What we measure

The bet, stated plainly

        ORGANIC AI MINDSHARE
              │
              ├──▶  AGENT TRAFFIC     (tools fetching data on a user's behalf)
              │
              └──▶  HUMAN TRAFFIC     (AI assistants sending users to your site)
                            │
                            ▼
                    REVENUE · USAGE · DISCOURSE
                            │
                            ▼
                    fuel for the next training cycle

Four metrics.
One output, three inputs.

One output number tells you where you stand today. Three input metrics tell you what is driving it and what to change. Findings come from the deltas between metrics; no single number in isolation is the headline.

How it works

┌──────────────────┐    ┌──────────────────┐    ┌──────────────────┐
│  1. SCRAPE       │    │  2. QUERY        │    │  3. TRIANGULATE  │
│                  │    │                  │    │                  │
│  Reddit · X      │───▶│  108-cell        │───▶│  cutoff-split    │
│  GitHub · arXiv  │    │  battery         │    │  velocity +      │
│  HN · Trends · G2│    │  3 models, 3rds  │    │  framing split   │
└──────────────────┘    └──────────────────┘    └──────────────────┘
   foundation              agent layer             cross-channel

1. 1

   Scrape your category's foundation channels

   Reddit, X, GitHub, arXiv, Hacker News, Google Trends, G2: the instrument draws the subset your category actually surfaces on, not all of them at once. The corpus-feeding ones are the same buyer signal already in the AI training corpora.

2. 2

   Run install-decision queries through the frontier-model panel

   Claude Sonnet, Claude Haiku, and GPT-5.5 (three always-on models): 108 cells per report across four query framings, 3 rounds. The panel is vertical-matched, so developer subjects add a fourth model (GPT-5.3-Codex, 144 cells) and consumer subjects add a vertical-matched fourth model once validated.

3. 3

   Cross-channel triangulate and cutoff-split

   Foundation activity from after each model's training cutoff feeds the velocity metric that predicts which brands enter the next training cycle.

What a real finding
actually looks like

Open Claude or ChatGPT right now. Ask the question a new buyer in your category would ask, without naming your brand. Watch what surfaces. If you don't appear, or a competitor confidently shows up in your place, you're looking at the most common pattern Korpora surfaces.

The findings that move the needle are deltas between metrics. The shape we surface most often: a brand leads on the foundation channels where buying-decision conversations happen, and trails on the AI-assistant layer that indexes those channels on a 6 to 12 month lag.

OPERATOR DISCOURSE vs AI-ASSISTANT RECALL

                  FOUNDATION CHANNELS       AI-ASSISTANT LAYER
                  (operator discussion)     (buyer research)

Brand             HN 924   Reddit 1,375   Organic mindshare   7.4%
Competitor A      HN  21   Reddit    93   Organic mindshare  86.1%
Competitor B      HN 534   Reddit    33   Organic mindshare  80.6%

The brand leads on the channels where staffing-decision conversations
actually happen. AI assistants index those channels on a 6 to 12 month
lag, so the recall catch-up arrives one to two training cycles out,
provided the right indexable signal is in place when the next corpus
closes.

What lands in your inbox

Three artifacts from a single measurement. The first is the shareable thing. The second is the actionable thing. The third is the optional thing you ask for if a specific decision-maker on your team needs the same data shaped for their lens.

We hand you the fix, not just the score

Every finding ships with a ranked, paste-ready engineer fix for the exact gap we measured: the hypothesis, the channel, the anchor terms, and the spec to prove it moved. The score always comes with the move that changes it.

Most AEO tools hand a marketing team a dashboard and leave the work to them. We aim each fix at the specific surface or channel your measurement says you are losing, so a founder without a content team still knows exactly what to ship next.

Why build this