AIStackIntel is an AI adoption intelligence database. This page explains how we collect, score and present the data you see. It is important to us that the intelligence is evidence-based and auditable — not guesswork presented as fact.

1. What we track

For each company we record the AI-related technologies and adoption signals we can observe from public sources:

• LLM APIs & model providers — e.g. OpenAI API, Anthropic/Claude, Hugging Face
• Agent & RAG frameworks — e.g. LangChain, LlamaIndex, vector databases like Pinecone
• AI product features — chat widgets, copilots, AI features announced publicly
• Infrastructure — hosting, CDN, framework and cloud provider (context, not AI itself)
• Hiring signals — AI-related roles (AI Engineer, ML Engineer, RAG Engineer, LLM Engineer…) collected from careers pages and public ATS feeds (Greenhouse, Ashby, Lever).

2. Data sources (all public)

• Company websites and public pages
• Public documentation (docs subdomains, API references)
• Job postings and public ATS feeds
• Public GitHub repositories and engineering references
• Product launches and public announcements

We do not access private accounts, private repositories, or non-public information.

3. Signal collection

• Each domain is probed with multiple signal layers: HTTP/DNS headers, cookies, HTML/JavaScript framework markers, documentation subdomains, hiring signals and public references.
• Pages protected by bot defense (e.g. Cloudflare) may yield fewer signals — we report lower confidence instead of guessing.

3.5 Two scores: Visibility & Maturity

Every company gets two complementary metrics:

• Signal Visibility Score (0-100) — how clearly a company's public signals show AI adoption today. Only direct AI signals count: LLM APIs, frameworks, vector databases, model platforms, AI-native docs and AI hiring.
• AI Maturity Score (0-100) — a deeper read of adoption depth. The inferred badge appears only for AI-native companies (e.g. OpenAI, Anthropic) whose Signal Visibility Score is lower than their true maturity — the +30 baseline is an inference, not a detected signal.
Weights reward deep integrations (frameworks & vector DBs ×1.5), discount light widgets (×0.3), add a +30 baseline for AI-native companies (OpenAI, Anthropic, Cohere, Mistral, Stability — marked inferred on profiles), and add hiring density (AI roles ÷ total roles) plus AI-native docs.

Maturity thresholds: Traditional 0-19 · AI Emerging 20-39 · AI Adopting 40-64 · AI Forward 65-100.

4. Confidence scoring

• Every detection carries a confidence score (0–100%) based on signal strength and corroboration.
• A single mention on a careers page counts less than an API integration referenced in documentation plus a GitHub repository.
• We never claim 100% accuracy. Absence of evidence is not evidence of absence: a low score may simply mean "no public signals detected yet".

5. AI Adoption Score™ (0–100)

The Adoption Score combines evidence-based dimensions — direct AI adoption signals only:

• AI Technologies — LLM APIs, agent frameworks, vector databases, model platforms, AI-native products
• AI Talent — AI-related hiring signals
• AI Product Integration — AI features and widgets on the product surface
• AI-Native Infrastructure — AI-specific platforms and documentation systems

Generic infrastructure does not count. Cloudflare, Next.js, SvelteKit, Astro, Vite, WordPress, nginx, AWS and generic ATS systems are shown as context but never add adoption points. Generic chat software (Crisp, Drift, Intercom, Tawk.to) is not treated as AI adoption unless there is explicit AI-specific evidence.

Representative component weights (full scoring visible on each profile):

Dimension	Weight
AI Technologies	~50
AI Talent (hiring)	~35
AI Product Integration	~10
AI-Native Infrastructure	~5

Maturity thresholds (published, exact):

Maturity	Score
Traditional	0–19
AI Emerging	20–39
AI Adopting	40–64
AI Forward	65–100

Evidence Confidence (0–100%) measures the strength and corroboration of detections: number of distinct AI signals, corroboration from documentation, and corroboration from hiring data. Coverage reports which channels were scanned for each company (website, docs, careers, job feeds). A low-confidence detection is never presented as fact.

The score is designed to be conservative: we would rather under-claim than over-claim.

6. Freshness

• Refresh cadence is quality-gated while historical tracking and change detection are being validated.
• Every profile page shows its last updated date.

7. Corrections

See something wrong? We audit evidence on request. Email [email protected] with the company and the signal you believe is incorrect, and we will re-check and correct it.