SEO Ranking Factors vs AI Citation Factors: What Moves Each One

Google still runs on links. Every large-scale test I have run since 2017 shows a page with the stronger link profile beating a near-identical page without one. AI assistants do not compute PageRank when they answer. They retrieve passages, and a passage with zero links behind it gets cited if it answers the question cleanly. What does matter to a model is the context a link sits in, which is a separate factor below. The authority score of the linking page is close to irrelevant; the words around the link are not.

Ask an assistant for the best agencies, tools or consultants in a category and watch where the names come from. It aggregates existing best-of lists and picks the names that recur. Appearing in those lists is now the most direct lever I know for category-level AI recommendations. Google gives you very little for it unless the list also links.

Google reads a well-linked cluster of pages as expertise and ranks the whole cluster higher, so this is one of the strongest SEO levers I have. For AI it matters far less than the industry assumes. A model retrieving passages does not check how many other pages on your site cover the topic. A single excellent, self-contained passage on a thin site gets cited over a mediocre one on a 300-page authority site. The modest AI score reflects the extra retrieval surface a cluster gives you, nothing more.

This is the single biggest split. AI systems lift passages, not pages. If your paragraph only makes sense after reading the three above it, it never gets quoted. Google rewards page-level relevance and can rank a page whose answer is buried in section six. For AI, the answer has to sit inside one extractable block with the entity named, the claim stated and the reason attached.

A model needs a sentence it can attribute. Subject, verb, object, stated plainly and bolded, gets cited far more than the same fact spread over a paragraph. Google likes a direct answer for snippets and AI Overviews, but it will still rank a page that rambles if the links and topical coverage are there.

Google still maps a query to a page, and the title tag and H1 still carry real weight. AI retrieval works on meaning. The same question gets rephrased into several sub-queries before retrieval, so an exact-match keyword matters less than covering the concept from multiple angles. Keywords are not dead for AI; they are just one of many ways to be about the topic.

Models learn what a brand is from how often, and in what context, it appears across their training data and the sources they retrieve at answer time. An unlinked mention in a Reddit thread, a listicle or a podcast transcript teaches the model that your brand belongs in that category. Google gives an unlinked mention very little direct weight; it wants the link.

Internal links tell Google which pages matter and what they are about, and moving a page higher in the architecture is still one of the fastest wins I have. A model retrieving a passage is largely indifferent to where that page sits in your hierarchy. The modest AI weight comes from crawl discovery and from hub pages that prove topical coverage.

Reddit and community threads feature heavily in the sources assistants cite, and OpenAI and Google both have data agreements with Reddit. A brand that real people recommend in threads shows up in answers. Google surfaces those threads in its own results, but they rarely lift your rankings directly.

This is the AI-side replacement for PageRank. A model does not ask how strong the linking page is. It reads the sentence the link or mention sits in and learns what you are: "Kasra Dash, the Manchester SEO consultant who recovers sites from algorithm updates" teaches it something, a bare URL in a footer teaches it nothing. The same link on the same page can be worth a citation or worth zero depending on the words around it. Google reads anchor text and some surrounding context, but the link's authority does most of the work there.

Schema earns rich results in Google but is not a ranking factor in itself. For AI it scores close to nothing, because in every test I have seen, and in the ones run by people I trust, the JSON-LD simply does not get read when a model decides what to cite. Assistants work from the visible text of the passage. The entity facts you would put in schema still matter, but they have to be in the copy on the page and repeated across the web, not tucked into a script tag.

Defining the entity is a core part of SEO, and co-occurrence is how you do it. When "Kasra Dash" keeps appearing in the same sentences, pages and lists as "SEO consultant", "Manchester" and "algorithm update recovery", Google's entity understanding files you under that topic and ranks you for it more readily. For AI it matters, but it is not core, and the reason is how the systems work. An assistant runs a search, pulls chunks of content that match the query, and synthesises an answer from those chunks. Most of the time it is not checking the entity at all; it is checking whether the snippet answers the question. Co-occurrence still earns a real score for AI because it shapes which chunks come back for brand and category questions, but the passage itself does the heavy lifting.

Retrieval-backed assistants such as Perplexity and ChatGPT with browsing lean hard on recency, and a visible last-updated date is one of the first things they check when two sources conflict. Google cares about freshness for queries that deserve it and largely ignores it for evergreen topics.

Google applies E-E-A-T most heavily on health, finance and other high-stakes topics, and its raters look for evidence of first-hand experience. For AI it barely registers. Across the 19,000 answers analysed, anonymous pages were cited alongside named experts with no visible penalty, because a model rarely looks at the byline before deciding a chunk answers the question. Author signals earn a small score for the tie-break cases where two chunks say the same thing, and nothing more.

Google has confirmed page experience as a lightweight ranking signal and a tiebreaker, and in my tests fixing Core Web Vitals rarely moves a page more than a position or two on its own. A model fetching your page for a citation does not care whether it painted in one second or four. It cares whether the text was there when it arrived.

Google builds a Knowledge Graph entry for you and ranks you against it, so the same name, location, titles and dates everywhere is foundational SEO work. If your facts disagree across your site, LinkedIn, Crunchbase and profile pages, Google is slower to connect you to your topic. For AI the weight is lower than most people claim. A model retrieving chunks for a question is mostly judging the chunk, not auditing who you are. Consistency helps when the question is about you or your category, because the retrieved snippets then agree with each other, but it does not get a passage cited on its own.

Original numbers earn links, which is why they score well for SEO. For AI they score higher again, because a model answering a question wants a specific figure with a source attached, and your page is the only place that figure exists. Publish the number, the sample and the date next to each other.

The Google antitrust exhibits made clear that click behaviour feeds ranking systems, and I have watched CTR changes move positions on client sites for years. AI gets a real score here too, and it is higher than most people expect. Once a brand is cited, the clicks on that citation appear to feed back into which source gets cited first next time. I have seen brands move from the third cited source to the first on the same prompt with nothing changing except engagement on the citation. Some of that engagement is being manufactured: there is a grey-hat, arguably black-hat, practice of running bots against a brand's own citations the moment it appears in answers, to hold or improve the position. I am describing it because it tells you the signal is live, not because it is a strategy worth the risk.

The only factor scored 100 for both, because it is not a weighting, it is a gate. Nothing else on this chart matters if the page cannot be fetched. For Google, an unindexed page has a ranking of zero. For AI, the gate is wider than one index: ChatGPT leans on Bing, Perplexity runs its own crawler, and Google AI Mode uses the Google index. Block GPTBot or PerplexityBot in robots.txt and you have opted out of citations from those tools.