How Citerecon Measures AI Search Intelligence
Large language models are non-deterministic, probabilistic retrieval engines. Citerecon rejects fake precision and single-shot snapshots. Here is the formal statistical framework we use to collect, extract, normalize, and attribute AI brand recommendations.
The Stochastic Observation Engine
REPEATED OBSERVATIONS & CONFIDENCE INTERVALS
When a buyer prompts an AI model, the model samples tokens stochastically from a probability distribution. Temperature settings, system prompt variants, RAG retrieval updates, and data center locations cause identical prompts to produce varying recommendations. Citerecon executes prompts across repeated observation batches (typically 20 to 100 iterations per prompt cluster) over 30-day sampling windows.
Every reported metric displays the exact sample size (“Based on N observations”). Low sample sizes (N < 15) carry explicit low-confidence markers.
We measure ordinal stability: how frequently your brand appears at rank #1 vs #2–3 across identical query variations and simulated browser sessions.
We test across OpenAI GPT-4o, Anthropic Claude 3.5 Sonnet, Perplexity Sonar, Google Gemini 2.0 Flash, and extended frontier models without cross-polluting datasets.
AI Market Share: 8-Part Decomposition
BEYOND SINGLE-SCORE VANITY METRICS
Single “visibility scores” hide the root cause of AI displacements. Citerecon breaks down AI Market Share into eight distinct, mathematically verifiable vectors:
Recommendation Share
Percentage of answers explicitly naming and recommending your product in the primary copy.
Citation Share
Share of linked footnotes and numerical citation references pointing to your domain.
Position Share
Ordinal rank weighted score (#1 position receives 3.5x weight vs #4+).
Competitor Share
Relative mention density compared directly to your designated 5–10 category rivals.
Source Authority
Algorithmic credibility weight of the 3rd-party domains citing you vs competitors.
AI Crawlability
Telemetry confirmation that AI bots (GPTBot, ClaudeBot, PerplexityBot) can index your core pages.
AI Referral Share
Verified inbound sessions originating from AI user agents and AI web search interfaces.
AI Revenue Share
Direct and assisted revenue pipeline attributed to verified AI session funnels.
Citation & Source Extraction Science
ZERO HALLUCINATED EVIDENCE
Unlike scrapers that guess citation origins, Citerecon captures raw model payloads, markdown reference maps, and grounding metadata. For engines supporting retrieval-augmented generation (RAG) like Perplexity and Gemini, we extract:
- Exact Source URLs: Fully qualified canonical URLs parsed from the model's grounding annotations.
- Query Fanout Traces: Recursive sub-queries executed by the engine during web retrieval (e.g. buyer prompt → comparative fanout query → source citation).
- Source Classification: Classification of each domain into Editorial, UGC/Community (Reddit, GitHub), Reference, Corporate, or Competitor Owned.
Attribution Boundaries & Integrity
NO UNFOUNDED REVENUE CLAIMS
We maintain absolute scientific honesty regarding traffic and conversion attribution:
Direct HTTP referrer from chatgpt.com, perplexity.ai, or claude.ai landing on your site with valid session parameters.
Users who first discovered your product via an AI citation, returned via direct or organic search within 14 days, and subsequently converted.
Statistical regression linking shifts in your Citation Share of Voice to overall inbound pipeline lift. Always explicitly labeled as modeled.
Known Scientific Limitations & Ethical Stance
Foundation model providers continuously update their internal system prompts, quantization kernels, and retrieval indices. Citerecon does not promise “guaranteed #1 AI placement” because genuine AI retrieval depends on broad web consensus and authoritative evidence. We equip software companies with empirical intelligence, automated schema generation, and source gap closure to win authentic consensus.