +0.05 | The top eleven sit between +1.65 and +2.20, so the gaps are small. Only the last two fall away clearly. Read it as "everything in the top group works; the bottom two are a different category." > References: > [Google AI Ranking Factors: What 130+ Experts Taught Us - Zyppy Signal](https://signal.zyppy.com/p/google-ai-ranking-factors) --- ## 2. The Top Three: Access, Query Match and Brand Memory ### Why crawl access came first Factor one, AI crawl access and snippet eligibility (+2.20), is whether Google can crawl the page and is allowed to use its content in snippets and AI features. Fili Wiese of Search Brothers called access "a hard gate ahead of any ranking consideration." The easy thing to miss is not noindex but the snippet controls. A page with `nosnippet` (or `max-snippet:0`) can rank without its text ever being used in an AI answer. A positive `max-snippet` value only caps how much text can be used, and `data-nosnippet` excludes only the marked section. Check for directives you did not mean to leave in place. With Google adding new user agents and exclusion controls, an audit of robots.txt and meta directives comes before any other tactic. ### Query-answer match is the same number one as classic SEO Factor two, query-answer match (+2.15), is how directly the page answers the user's question or the question Google is trying to resolve. In Zyppy's parallel survey of traditional ranking factors, search intent match was also number one. Natzir's line sums it up: "without a query-answer match, there is nothing to cite." ### LLM memory: the most powerful factor and the hardest to move Factor three, brand or entity in LLM memory (+2.08), is how strongly the brand, person or entity is represented in the model's existing knowledge. Dan Petrovic of DEJAN calls this parametric memory, or "primary bias," and notes that "even if you're a grounding source, if you don't have the model's love you won't be recommended as much. Entering the model's training data is hard to do on your own, much harder than link building." Jan-Willem Bobbink of GSC Wizard adds that you first have to be considered to make it into the fan-outs, and that consideration is memory-based. Here's the thing: this factor is third on the "what works" list and close to last on the "what you can change this quarter" list. The only path is becoming a defined entity that multiple sources describe consistently, over time. We cover how in "[Entity Authority Decides AI Search Visibility](/blog/en/entity-authority-ai-search-2026/)." > References: > [Google AI Ranking Factors: What 130+ Experts Taught Us - Zyppy Signal](https://signal.zyppy.com/p/google-ai-ranking-factors) > [AI Features and Your Website - Google Search Central](https://developers.google.com/search/docs/appearance/ai-features) --- ## 3. Rankings Still Matter, but Not Only for the Visible Query Factor five, organic fan-out coverage rankings (+1.91), and factor six, organic search ranking (+1.89), are a near tie. What interests me is that the experts put fan-out slightly ahead. Fili Wiese's explanation was the clearest: AI Overviews and AI Mode run on retrieval-augmented generation grounded in the core Search ranking systems, with query fan-out generating related subqueries behind each response. "Pages get cited because they rank, either for the visible query or for the fan-out queries underneath." Natzir went further: "Ranking good for the subqueries is literally how you get in." Jamie Indigo of Not a Robot added that organic search "is one of the few models that retrieves for every interaction," which is why it plays such a large role in the retrieval step. Abandoning rankings to chase AI visibility is not an option today. The practical shift is in the unit of optimization. In Maryanna Franco's words, it moves from "rank the page for the query" to "be the best supporting source for one or more information needs generated by the query." Map the subqueries your main questions break into, and hold a heading that answers each one. Our "[Beginner's Guide to Query Fan-Out](/blog/en/query-fan-out-guide-2026/)" walks through the mechanics, and [LinkSurge](https://linksurge.jp/)'s query fan-out analysis shows how a given query decomposes. > References: > [Google AI Ranking Factors: What 130+ Experts Taught Us - Zyppy Signal](https://signal.zyppy.com/p/google-ai-ranking-factors) > [What is Query Fan-Out? Understanding the Hidden Queries Driving AI Search - Ahrefs](https://ahrefs.com/blog/query-fan-out/) --- ## 4. Facts Must Be Specific, Unique and Corroborated Factor four, citable and specific facts (+2.07), factor seven, unique first-party information (+1.85), and factor eight, cross-web consensus (+1.81), belong together. Geoff Kenyon of Pomar: specific facts and first-party information "are evaluated separately here, but when they're combined it is particularly powerful in earning citations. A specific number that's already everywhere is not helpful." Melissa Popp of RicketyRoo was blunter: if ten pages already say roughly the same thing, "page eleven needs to bring something those ten didn't. Original data, a test somebody actually ran, a firsthand observation, a process with a name and steps." Factor eight is the counterweight. If you say your brand is amazing and 99 other sites disagree, the AI will not trust you. Basic facts should agree with the rest of the web, and then you add the one thing nobody else has. Everett Sizemore named the tension directly: "The consensus thing can be a trap. How do you offer information gain if you're saying the same thing everyone else is? And what if everyone else is wrong?" In my own work the rule of thumb is "one first-party fact, two corroborations." Put one number you measured or one process you tested on the page, and keep the surrounding general facts aligned with authoritative sources. Whether that fact gets lifted depends less on where it sits on the page than on whether the block it sits in stands alone, which is what our own study of 277 real AI Overviews citations found: "[Where Does AI Search Actually Quote From?](/blog/en/ai-citation-block-position-research-2026/)" --- ## 5. Reading the Bottom Factors: Structured Data +0.80, llms.txt +0.05 ![Three Stages to Get Picked by Google AI](/blog/images/google-ai-ranking-factors-expert-survey-2026/en-02.png) ### Extractable structure and answer prominence are ordinary hygiene Factor ten, extractable content structure (+1.69), and factor eleven, answer prominence (+1.65), close out the top group. The interesting detail is chunking. Zyppy notes that respondents mentioned "chunking" eight times, and almost all agreed you do not need to chunk content for Google to understand it. Clearly organized content, meaning headings, tables, lists and proper sentence structure, remains "hugely important." On prominence, Fabio Ricotta of Agência Mestre advised to "answer the query head-on and structure the answer so the system extracts it without effort," and Will Scott of Search Influence noted that "humans and machines like quick answers." This is not the same as "put the conclusion at the top and stop." In our analysis of 277 real AI Overviews citations, only 30.7% came from the page's opening lead; 69.3% came from heading blocks in the body. A block that answers on its own beats a summary at the top. ### Structured data: not zero, but a different order of magnitude Factor twelve, structured data (+0.80), scored less than half of the leaders. Zyppy writes that there is a new debate every week about whether AI systems use it, that Google definitely uses it to understand pages for organic ranking, and that it "likely has at least some influence" on AI answers. Andy Chadwick drew the useful boundary: "The exception is product and merchant data, which feeds the Shopping Graph and does real work in AI Mode shopping results." That lines up with Google's own guidance, which says no special schema is needed for AI Overviews or AI Mode. It is also why we removed an unverified FAQ-schema citation multiplier from our own product, a story told in "[Does FAQ Schema Really Get You Cited 3.2x More?](/blog/en/faq-schema-ai-citation-myth-2026/)" ### llms.txt: +0.05 Factor thirteen, the llms.txt file (+0.05), is effectively zero. Andrew Shotland of Local SEO Guide offered "An LLMs.txt file and $8 will get you a decent Boba," and Przemysław Charchan reported collecting seven independent studies covering nearly 500,000 domains, all finding that AI bots do not check the file on their own. The claim is specifically about Google AI citations. llms.txt is published by Anthropic, Perplexity and others for their own documentation and is used to hand site structure to AI agents. "Doesn't help search citations" and "has no use" are different statements; we separate the two in "[What Is LLMO? A Complete Beginner's Guide](/blog/en/llmo-guide-beginners-2026/)." > References: > [Google AI Ranking Factors: What 130+ Experts Taught Us - Zyppy Signal](https://signal.zyppy.com/p/google-ai-ranking-factors) > [AI Features and Your Website - Google Search Central](https://developers.google.com/search/docs/appearance/ai-features) --- {{linksurge-card}} ## 6. The Order to Work In Zyppy groups the factors into three stages: make your content eligible (access, query match, rankings and fan-out), give Google specific facts it can use (specific facts, first-party information), and make Google trust you (LLM memory, consensus, reputation). The grouping holds up, and this is the sequence I recommend. **First, audit eligibility.** robots.txt, noindex, the nosnippet family, and JavaScript-dependent body text. If this gate is closed, nothing else counts. That is why factor one is called a hard gate. **Second, hold a direct answer to each main question at the heading level.** This satisfies query-answer match and the structure and prominence factors at once. One heading per question, with the answer complete inside that block. Completeness per block beats a conclusion at the top. **Third, cover the topics fan-out will generate.** Factor five. List the subqueries your main query breaks into, related topics, implicit concerns, comparisons, recency and rephrasings, and hold a heading or a page for each. **Fourth, add one first-party fact.** Factors four and seven. A number you measured, a process you tested, a real case. Something that is not on the pages already covering the topic, with concrete figures and names. **Fifth, build entity consistency.** Factors three, eight and nine, and the slowest to move. Get your company, product and people described the same way across sources you do not control: press coverage, industry bodies, third-party media, reference sites. **Last, structured data and llms.txt.** Harmless, but there is no reason to do them before the five steps above. The exception is product and merchant data if you sell things; that one moves up the list. To see where you stand, start by checking whether AI Overviews appear for your key queries and who gets cited. [LinkSurge](https://linksurge.jp/)'s AI Overview analysis shows AIO presence and citation sources per keyword. --- ## 7. Caveats: Expert Opinion, Not Measurement Keep these limits in mind when you use the numbers. - The scores are means of subjective ratings from 130+ experts, not Google's weights or measured data. Will Critchlow of SearchPilot said "lots of these should really be 'don't know'," and Victor Pan of HubSpot said "this will likely age poorly" - Contributors are mostly from the English-language market; the ratings are not based on Japanese or other non-English results - The factor definitions were set by Zyppy, and the factors are not independent (rankings versus fan-out rankings, specific facts versus first-party information) - No negative factors appear among the 13 published; penalties are outside this article's scope - The survey was published on September 16, 2026, and Google's AI search changes month to month What I value in this survey is not the decimals but the agreement among 130 practitioners that being readable, relevant and known outranks any AI-specific trick. Use it as consensus on direction, not as a weight table. --- ## Frequently Asked Questions ### What is the most important Google AI ranking factor? In Zyppy's September 2026 expert survey, AI crawl access and snippet eligibility scored highest at +2.20, followed by query-answer match at +2.15 and brand or entity presence in the LLM's memory at +2.08. Google can read the page, the page answers the question directly, and the brand is already known. ### Does structured data help with AI Overviews? Experts rated it +0.80, less than half of the top factors (+1.65 to +2.20). Google's own guidance says no special schema is needed for AI Overviews or AI Mode. The exception is product and merchant data, which feeds AI Mode shopping results. ### Should I add an llms.txt file? Its rated effect on Google AI citations is +0.05, effectively zero, and studies covering nearly 500,000 domains found AI bots do not read it on their own. It is still used to hand site structure to AI agents, so "no effect on search citations" is not the same as "no use." ### If I rank well organically, will I be cited in AI Overviews? Organic ranking scored +1.89 and fan-out rankings +1.91, both top-group factors. The experts' reading is that pages get cited because they rank, but ranking is a necessary condition, not a sufficient one. Specific facts and brand trust have to be there too. --- ## Conclusion: Keep the Order and Most of AI Search Is Just Good SEO AJ Kohn of Infinite Visibility Group put it well: "If you were doing user-centric SEO before the AI craze, then you're already set up to succeed in an AI search ecosystem." Confirm access, answer questions directly, cover the topics fan-out will ask, add one first-party fact, and build entity consistency over time. Work in that order and the AI-specific part is a small remainder at the end. Start by checking whether AI Overviews appear for your key queries and who is being cited. LinkSurge's [AI Overview analysis](https://linksurge.jp/aio-v5) shows AI Overviews presence and citation sources per keyword, and its [query fan-out analysis](https://linksurge.jp/) shows the subqueries your main queries decompose into.