Why traditional ranking factors do not explain AI visibility
Traditional ranking factors explain position in a list, and a generative answer has no list. A ChatGPT or Perplexity response is a synthesised piece of text, so the useful question shifts from "where do I rank?" to "is my information part of what the system says, and how much of it?"
Uriel de Oliveira makes this argument in a conceptual paper in Information Research (2026). He contrasts classic search, where visibility comes "through navigability", with generative systems, where visibility "is therefore not primarily positional or extractive but representational: information becomes influential to the extent that it shapes the semantic content of generated outputs." From that comparison he derives three mechanisms of generative visibility: selection, contribution and consistency. They describe recurring mechanisms, in his words, "rather than prescriptive tactics", which is why they work well as a diagnostic frame.
The empirical work points the same way. Kargaev (2026) found a "near-null" correlation between traditional technical SEO factors such as HTTPS and page speed and GEO performance. The signals that did predict AI citation frequency were Brand Entity Mentions (NIS 0.918), Statistics (NIS 0.747) and Citations (NIS 0.671).
That does not make SEO irrelevant. Google states in its guidance on optimizing for generative AI features that AI Overviews and AI Mode run on its core ranking and quality systems, and that a page must be indexed and eligible for a snippet to appear at all. SEO gets a page into the candidate pool. The three mechanisms decide what happens once it is there. For the full signal comparison, see SEO vs GEO.
| Mechanism | Question it answers | Main drivers | How to measure it |
|---|---|---|---|
| Selection | Is the source used in the answer? | Entity clarity, semantic relevance, crawl access, structured content | Inclusion rate across repeated queries |
| Contribution | How much does it shape the answer? | Attributed evidence, clear concepts, precise positioning, topical depth | Influence score: semantic similarity, perturbation tests |
| Consistency | Does inclusion hold up? | Cross-platform entity data, multilingual content, coverage of all intents | Variation across engines, intents and paraphrased queries |
What selection means and what drives it
Selection is whether a source is incorporated into a generated response at all. De Oliveira calls it the replacement for ranking as "the primary visibility threshold": a source is either in the synthesised answer or excluded, and in information science terms selection "functions as algorithmic gatekeeping".
The closest SEO analogy is indexation, with one difference. Indexation is mostly a technical question. Selection is also a confidence question: does the system know clearly enough what this source is and what it covers to use it? The paper lists semantic relevance, conceptual clarity, credibility signals and contextual alignment as the factors that studies link to selection.
Four drivers matter in practice.
- Entity clarity. Kargaev ranks Brand Entity Mentions as the strongest GEO signal (NIS 0.918). A brand described in different ways on its own site, its Google Business Profile and third-party pages is harder for a model to pin down.
- Access and the organic foundation. Kargaev also documents an organic foundation effect: AI systems draw from the indexed, organically visible web. Crawler access is part of this. In our study of 310 Dutch SME websites, GPTBot had no or insufficient access to 20% of the sites we could test (60 of 296).
- Structured content completeness. Haddad (2026) measured +8.7% AI-assisted inclusion for structured content completeness. Pages that spell out the specifics (what, where, for whom, at what price) give a system something it can safely summarise.
- Machine-readable identity. In the same SME study, 52% of sites had no Organization or LocalBusiness schema and 89% had no machine-readable address. Google says no special markup is required for its AI features, so schema is not a switch. It does remove ambiguity about who you are.
Selection does not require being the biggest authority in a category. Specialist businesses can cross the threshold for their own narrow queries, because a precise and well-described source is easier to include than a vague, broad one. The entity side is covered in more depth in brand entity SEO.
What contribution means and why it is the value layer
Contribution is the degree to which a selected source shapes the meaning and framing of the generated answer. De Oliveira is blunt about the gap between the two: "Inclusion alone does not guarantee influence." He cites research showing that semantic contribution can diverge from citation frequency, and describes contribution as epistemic authority, "the capacity of information to structure interpretation within synthesised outputs."
A source can be one of twenty inputs and add half a sentence. Another source can supply the definition, the key numbers and the recommended next step. Both were selected. Only the second one shaped what the user took away, and that is where commercial value sits.
What raises contribution:
- Specific, attributed evidence. Iyappan (2026) reports that content with statistics and citations achieves 85% AI citation rates. A claim like "we are experts in AI visibility" gives a model nothing to reuse. A sourced figure with a named study does.
- Conceptual clarity. Drawing on knowledge organisation research (Bowker and Star, 1999; Hjørland, 2002), de Oliveira notes that "structured and conceptually coherent content remains more generatively legible." Consistent terminology and one idea per section make a passage easier to lift into an answer.
- Precise positioning. A clearly defined offer for a clearly defined buyer gives the system an obvious role to assign you in the answer. Luther and Touboul-Cohen (2026) describe this as a category positioning effect: precisely positioned brands achieve a disproportionately strong average position in AI responses, because specific positioning creates high-confidence semantic matches.
- Topical depth. Iyappan identifies topical authority as a Very Strong cross-paradigm signal. Deep coverage of one domain gives a model more material to draw on than thin coverage of many.
Iyappan's citation rates by content format show the same pattern: long-form contextual content 92%, entity-rich 89%, structured data 85%, FAQ-format 67% and keyword-focused 41%. Formats that explain and attribute outperform formats built around matching a query string. For the content side in detail, see AI content optimization.
What consistency means and how it is built
Consistency is the stability of selection and contribution across query phrasing, user intent, language and platform architecture. Without it, AI visibility is a set of lucky moments rather than a position. A brand can appear for one wording of a question and vanish for a synonym, or show up in ChatGPT and not in Google AI Overviews.
De Oliveira points to Chen et al. (2025) and Wang et al. (2024), who document "substantial variation across engines and across types of intent", and names linguistic robustness as an operational test: does a source keep its inclusion and influence "across paraphrased or translated queries"? For businesses working in several languages, Haddad (2026) adds a concrete figure: structured bilingual content produced a 9.4% attention gain in mixed-language sessions.
Luther and Touboul-Cohen (2026) show what consistency looks like in data. In their ten-week study of tea brands, Twinings held the best average position on ChatGPT at all five measurement intervals, the only case of sustained single-metric leadership in the study, while the visibility of other brands moved around. Beneath that surface volatility the competitive order stayed stable, with Kendall's W concordance values of 0.785 on ChatGPT and 0.743 on Google AI. A durable hierarchy exists, and consistency decides where a brand sits in it. What Twinings shows is not the highest peak but the most reliable floor.
Consistency rests on three things.
- Identical entity data everywhere. Name, address, category and description should match on your site, in your schema, in business profiles and in editorial mentions. Each AI platform draws on a different mix of sources.
- Coverage of every intent. De Oliveira separates informational, procedural, comparative and evaluative queries. A brand with strong how-to content and no comparison content will be selected for one type and missing from the other.
- Maintenance. Facts change: prices, services, addresses, team members. Content that is not kept current drifts out of step with other sources, and that inconsistency weakens selection over time. The volatile AI landscape that Luther and Touboul-Cohen (2026) measured means signals have to be monitored and refreshed, not built once.
Concentration makes this harder for smaller players. Aral, Li and Zuo (2026) found that AI search surfaces significantly fewer long-tail information sources than traditional search. When fewer sources get through, the ones already established keep getting picked. Tracking this across platforms is covered in AI search monitoring.
How the three mechanisms reinforce each other
The three mechanisms form a sequence and a loop. Selection is the prerequisite, contribution is where the value lies, and consistency turns both into a durable position.
De Oliveira describes this as the authority loop: "Information that is structurally coherent, semantically explicit, and institutionally recognized is more likely to be selected in generative outputs. Once incorporated, it gains visibility and perceived credibility. This enhanced legitimacy increases the likelihood of future inclusion, reinforcing representational alignment within model embeddings."
Two practical consequences follow. A programme that only asks "do we appear?" is optimising the minimum condition and ignoring the part that pays. And because the loop compounds, early and steady work counts for more than a one-off sprint. A brand that gets selected and shapes answers today is more likely to be selected tomorrow.
Kargaev's NIS hierarchy fits this frame. Entity mentions mostly drive selection, since they let a system name a brand with confidence. Statistics and citations mostly drive contribution, since they are the material a model reuses when it explains a topic. Haddad's structured completeness finding sits on the selection side, the bilingual finding on the consistency side. The Twinings result from Luther and Touboul-Cohen (2026) belongs to consistency: stable signals kept the brand at the top of ChatGPT answers across every measurement interval.
How to prioritise GEO work across the three mechanisms
The sensible order is selection first, then contribution, then consistency, while keeping in mind that much of the work overlaps.
| Priority | Mechanism | Typical work |
|---|---|---|
| 1 | Selection | Allow AI crawlers, fix indexation, Organization or LocalBusiness schema, consistent business data, complete service and product pages |
| 2 | Contribution | Add sourced figures and named studies, define concepts clearly, narrow positioning, build topical depth, earn editorial mentions |
| 3 | Consistency | Track ChatGPT, Perplexity and Google AI features separately, cover all query intents, maintain multilingual versions, refresh key pages |
Contribution work tends to help selection too. Content that is specific and well attributed gives a system more reason to include it in the first place. The reverse is weaker: schema and entity cleanup alone do not make a page shape answers, because shaping requires substance on the page. For a full action list, see the GEO checklist, and for planning across a year, the AI visibility strategy framework.
Frequently asked questions
What are the three GEO ranking factors?
The three GEO ranking factors are selection, contribution and consistency, as set out by de Oliveira (2026) in Information Research. Selection is whether a source is used in a generated answer. Contribution is how much that source shapes the answer's meaning. Consistency is whether both hold across phrasings, intents, platforms and time.
How do GEO ranking factors differ from SEO ranking factors?
SEO ranking factors decide position in a list, while GEO mechanisms decide inclusion and influence inside a synthesised answer. Kargaev (2026) found near-null correlation between technical SEO factors and GEO performance, with entity mentions, statistics and citations as the strongest predictors. SEO still matters as the entry ticket, because Google's AI features only draw on indexed, snippet-eligible pages.
Which GEO ranking factor matters most?
Selection matters first, because nothing else is possible without it. Once a brand appears in AI answers, contribution usually offers the biggest return: shaping the answer is worth more than being one of many background sources. Consistency then protects that position.
How do I measure selection, contribution and consistency?
Selection is measured as inclusion rate: how often your source appears across repeated, category-relevant queries. Contribution is harder; de Oliveira mentions semantic similarity analysis and perturbation testing as ways to estimate an influence score, and a practical proxy is reviewing what the answer actually says about you. Consistency is the variation in both measures across engines, intents and reworded queries.
Can a small business compete on GEO ranking factors?
Yes, especially on selection and contribution in a narrow niche. Selection depends on clarity and relevance more than on overall domain size, and contribution depends on the quality of evidence on the page. A specialist with precise, well-sourced content can outweigh a generalist on its own topic. The category positioning effect documented by Luther and Touboul-Cohen (2026) confirms this empirically: narrowly positioned brands earn a better average position in AI answers within their own territory.
Does structured data guarantee AI citation?
No, structured data does not guarantee citation. Google says no special markup is needed to appear in its AI features. Schema helps by making identity and facts unambiguous, which supports selection and consistency, but contribution still depends on the content itself.
How do GEO ranking factors relate to E-E-A-T?
E-E-A-T describes the qualities that make content trustworthy; the three mechanisms describe how those qualities turn into AI visibility. Experience and expertise mostly feed contribution. Authoritativeness supports selection and consistency, and trustworthiness runs through all three.
How long does it take to improve each mechanism?
There is no fixed timeline, because it depends on how often systems recrawl and refresh their sources. Selection fixes such as crawler access and entity data are usually the quickest to show up. Contribution and consistency build more slowly, as content accumulates and the authority loop starts to work in your favour.
If you want to know where your site stands on selection, contribution and consistency, start with a free AI visibility analysis.
