The prompts that decide who AI recommends.
How do you know which prompts matter? We map them — Audience × Product → Topic → Prompt — into one funnel-sorted matrix, then measure where you stand across the five engines. Not guesswork. A method.
The Prompt Matrix is our method for finding the exact questions your buyers ask AI — mapping Audience × Product → Topic → Prompt into one funnel-sorted grid, so you optimise for the prompts that lead to revenue, not vanity keywords.
Every prompt traces back to a buyer.
APTP is the targeting system. You don't start from keywords — you start from who buys and what they buy, and derive the exact questions they ask an AI. Four steps, in order.
Audience
2–4 buyer segments — role, awareness stage (problem / solution / product-aware), the job they're doing, and the exact words they'd type. Your customers' language, not category jargon.
Product
Each offer and the concrete outcome it delivers, mapped to the segment it serves. This is the anchor every prompt has to earn its way back to.
Topic
Pair audience with product to derive topic clusters — one pillar per core theme, spokes underneath. Gaps in the cluster are your content opportunities.
Prompt
The exact buyer question — a full natural-language prompt someone asks ChatGPT or Perplexity, not a keyword fragment. Classified by funnel stage, each mapped to the page that should own the answer.
✦ The golden rule: every prompt must trace back through Topic → Product → Audience. A prompt with no path to an outcome is vanity — we cut it. Relevance beats volume; conversion beats vanity.
Three stages, one priority order.
Prompts are classified by where the buyer is — and worked in the order that pays. Money prompts first.
Worked first
"best … agency", "… vs …", "… cost", "… alternative". Comparison, case-study and pricing pages. The prompts closest to a decision — highest leverage, done first.
The bridges
"how to …", "… strategy", "measure …", "… framework". How-to, method pages and checklists that move a solution-aware buyer toward you.
Authority & AEO magnets
"what is …", "why …", "… explained". Guides, FAQ and pillar pages. Many carry near-zero Google volume but AI answers them anyway — we keep them, flagged AEO-only.
Your filled-in Prompt Matrix.
The non-negotiable output of every audit. One row per prompt — traced to a buyer, classified, volume-estimated, and mapped to an owning page. Example rows below use our own data; yours replace them.
| # | Audience | Product | Topic | Prompt | Stage | Vol (est.) | Owning page |
|---|---|---|---|---|---|---|---|
| 1 | A3 | P1 | Buying a GEO solution | "best GEO agency DACH" | BOFU | measured in audit | /geo-agency |
| 2 | A2 | P1 | Measuring AI visibility | "how to measure AI visibility" | MOFU | AEO-only | /ai-visibility-index |
| 3 | A1 | P1 | Understanding GEO | "what is generative engine optimization" | TOFU | measured in audit | /geo |
Example rows are generic and illustrative — no invented figures. Real volumes are estimated in the audit (DataForSEO, always labelled); prompts with no Google volume are kept and flagged AEO-only.
Then we measure where you stand today.
The matrix's top prompts become a citation test. We run each across the five engines and score it — a dated, screenshot-backed baseline, not a promise.
- 8–12 buyer prompts (BOFU-weighted) plus your three named competitors form the test set.
- Run across all five engines: ChatGPT, Gemini, Perplexity, Claude and Copilot.
- Every result is screenshotted and dated — the receipt, and your before/after baseline.
- Each prompt is marked: does the engine cite you, a competitor, or no one? That gap list is the work.
ARR rubric per engine: Cited-at-all 40 · Rank-position 25 · Share-of-voice 20 · Accuracy 10 · Sentiment 5 → averaged into a headline ARR.
Honesty guardrail: ARR is Agent Citara's own self-defined, measured methodology — never a universal external score. No guaranteed rankings, no "100/100". It's a tracked, dated trend.
The matrix is a work-queue, not a report.
Each un-cited prompt becomes a task, routed to the engine that can win it. Then the loop turns — and Studio OS runs it.
Target
Pick the prompts that matter — BOFU money prompts first, ordered by ARR-score-per-effort.
Act
Route each prompt to its C-lever — Site, Shop, SEO or GEO Engine — to build the asset, get found, or get cited.
Measure
Re-run the ARR baseline. Citations move; the delta is the proof, dated and screenshotted.
Re-target
The gaps feed back into the matrix. Studio OS spins the loop nightly; you approve in the morning.
Advised by an operator,
not a pitch.
Agent Citara was founded by Thorsten Keller — an economic journalist in Asia (Taiwan correspondent for the Süddeutsche Zeitung), then a C-level operator in global tech, today an AI architect and certified trainer in Neuroresonance, emTrace and Mimikresonanz.
Three perspectives that make AI visibility in commerce serious: a feel for the business, the engineering of AI systems, and the language that convinces people — and machines.
C-level in global tech
Chief of Staff Europe at Salesforce, EMEA COO at UiPath, plus Microsoft, Siemens and Atos. He knows how visibility turns into revenue — and orders.
Builds agent systems
Today he builds AI companies and agent systems — Agent Citara grew out of that. Agent E-commerce isn't a buzzword, it's his craft.
Language that lands
Certified trainer in Neuroresonance, NLP and emotional coaching. That's why your brand stays your brand — even for the machine.
Frequently asked questions.
How is this different from keyword research?
Keyword research starts from search fragments. The Prompt Matrix starts from the buyer — who they are and what they buy — and derives the full natural-language questions they ask an AI. Every prompt must trace back to an audience and a product, or it's cut. LLMs answer questions, not fragments.
What is the AI-Recommendable Rank (ARR)?
ARR is our self-defined, measured score for how often AI engines cite you when buyers ask. It's measured across ChatGPT, Gemini, Perplexity, Claude and Copilot using a fixed rubric (Cited 40 / Rank 25 / Share-of-voice 20 / Accuracy 10 / Sentiment 5), and presented as a dated, screenshot-backed trend — not a universal or guaranteed external score.
Do you promise a ranking?
No. AI outputs are non-deterministic. We influence the grounding data engines read; we do not control their answers. ARR tracks the trend honestly, with dated evidence — never a guaranteed "100/100".
What do I actually get?
Your filled-in APTP matrix (audience × product → topic → prompt, funnel-sorted, each mapped to an owning page), a measured ARR baseline with dated citation screenshots, and a prioritised work-queue routed to the engines that execute it.
See the system work on your case.
The free Check produces your first APTP matrix and measures your ARR — where you stand in ChatGPT, Perplexity and Gemini, against three competitors. From there, the machine takes over.