FoundByAI is a Generative Engine Optimization (GEO) platform that provides businesses with an AI Visibility Index (0–100). We measure and optimize discoverability across major AI assistants including ChatGPT, Perplexity, Gemini, Claude, and Copilot.
Our platform leverages proprietary live-inference testing and calibrated signal modeling to provide businesses with actionable, high-impact visibility roadmaps. The system is designed to function not merely as an SEO tool, but as a continuous trust engine for the AI knowledge graph era.
The audit engine operates on a deterministic, four-phase pipeline that ensures data consistency while utilizing LLM-based inference for real-world validation.
Before any processing, all input data (URLs, business names, city names) is canonicalized to ensure accurate entity matching and cross-referencing against external databases. Boolean fields are standardized and URLs are stripped to their canonical form.
The engine generates 12 highly specific, context-aware queries tailored to the business's audience type (Local, Blog, or SaaS). These queries span four categories: Discovery, Comparison, Brand, and Problem-Solving, ensuring comprehensive coverage of the intent spectrum.
The system executes these 12 queries via a live LLM API call. Responses are analyzed through a structured classification prompt that applies strict rules to categorize the business's presence as Visible, Partial, or Invisible — and also extracts sentiment (positive / neutral / negative) and cited source domains from the response. This provides empirical ground-truth data on how the model evaluates the brand. Claude is the only engine scored from a live test; the other four engines use calibrated estimation (see Section 7).
For engines that do not offer public, real-time search-visibility APIs (ChatGPT, Perplexity, Gemini, Copilot), we apply a Generative Engine Optimization (GEO) calibration model. This model assigns weighted point values to verifiable trust signals — including review counts, Google Business Profile completeness, schema markup, Wikipedia footprint, and directory citation density — mapped against each engine's publicly documented ranking behavior. Full calibration weights and accuracy methodology are detailed in Section 7.
The master visibility index is a weighted composite, optimized by business type (Local, Blog, SaaS). This ensures that a restaurant is judged on its Google Business Profile, while a SaaS platform is judged on its entity strength across professional directories.
Additionally, a disambiguation check (runDisambiguationCheck) evaluates whether the business name shares overlap with other known entities, applying a penalty to the master score when collision risk is High or Medium.
We do not rely on a single, fragile metric. The engine triangulates visibility through three independent layers: real-time LLM interaction (Claude API), heuristic calibration of other AI engines (GEO Model), and technical signal analysis (Schema / Technical SEO). This ensures scores are both grounded in live reality and scalable to every business category we support.
Every audit record is tagged with an estimation_method_version and an audit timestamp. This allows buyers and stakeholders to see that the audit engine is not static — it evolves as AI ranking patterns change, and every historical score carries provenance metadata.
The engine is architected with a built-in fallback mechanism. If a live LLM inference query fails, the system automatically defaults to the calibration model, ensuring the user always receives a report rather than an error state. Failed runs are tagged real_engine: 'fallback' for full transparency.
Our proprietary getPriorityActions engine translates every score deficit into a specific, time-bound remediation task. Each task includes a time estimate, impact weighting, and copy-paste template — directly converting audit data into a project management workflow for the user.
The audit pipeline runs as a stateless serverless function (runAudit), allowing parallel processing for high-volume audit loads without maintaining persistent infrastructure. Query execution is sequential to respect external API rate limits while minimizing total latency.
The query inference prompt is optimized to produce structured JSON classification in a single pass, eliminating the need for multi-turn conversations or expensive chain-of-thought reasoning. This keeps per-audit token costs predictable and low.
Ongoing subscription value is delivered through automated monthly re-audits. This provides recurring monitoring, Freshness Decay alerts, and historical performance tracking — all critical needs for professional-grade business owners — without requiring manual re-runs of expensive live tests.
A one-time audit report ($29) serves as a high-velocity customer acquisition tool, focused on immediate 'fix-it' value. Users receive their full action plan, schema templates, and priority remediation roadmap.
A subscription model ($49/mo) built on the reality that AI citation patterns are dynamic. Subscribers pay for continuous monitoring, Freshness Decay alerts, competitor benchmarking, and historical score trajectory. This ensures long-term LTV and profitability, as the product provides continuous value rather than a one-time snapshot.
The combination of a low-friction entry point and a high-retention subscription creates a healthy unit economics profile with strong gross margins, driven by the efficiency of the serverless audit pipeline.
FoundByAI's AI Visibility Index uses a hybrid data architecture: one live API-tested engine (Claude) and four calibrated estimation engines (ChatGPT, Perplexity, Gemini, Copilot). This section details exactly how data is sourced, how calibration weights are derived, and how accuracy is assured.
The Claude engine score is the only score derived from a real-time API call. During each audit, the system sends 12 business-specific queries directly to the LLM. Each response is classified in real-time using a structured prompt that returns: (a) visibility classification — Visible (business named and recommended), Partial (mentioned but not recommended), or Invisible (not mentioned at all); (b) sentiment classification — Positive, Neutral, or Negative; and (c) cited source domains — the websites the AI referenced in its answer. This live data is tagged with real_engine: 'claude' and estimation_method_version: 'v3' in every audit record for full provenance.
These four engines do not currently offer public, real-time search-visibility APIs that can be integrated for per-audit live testing at scale. Instead, FoundByAI applies a GEO calibration model — a weighted point system based on each engine's publicly documented ranking signals. The signals are drawn entirely from verifiable data the business provides during the intake form (GBP status, review count, schema markup, directory presence, etc.). No scores are fabricated; they are deterministic functions of the business's submitted digital footprint.
ChatGPT is weighted conservatively toward local trust signals and structured data. The calibration assigns: +20 pts for a verified Google Business Profile with at least 1 review; +15 pts for 10+ reviews; +10 additional pts for 50+ reviews; +10 pts for schema markup; +8 pts for an FAQ page; +15 pts for Wikipedia presence or press coverage; +7 pts if the business has been operating for 3+ years. Maximum achievable: 100 points (clamped).
Perplexity is weighted heavily toward live web citations and cross-platform entity consistency. The calibration assigns: +20 pts for press coverage or 2+ directory citations; +15 pts for 3+ consistent directory listings; +20 pts for Wikipedia presence; +15 pts for press coverage; +10 pts for an active content hub/blog; +10 pts for a LinkedIn company page. Maximum achievable: 100 points (clamped).
Gemini is weighted toward the Google ecosystem — Google Business Profile completeness is the dominant signal. The calibration assigns: +25 pts for a verified GBP; +10 pts for a GBP with an FAQ page; +5 pts for 1–9 reviews, +12 pts for 10–49 reviews, or +20 pts for 50+ reviews; +8 pts for an HTTPS-indexed website; +4 pts for structured schema data. Maximum achievable: 100 points (clamped).
Copilot is weighted toward professional, B2B, and Bing-indexed signals. The calibration assigns: +20 pts for a LinkedIn company page; +10 additional pts if LinkedIn exists and the business is SaaS or professional services; +10 pts for an HTTPS website (Bing indexing assumed); +15 pts for Crunchbase or Product Hunt presence; +15 pts for press coverage; +8 pts for other directory listings; +10 pts if the business operates in a professional or SaaS category. Maximum achievable: 100 points (clamped).
Calibration is the process of aligning estimated scores with real-world observed AI engine behavior. Because engines like ChatGPT and Perplexity do not expose a 'visibility score' via API, FoundByAI infers visibility from the publicly known trust signals each engine uses in its ranking pipeline. The point weights assigned to each signal (e.g., +25 for GBP on Gemini) are calibration values — they represent how strongly that signal correlates with actual AI recommendation behavior, based on industry research, published AI ranking model documentation, and competitive analysis.
The calibration model is implemented as a deterministic function (estimateOtherEngines) in the runAudit backend pipeline. Each engine starts from a conservative baseline score (10–20 pts). Points are added for each verified signal the business has, using engine-specific weights that reflect that engine's known ranking priorities. Every score is clamped to 0–100 and tagged with estimation_method_version: 'v3' so that the calibration methodology version is traceable on every audit record. The current model is static (hardcoded weights), which is standard for an early-stage platform and ensures full reproducibility — every business with the same input profile receives the same score.
The current calibration weights are static and versioned (v3). As the platform scales, the calibration will move toward dynamic recalibration — periodic re-weighting based on aggregate live Claude benchmark data and user-reported visibility outcomes. Each recalibration increments the estimation_method_version, and all historical audit records retain their original version tag for full provenance and auditability.
Accuracy is assured through a three-pillar strategy: (1) The Truth-Check Loop — because Claude is live-tested, its results serve as a ground-truth anchor. FoundByAI can periodically cross-reference estimated engine scores against live Claude benchmarks to detect calibration drift; if estimated scores diverge significantly from live results across a sample set, weights are adjusted. (2) Signal-to-Visibility Correlation — scores are not ranking predictions but signal health indices. The platform measures whether a business possesses the trust signals AI engines require, not whether it will rank #1 — a more defensible and durable claim. (3) Feedback-Driven Tuning — Growth subscribers receive monthly re-audits. When a business fixes an action item and their score rises, and they report real-world visibility improvement, that confirms the calibration weighting was correct; if not, the weight is flagged for review.
Every audit record stores the full provenance of its scores: which engine was live-tested (real_engine), the estimation method version (estimation_method_version: 'v3'), the data completeness percentage (dataCompleteness), whether a reality check was triggered (realityCheckTriggered), and any disambiguation penalty applied. This means any stakeholder — including an acquirer — can trace exactly how every number on a report was produced. No score is presented without a source label; users see clearly which engine is live-tested and which is estimated.
FoundByAI — Proprietary & Confidential. This document contains technical methodology for acquisition evaluation purposes only.