How the numbers are built

You'll see a lot of numbers on this platform: pillar scores, an overall score, mindshare percentages, agent visits, human impressions, and a dollar range for "revenue at stake." Every one of them is a forecast, not a measurement — because AI answers change day to day. This page walks through where each number comes from and how to read it without needing the underlying formulas (which stay inside the product).

Reviewers and analysts ask the same question: where do these numbers come from? This page answers it at the level of concept and chain of reasoning — what feeds what, what each output is for, and how to read it. The underlying formulas, weights, prompt sets, baselines and per-category tuning are not published. They are what makes the output defensible and they stay inside the product.

Raw agent signals roll up into per-pillar scores. The pillar scores roll up into a single Appkitekt AI Index. From there, the Index and per-agent breakdown feed four downstream views: mindshare (your share of agent answers), agentic visits (traffic from agent surfaces), human impressions (the downstream funnel), and a revenue-at-stake band in USD. Every step is a separate, auditable stage — so when something moves, you can see where in the chain it moved.

Each pillar (AEO, GEO, AAIO, AXO, DAO, PULSE, RIVAL, PROTO, BUG, SIGNAL, MCP, SOCIAL) is a 0–100 readiness number for one slice of agent-readiness. Factor lists are public — you can see what we look at. How those factors combine into the score is tuned per category, per surface and per agent generation.

See The 12 pillars for what each pillar measures.

The Index is the single 0–100 roll-up across all pillars. Pillars don't matter equally everywhere — a retail app, a B2B SaaS site and an MCP server have very different agent funnels — so the roll-up shifts by surface and by category. The Index is a readiness number; it does not claim a revenue figure on its own.

See Scoring & weighting and Web vs App.

Mindshare is your share of category prompts where you appear in the agent's answer, weighted by where you appear (being named first counts more than being named tenth). The prompt set rotates on a regular cadence so the metric resists gaming. We project traffic from mindshare — not from the raw Index — because two brands with the same Index can have very different real-world visibility in the prompts users actually run.

Agentic visits combine three ideas: how visible you are in agent answers, how much demand there is for your category across agents, and how cleanly the handoff works once an agent recommends you. When Agent Intelligence connectors are live, this estimate is replaced by measured agent-referrer and AI-overview impressions, and the confidence label flips upward.

Per-agent reach answers "how many people see my brand through this agent, on a typical day?". It blends how many people use the agent, how often any prompt falls into your category, and how likely your surface is to be cited or invoked when it does. The output is a range with a documented source for each input.

The revenue band is the dollar version of the mindshare gap: how much agentic demand sits between you and the category leader, mapped through category conversion behaviour to a USD range. It defaults to category benchmarks and replaces them with your own numbers as soon as a connector (Stripe, store analytics, GSC) goes live. Every output is a low–high band, never a single number — and the band narrows mechanically as inputs sharpen.

Full reasoning: Traffic & revenue.

The market band is the total agentic opportunity inside your category. Your brand band is your slice of it, sized by how your mindshare compares to the category leader. Both are shown side-by-side so board-level conversations can separate "is the category big enough?" from "are we winning our share of it?".

• Coarse — audit only. Wide band, directional. Good for "is this worth fixing?".
• Partial — audit + at least one live connector. Band narrows meaningfully.
• Refined — all relevant connectors live with enough history. Band narrows again and is safe to share externally.

Confidence is shown on every revenue tile in the product. We never hide it — a wide band labelled honestly is more useful than a narrow band labelled confidently.

• Not a guarantee. The band is a planning aid.
• Not attribution. No tool can prove a specific sale came from a specific agent answer — yet.
• Not comparable across vendors. Different prompt sets and assumptions will not match ours.

• Scoring & weighting — pillar → Index roll-up.
• Traffic & revenue — how readiness becomes a USD band.
• Measuring AIAO impact — pillar-by-pillar impact mapping.
• Reading the numbers — bands, percentiles, confidence labels.
• Methodology — measurement frame and data sources.