Plain definitions of the terms that decide whether AI recommends a brand — and where each one sits in the JefurryAxis method.
Maintained by the JefurryAxis research team at Jeffery Asia. Updated 21 September 2026.
Gets a page found in a list of links.
Get a brand named inside the answer an AI writes.
That answer depends on who is asking and what they are going through — so we optimise for people and situations, not keywords.
The AI Visibility Score tracks it on fixed, brand-free questions. HCI grades content before it ships.
GEO is the practice of making a brand understood, cited and recommended in answers produced by generative AI models such as ChatGPT, Gemini, DeepSeek and Doubao.
In JefurryAxis: GEO starts from the buyer, not the keyword. We model who is asking and their situation, test the questions they actually type, and change the evidence the models read.
AEO is the practice of shaping content so that answer engines — AI assistants, AI search and in-app “ask” features — give a direct answer that includes the brand.
In JefurryAxis: we treat GEO and AEO as one job measured one way. AEO also covers closed platforms such as RedNote Ask, where only in-platform content counts.
AI SEO is the umbrella term for search work aimed at AI-generated results as well as ranked links. It usually combines classic SEO foundations with GEO and AEO.
In JefurryAxis: the SEO foundations still matter — crawlable pages, clear facts, structured data — but they decide whether AI can read you, not whether it chooses you.
SEO is the practice of improving a web page’s position in a search engine’s ranked list of results for a query.
In JefurryAxis: a page can rank first and still be missing from the AI answer to the same need. We measure the answer, not the ranking.
A zero-click search is one where the user gets what they need on the results or answer screen and does not click through to any website.
In JefurryAxis: when the answer is the destination, being named inside it is the visibility that counts.
An answer engine is any system that responds to a question with a composed answer rather than a list of links: AI chat assistants, AI search features and in-app AI answers.
In JefurryAxis: we measure ten models automatically and cover Tencent Yuanbao, RedNote Ask and Google AI Overviews through team-assisted runs.
B = f(P, E) states that behaviour (B) is a function of the person (P) and their environment (E). Kurt Lewin proposed it in 1951.
In JefurryAxis: B is the question a buyer asks an AI and the choice that follows. We model P and E, derive B, and check whether the brand is in the answer. See it turned by hand
A persona is a decision-maker defined by profile, goals, resources, constraints and preferences — a person with a motive and a timeline, not an age band.
In JefurryAxis: personas are derived from the brand’s own material, tagged across four segmentation dimensions and clustered, then confirmed with the client. P is the constant.
A scenario is the situation behind a question: its trigger, urgency, evaluation criteria and decision stage, plus outside signals such as policy, prices, sentiment and competitor moves.
In JefurryAxis: E moves with time, so it is written Et and watched by the Insights engine. How E is tracked
The P × E matrix crosses every persona with every situation it buys in. Each cell is one decision context with its own questions and its own winning answer.
In JefurryAxis: a standard slice is 5 personas × 2 situations × 3 questions = 30 questions per run.
A brand-free question is a test prompt that never names the brand being measured, written the way a buyer types before they know the answer.
In JefurryAxis: every measured question is brand-free. A question that names the brand measures the prompt, not the market.
Dual-track measurement scores international AI models and Chinese AI models separately, because they read different corpora and trust different sources.
In JefurryAxis: the two tracks are never averaged into one number. One brand scored 83.0 on one track and 32.6 on the other.
A closed platform is an AI answer surface that only draws on content published inside it, such as RedNote Ask or Tencent Yuanbao.
In JefurryAxis: websites and press releases cannot reach these answers; in-platform content is the only route in. The RedNote case
A query freeze fixes the personas, situations, questions, models and scoring before the first measurement, so later runs are comparable.
In JefurryAxis: a re-test uses exactly the same questions. New situations are added as new columns, never swapped in.
The AI Visibility Score is a 0–100 measure of how visible a brand is in AI answers to a fixed set of brand-free questions, built from mention rate, occurrence frequency, average rank and brand content ratio.
In JefurryAxis: the four measures are normalised, weighted and mapped so the score is most sensitive in the middle of the range. It is not market share. Formal definition
Mention rate is the share of sampled AI answers in which the brand is named at all.
In JefurryAxis: MR carries the largest weight in the score, because nothing makes up for being absent.
Occurrence frequency is how many times the brand is mentioned within an answer, adjusted for answer length.
In JefurryAxis: it separates a brand named in passing from a brand discussed throughout.
Average rank is the brand’s mean position in the lists AI answers produce, with an unlisted brand counted as last.
In JefurryAxis: counting absence as last stops a brand’s average from improving just because it disappeared.
Brand content ratio is the share of an answer’s sentences that are about the brand.
In JefurryAxis: it shows whether the brand is the subject of the answer or a line inside someone else’s.
A citation is a web page or post that an AI answer lists or relies on as a source.
In JefurryAxis: citation data decides where content is published — in one diagnosis, owned domains were cited zero times on the Chinese track.
Attack, defend and monitor are the three verdicts given to each persona after a run: attack where a rival leads, defend where the lead is shrinking, monitor everywhere else.
In JefurryAxis: verdicts are set per track and decide where the budget goes. Reading the radar
HCI is a content score that checks, before publication, whether a piece is written for the right persona, handles the right situation, offers a credible solution and is easy for AI to cite.
In JefurryAxis: four layers — persona fit, scenario fit, solution maturity, content quality — and twelve checks. HCI grades the input; VS measures the result. The four layers
These are the three stages content passes before an AI names a brand: it is retrieved for the situation, selected for the person, and recommended only if it carries brand-specific evidence.
In JefurryAxis: the model comes from our peer-reviewed paper accepted at ANZMAC 2026. The three gates
Source analysis identifies which domains and pages AI models actually cite for a set of questions, and why those pages were worth citing.
In JefurryAxis: cited pages are pulled, sorted by persona and scored, then turned into a content spec. How it works
Structured data (schema.org JSON-LD) describes a page’s facts in machine-readable form; llms.txt is a plain-text index of a site written for language models.
In JefurryAxis: both help AI read a site accurately. Neither makes AI recommend it — that depends on evidence fitting the person and the situation.
| SEO | AEO | GEO | The JefurryAxis approach | |
|---|---|---|---|---|
| Goal | Rank a page | Be the direct answer | Be cited and recommended | Be recommended to a named persona in a named situation |
| Where it shows | Ranked links | Answer boxes, assistants, in-app AI | AI-generated answers | Both tracks of AI models, plus closed platforms |
| What is optimised | Keywords, pages, links | Question-and-answer content | Entities, evidence, sources | Persona × situation, then evidence and sources |
| How it is measured | Position, clicks | Answer inclusion | Mentions, citations, share | AI Visibility Score on frozen brand-free questions, per persona and per track |
They overlap more than they compete: SEO makes a site readable, GEO and AEO make a brand the answer.
Nearly. GEO is usually used for generative AI answers and AEO for any answer engine, including in-app features. Both aim to get a brand into the answer rather than onto a list, and JefurryAxis measures them the same way.
No. SEO decides whether AI can read and trust your pages; GEO and AEO decide whether it chooses you for a particular person. A site needs both, and a strong SEO site can still be missing from AI answers.
There is no universal threshold. A score is only comparable with the same questions, models and competitors, which is why JefurryAxis reports the trend on a frozen question set and the gap to named rivals, per persona and per track.
No. They help AI read a site accurately. Recommendation depends on whether the content fits the person and the situation and carries evidence the model can cite.
Because real buyers do not name the brand before they know it. A question that includes the brand makes the model talk about the brand, so it measures the prompt instead of the market.
Knowing the terms is the first step. Seeing where AI places your brand today is the next — and it takes a single run.