Is it written for this person?
- PFSPersona match
- TSMTone & style
Optimization Engine
Turn what you learn from AI answers into better content. Fill information gaps, explain what makes your product useful, and support important claims with evidence people can check.
Part of JefurryAxis, the AI SEO, GEO and AEO product by Jeffery Asia.
Answer the decision.
Back it up with facts.
Detection tells you which squares of the grid you lose. This engine fills them. Every brief starts from the same two questions: who is asking, and what is going on in their week — B = f(P, E) applied to a page rather than a paper.
Then one more question, the one most content skips: what would this person need to read before they felt safe choosing you?
From a broad claim to something quotable
Nothing to quote
“The best programme for ambitious professionals.”
Something an AI can cite
Class times. Total fees. Entry requirements. Where last year’s graduates went. Dated, sourced, and specific enough to be wrong.
A model quotes what it can attribute. Adjectives are unquotable — that is the whole reason they fail.
Written for a persona.
Not for a prompt.
Most GEO work picks a few intents or prompts and writes until the brand shows up under them. We write for the buyer behind the prompt, so each piece reaches a whole group of people you want, in whatever words they ask. It is traditional GEO with an aim: less budget spent on readers who were never going to buy.
Prompt-first content has a cost of its own. Guessing the wording yields thin, near-identical pages, the wins fade when the phrasing moves, and mass-produced prompt bait pollutes the sources every AI reads.
The method is written up in Persona-Driven AI Engine Optimization in Zero-Click Search, accepted for an oral presentation at ANZMAC 2026 after peer review.
Two ways to chase the same answer
Prompt-first GEO
Pick twenty prompts. Write a page per prompt and repeat the phrase until a model picks it up. Visibility with no idea who it reached.
Persona-first, with JefurryAxis
Pick the buyer and their situation. Write the evidence they need before choosing you. Cited whenever that kind of person asks.
SMU case: the AI Visibility Score rose from 69.1 to 77.2 in seven days, and the articles were cited only under the target persona’s questions. Read the case.
Content gets a score too
VS tells you where you stand. It cannot tell you what to write. So before anything ships, we grade it with HCI — Human-Centric Insight: the right person, the right moment, a solution worth trusting, written so AI can quote it.
Does it understand the person and the moment?
Is it worth an AI quoting?
Does it handle the decision they face right now?
- SCScenario coverage
- SRSAnswers the core question
- SEAEmotional fit
- MADDecision stage
Is the answer credible, honest about limits, and fair?
- PPCPain points solved
- RAMSRisks flagged
- SOWho it fits, and who it doesn’t
Can AI lift it cleanly and trust it?
- SRClear structure
- CVDInformation density
- ASCAuthoritative sources
Four layers, twelve checks. HCI grades the input; VS measures the result. In our tracked work they move together — a working relationship, not a law.
Three platforms. Three rulebooks.
One GEO package for all of them is the most expensive mistake in this category.
Open web · International
ChatGPT · Claude · Gemini · Perplexity
Reads official sites, established media, rankings.
Your siteSchemaTrade pressRankingsReddit / LinkedIn
Open web · Chinese
Doubao · DeepSeek · Qwen · Kimi
A different corpus, with its own trusted platforms.
Content platformsLong-form Q&ATrade mediaYour site
Closed platforms
RedNote Ask · Yuanbao
In-platform posts only. Nothing else gets in.
Creator contentUser content
We don’t guess what AI likes.
We read what it cited.
Two questions, in that order. Where does the model go for content — and why was that page worth quoting?
Citation counts from one client diagnosis, same week, same thirty questions. On one track the owned domain is the battleground and it is losing to three rivals. On the other it is not in the contest at all. Two citation economies that share nothing — so the plan gets written twice.
Then we take the cited pages apart
- 01Pull the page
Read what was actually cited, before looking at whose domain it sits on.
- 02Sort by buyer
One domain often serves three different people. Only one of them is yours.
- 03Score why
Genre, structure, density of checkable numbers, authority of the source.
- 04Write the spec
A brief another writer could execute — not a note saying “write more”.
This is not copying. What gets reproduced is the structure of why something was worth quoting: the genre, the evidence density, the way a fit boundary is written down.
How a piece gets made
- Inputs
Market signals, the source landscape, brand rules, real prospect questions.
- Draft
Combined into one draft, on-brand from the start.
- Fact check
Every number sourced. Fails, goes back.
- Human sign-off
A person decides. Not a model.
- Re-run
Same questions. Did it move?
What accumulates is not the articles — it is the record of what got cited. That becomes the next input, and it is why month ten beats month one.
Two cases, two rulebooks
One on open-web Chinese models. One inside a closed platform, where nothing published elsewhere could have helped.
Finance master’s: 69.1 → 77.2 ↗ Sentosa: absent → cited inside RedNote ↗What makes a result worth reading
A named audience
Which profiles, which situations — not “consumers”.
An unchanged basis
Same questions, models and scoring, before and after.
An honest limit
Visibility is visibility. Not enrolments, not revenue.
Questions about this engine
What does HCI stand for?
Human-Centric Insight. It grades a piece of content before it ships, across four layers — persona fit, scenario fit, solution maturity and content quality — twelve checks in all. It tells us what to fix first; whether the content actually gets cited is checked in the next run.
Can you control what an AI model recommends?
No. We can improve the information available to people and search systems, then test for changes. The AI platform controls its retrieval and answer generation.
Will you automatically post everywhere?
Publishing channels and approvals are agreed for the engagement. A website update, editorial PR placement and social post are different activities. No placement, platform endorsement or automatic distribution to every channel is implied.
Does adding schema or llms.txt guarantee results?
No. Structured data helps describe visible content, and llms.txt can provide a reading guide. Neither guarantees indexing, citation or a recommendation. Clear content and accessible pages remain the foundation.
Be the AI
Recommendation
AI tends to quote the brand with the clearest, most specific answer — not always the biggest one. With the right content in the right places, that brand can be yours.