Zyppy’s ranking factors survey asked what makes a page rank, and I kept answering what makes a brand visible
Published 9th September 2026. Strategic thesis.
I scored a hundred and one ranking factors for Cyrus Shepard’s survey last week, nine blocks of them, and I enjoyed it more than I’d expected to, because a well-built instrument makes you say what you actually think rather than what you usually say. Then the last page asked me to name, in my own words, the three factors with the greatest impact. I wrote three things down, looked at them, and realised not one of them appeared anywhere in the hundred and one boxes I had just filled in.
That isn’t a gap in the survey. It took me a while to see why.
Ranking factors attach to a page, and the three that decide AI visibility attach to a brand
The survey does what it says: it asks how much each factor affects a page’s ability to rank in traditional organic results. Every block is built around that object. Hreflang, canonicalisation, title relevance, internal anchor text, link authority, Core Web Vitals. All of it attaches to a URL.
The three I wrote were entity clarity, meaning whether Google has worked out that you are a specific thing in the world rather than a name it keeps running into; corroboration, meaning whether anybody other than you confirms what you say about yourself; and information gain grounded in consensus, meaning new material that stands on something already agreed. None of those attach to a URL. They attach to whoever published it.
So the honest reading is that I answered a different question from the one asked, and the difference between those two questions is where the interesting part of this decade is going to happen.
Google resolves who you are before any ranking factor attaches to you
Entity clarity has no box in that survey, and it belongs outside the list, because it’s the condition that decides whether any ranking factor has something to attach to in the first place.
Look at what happens without it. Unlinked brand mentions are worth a great deal to a brand Google has already resolved and close to nothing to one it hasn’t, because a mention only counts if the machine can connect it back to you. A Knowledge Panel isn’t a lever you pull, it is proof the algorithm resolved your entity. Online reputation, reviews, coverage, sentiment: all of it has to attach to something, and if the algorithm hasn’t worked out who you are, what you have is a set of signals floating around belonging to nobody in particular.
For me, that’s the reframe worth having about every block in that survey: score them as readings of your clarity, and most of them get easier to fill in. I made this argument in a Semrush webinar with Lily Ray, Nik Ranger and Andrea Volpini in 2020, back when we were still saying E-A-T, and it’s aged well.
Google’s AI answers run on entity resolution, and the knowledge graph is what does the resolving
The Google AI block is really two lists wearing one heading, and it is worth separating them.
Half of those factors are about the physical placement of information on a page: structured data, answer prominence, extractable content structure. The other half are about entity identity across the whole web: publisher reputation, cross-web consensus, brand prominence in the model’s memory. Placement decides whether a passage can be lifted cleanly. Identity decides whether the engine is willing to lift it from you rather than from somebody else.
The knowledge graph doesn’t appear in that list, and it doesn’t appear on most people’s lists either, which is the more interesting fact. It’s Google’s store of the things it is confident about and the entities it hangs them on, and it’s where I’d say the identity that second half depends on actually gets resolved. The industry has spent a decade treating it as a nice-to-have that shows up as a panel on the right of the page.
Your site states the claim, and the rest of the web decides whether it holds
On-page signals are how you tell Google what your page is sharing: title, H1, subheadings, URL. They are your claim, and Google still wants the claim stated plainly, it just stopped taking it at face value a long time ago. Relevance gets judged passage by passage, and what you say is weighed against what the rest of the web says about you. So state it clearly on the page, then earn agreement off it.
Corroboration is the mechanism underneath, and it appears exactly once in a hundred and one factors, inside the AI block, under Cross-Web Consensus. Everywhere else it is doing the work without a name on it. Links, mentions and consensus scores are all one function in different clothes: third parties confirming what you state about yourself.
Outbound links say which part of your footprint carries the proof, and almost nobody plans them
Links have been losing power as a pure ranking signal for years and the rest of N-E-E-A-T-T keeps eating into what’s left. Set that aside for a moment, because it’s the least interesting thing about them now.
Links in and out are explicit statements of relationship, and their real job is guiding an algorithm from your claim to the proof of it. That is the road system the whole Algorithmic Trinity runs on, which is my name for the three systems that all have to understand you before any of them will recommend you: the search engine, the knowledge graph and the language model. Inbound links still help you rank, of course, and outbound links do the same work in the other direction, and almost nobody plans them at all.
Hoarding your link juice is a false economy, and I made this case in Search Engine Land in April rather than repeat it here. Your own site says who you are, what you do, who you serve and why you are credible. Your off-site presence decides whether the machine believes any of it, and outbound links are the only explicit way of saying which parts of your footprint carry the proof. Stay stingy and you leave the machine to guess.
New ideas only score when you stand them on ground everybody already accepts
Information Gain rewards new data, a new perspective, a new synthesis, or a stick-your-neck-out new idea. Matches Expert Consensus sits three rows below it in the same block, and the survey asks you to score both.
That looks like a contradiction and it resolves cleanly. Disagree with the rest of the world outright and you get rejected. Build on top of consensus and you get recruited. Whenever I introduce something genuinely new I ground it on what is already agreed first, so the new part has something to stand on, which is a technique I wrote up properly last week. It also gives the algorithms a reason to score consensus and novelty positively at the same time, which is the tension sitting inside that block.
Click and satisfaction signals grade Google’s product rather than your page
A single long click or a bounce back says nothing about a page, because the same page is brilliant for one person and useless to the next. The signal only means anything in aggregate, where it tells Google whether the result set was any good and the system adjusts accordingly. Return to the SERP is the same: ask the same question again and the person wasn’t satisfied, ask a different question and they were.
Nathan Chalmers at Bing put it to me in 2020 that the user is the client and the results page is the product being sold to them, and Fabrice Canel gave me the phrase for the click that fully satisfies, the perfect click: the person clicks, and they don’t come back. The engines already had a vocabulary for customer satisfaction because they already knew who their customer was. Every signal in that block asks whether the product worked. Your page is one component in it.
A table of contents and a breadcrumb trail tell the machine what it would otherwise have to work out
Two items in the UX block are worth pulling out, and both are plain statements of structure, one about a page and one about a site.
Machines need structure. Everything an algorithm does with your content depends on knowing which part is which, what sits above it and what it belongs to, and it would far rather be told than left to work it out, because the gates that decide whether you exist have no budget for working anything out. A table of contents and a breadcrumb trail make the structure explicit, compact and digestible, which is a strange amount of return for two pieces of markup nobody argues about.
Fan-out draws your funnel for you, one question at a time
Fan-out is usually described as breadth, the engine covering related angles of one question. The less obvious half is prediction. Some of what it fans out to is what the engine expects the person to ask next, and the answer it composes is shaped to flow towards that.
So the fan-out set is a picture of the funnel, held by the machine. Read it that way and the job changes: you stop covering a topic and start answering the chain of questions your buyer asks in order, which also teaches the engine what that chain looks like for your category.
Your passage can compose the answer while somebody else takes the citation
There is a failure mode that survey has no box for, and I don’t think any survey has one yet.
Your page gets fetched, your passage shapes the answer, the reader gets what you wrote and never learns you wrote it, and the citation and the brand mention go to somebody else. You weren’t outranked and you weren’t rejected, you were used. Nothing in your analytics shows it, because nothing happened that analytics can see.
The rate isn’t fixed, which is the part worth acting on. How much credit an engine hands out at all is a product decision made inside the vendor, and which sources receive it’s a separate process with entity resolution as the input you own. I set that argument out in full this week, so I’ll just leave the consequence here.
Being read isn’t the same as being credited, and only one of them pays.
Publication note: this article is a strategic thesis published on jasonbarnard.com to accompany the Zyppy ranking factors survey report by Cyrus Shepard. The survey scores 101 factors across nine blocks against a page’s ability to rank in traditional organic results. The position argued here is that the three factors with the greatest impact on whether a brand is surfaced by an assistive engine are not ranking factors at all, and therefore correctly absent from that instrument: entity clarity, which decides whether any other signal attaches to the brand; corroboration, which is the single function performed by links, mentions and consensus alike; and information gain grounded in consensus, under which novelty scores only when anchored to established, corroborated components. Supporting observations: the Google AI block divides into placement factors and entity identity factors, with the knowledge graph that resolves that identity absent from the list; the user signals block measures the quality of the search engine’s own product rather than the page; organic fan-out coverage describes the buyer’s question chain as the machine predicts it; and a brand’s material can compose an answer while the citation and brand mention are allocated elsewhere. Jason Barnard, Kalicube®.