Trust Score Methodology

How We Calculate Trust Scores

Each AI tool and agent with enough verifiable public data receives a Trust Score from 0 to 100, built from five weighted pillars: Compliance & Verified, Operational, Reliability, Market Proof and Company.

Our methodology combines market-adoption signals (named notable customers), company and funding signals, operational-reliability signals (measured uptime, status page, data residency, incident record), independent verification, and self-reported compliance signals to create a comprehensive trust profile. Third-party review-platform star ratings (Capterra, GetApp, and others) are never fed into the Trust Score and are not publicly displayed anywhere; only npm and PyPI package-download counts are shown.

The 5 Trust Pillars

Each pillar contributes a specific weight to the total score of 100 points. An item is scored only on the signals that apply to it — a missing signal is not counted against it, with three deliberate exceptions every item is expected to have: a named company, Intelloro verification, and measured uptime. Items with very little verifiable public data are marked "Insufficient data" (shown as "—") rather than given a misleadingly low score. The score is also capped by how much a vendor discloses publicly: an item with thin public data cannot score above 60 or 75, and one with too little shows "Insufficient data" instead.

Trust Tiers

Based on the total score, each tool is assigned a Trust Tier that gives you a quick visual indicator of overall trustworthiness.

Beyond Trust Scores, every tool and agent receives Dimension Scores and Task Scores, each rated 1–10. Tools are scored on 6 dimensions (4 universal + 2 AI-specific). Agents are scored on 7 dimensions (the same 6 plus 1 agent-specific dimension). Task Scores are drawn from per-category rubrics described below.

The Scoring Dimensions (6 for Tools, 7 for Agents)

Grouped into three tiers: Universal SaaS (4 dims, all products), AI-Specific (2 dims, all products), and Agent-Only(1 dim, agents only).

Task Scores rate how well a tool or agent performs at specific tasks within its category. Intelloro maintains a per-category set of scoring rubrics — one per vertical category. Each rubric defines 10 task capabilities that are the prime evaluation axes for that vertical (e.g., a Search & Discovery rubric scores search accuracy, semantic understanding, real-time indexing, filter refinement, source attribution,and 5 more). Each task is scored 1–10 against a 4-band anchor (1–3 / 4–6 / 7–8 / 9–10) — modeled after per-category feature-rating patterns used across the software-review industry.

Rubric selection is rule-based: an item is scored on the rubric of its category_id(e.g., data-analytics → the Data Analysis & BI rubric) plus one rubric for each task it does, so an item can have several rubrics. Every item scored on a rubric is graded on the same 10 canonical tasks for fair, apples-to-apples comparison. Rubric definitions are locked in code and protected against drift by automated audit checks.

Scoring follows one fixed procedure for every tool and agent. AI models (Google Gemini and Anthropic Claude) read the vendor's own website (pricing, features, documentation, security pages) together with the verification sources described above, and grade each task against the canonical 4-band rubric for the item's category. Dimension scores are mostly calculated by fixed formulas from the facts recorded for the item; only an agent's Planning Quality is judged by an AI model. Scores are not vendor-supplied. A person approves each listing before it goes live, and scores update whenever the underlying data changes.

The rubrics are the fixed comparison targets, but the free-text task labels our extraction pipeline produces (and admins enter) vary in wording. We normalize them through a controlled alias map — a curated set of aliases — so synonyms like “content writing” and “copywriting” resolve to one canonical rubric, keeping every item comparable. Aliases are grounded in recognized, independent standards rather than vendor marketing: W3C WCAG 2.2 for accessibility tasks, the NIST AI Risk Management Framework for governance, OpenTelemetry semantic conventions for observability, and the DAIR.AI taxonomy for prompt engineering — supported by public reference encyclopedias for general software categories. Every alias must map to an existing rubric: a continuous-integration drift gate rejects any that points to a non-existent category, keeping the map and the rubric set in lockstep, and new aliases require a cited source before they are merged.

The 60-rubric architecture aligns with the per-software-category feature-rating patterns used across the software-review industry. Forrester Wave and Gartner Magic Quadrant use analyst-determined criteria per evaluation, which is more nuanced but does not scale beyond ~50 vendors per cycle. Intelloro's LLM-judged fixed-rubric approach is optimized for breadth — every item scored on a rubric is graded on the same canonical task set, keeping comparisons consistent across the whole catalogue.

Every data point on Intelloro is tagged with its source so you know where the information comes from.

Confidence Levels

Each score and data point is assigned a confidence level indicating how reliably it has been verified.

Every listing carries a verification-level badge (Fully Verified, Largely Verified, Partially Verified, Limited Data, or Unverified) reflecting how many of its data points we have completed via our extraction + independent verification pass. The badge describes the state of our data about a tool, not a rating of the tool itself.

For vendors: if your listing carries a lower verification level, you can contact us to have it re-verified. The badge updates automatically after each extraction + verification pass. Industry pattern reference: LinkedIn Profile Strength (5 levels), Crunchbase Profile Strength, Wikipedia Content Assessment letter grades.

Verified listings on Intelloro are cross-checked against a checklist of up to 37 public sources, including public registries as well as the vendor's own pages. Not every listing has completed this process yet — each listing shows a verification-level badge (see above) indicating how much of its data has been cross-checked. This cross-checking goes beyond what a vendor's website alone says.

Every classification on Intelloro — task categories, supported countries, supported languages, data residency regions — runs through a controlled vocabulary and a multi-layer normalizer before reaching a tool or agent listing. The same rules apply to every entry on the platform: claimed, unclaimed, free, paid, sponsored or organic. Vendors cannot purchase a different vocabulary, override a normalizer rule, or skip the drift-detection cycle. The pipeline is summarized below.

The same controlled vocabulary, normalizers, and audit gates apply to every listing — there is no paid pathway to bypass them.

AI Origin Classification — How We Label AI-Native vs AI-Layer

For some tools and agents, Intelloro publishes an AI Origin label. This is Intelloro's editorial opinion — a transparent, subjective classification based on publicly available information, not a statement of fact and not a claim that any vendor misrepresents its technology. We apply one simple, disclosed test.

This label reflects Intelloro's opinion from public information and may be updated as products evolve. Vendors who believe a classification is inaccurate can request a review via our contact page.

Frequently Asked Questions