How We Rate AI Tools

One9Founders uses a uniform, transparent methodology to evaluate AI tools in our directory. Unlike other directories that rely on affiliate relationships or popularity metrics alone, we apply consistent criteria — and we only publish a numeric score when enough of that framework has been completed.

What runs on every tool is automated collection of published pages, then a score for each criterion that those pages actually support. Each score cites a source URL. Ease of Use, Reliability & Performance, and hands-on security testing are not automated: they need a person using the product.

Today, 357 of 12,565 tools have a Provisional score from published evidence. 0 tools have a full Rated review after hands-on testing. Listings that are not yet scored show “Not Yet Rated” rather than a placeholder number.

Our commitment: Zero affiliate bias. Every scored criterion cites a URL. We do not claim tests we did not run.

Rating and security states

Not Yet Rated

Fewer than 6 of 10 criteria have a citable source. No numeric score is shown anywhere — not on cards, comparison tables, FAQ copy, or structured data.

Provisional

Six or more criteria evidenced from published pages, and no hands-on testing yet. The number is an unweighted mean of the evidenced criteria, shown as e.g. “3.8/5 (Provisional — 7/10 criteria assessed)”. Automated scoring cannot reach 10/10, because Ease of Use and Reliability are never filled in by the crawler.

Rated

All 10 criteria scored, including hands-on testing by a person who has actually used the product. Only that short list qualifies. We show the numeric score with a tier: Outstanding (4.5–5.0), Excellent (4.0–4.49), Strong (3.5–3.99), Good (3.0–3.49), Fair (2.0–2.99), or Needs Improvement (below 2.0).

Security status

We check published security posture — encryption in transit, a reachable privacy policy, and stated compliance commitments such as SOC 2 or GDPR. We do not perform security testing. Until that criterion is scored: “Security: Not Yet Assessed”. Below 12/20: “Security: Flagged”. 12/20 or above: “Security: Published posture”.

The ten criteria

1. Security & Data Privacy

Automated

Published posture only. We do not test anyone's controls.

  • Encryption in transit (HTTPS on the live site)
  • A reachable privacy policy
  • Stated compliance commitments such as SOC 2, GDPR, or a DPA
  • Stated data retention or training-data commitments, when published

2. Functionality & Features

Automated
  • What the product or features page actually lists
  • Stated capabilities compared with the tool's own positioning
  • API or product surface mentioned on the site

3. Ease of Use

Hands-on

Not automated. Needs a person using the product.

  • Onboarding experience (time to first value)
  • Interface intuitiveness
  • Learning curve for new users
  • Documentation and tutorial quality in actual use

4. Pricing & Value

Automated
  • A reachable pricing or plans page
  • Free tier or trial, when published
  • Whether prices are listed rather than "contact sales" only

5. Reliability & Performance

Hands-on

Not automated. Needs a person using the product.

  • Uptime in actual use
  • Response speed and latency
  • Output quality consistency
  • Error handling and recovery

6. Integration Capabilities

Automated
  • An integrations, apps, or marketplace page
  • Named connectors published by the vendor
  • API or webhook mentions on the site

7. Customer Support

Automated
  • A support, contact, help, or docs URL that resolves
  • Stated support channels on that page

8. Company Stability

Automated
  • What the vendor publishes about the company, team, or funding
  • For open-source rows: whether the repository is archived

9. Update Frequency

Automated
  • A changelog, releases, or what's-new page
  • For open-source rows: last commit date

10. Startup-Friendliness

Automated
  • Free tier, student, or startup programme pages
  • Published credits or small-team pricing

What actually happens

1

Resolve the live site

We follow redirects and record whether the final URL is served over HTTPS. A dead or parked site is not scored.

2

Collect published evidence

We fetch a small set of pages from the tool's own domain — privacy, pricing, integrations, docs, changelog, and the homepage. Open-source rows use GitHub facts (licence, last commit, archived) instead. Each page is truncated; we do not crawl the whole site. We read the HTML we are served, so a client-rendered app may yield thin text — the URL still has to exist to be cited.

3

Score only what the pages support

A model reads those pages and scores a criterion only when it can cite one of the fetched URLs. No citation, no score. Guessing is discarded.

4

Leave the rest unassessed

Ease of Use, Reliability & Performance, and the hands-on half of Security stay null until a person uses the product. Absence is labelled, not filled in.

5

Hands-on testing (Rated only)

A listing becomes Rated only after someone on the team has actually used it and scored the two hands-on criteria. That list starts small on purpose.

6

Refresh on a schedule

The automated pass is re-run so evidence URLs and scores can move when a vendor publishes a privacy policy or a changelog. It is not a quarterly recertification of every listing.

Our Zero Affiliate Bias Commitment

One9Founders does not accept affiliate commissions from any tool listed in our directory. Our revenue comes from optional premium listings and enterprise partnerships - never from influencing which tools rank higher.

This means when we recommend a tool, it's because it genuinely scored well in our evaluation - not because we earn money when you click.

How We Evaluate LLMs

Our LLM Explorer tracks 250+ models across multiple dimensions. Unlike tool ratings, LLM data is sourced from public benchmarks and provider documentation.

What We Track

  • Arena Elo rankings (from Chatbot Arena)
  • Input & output pricing (USD and INR)
  • Context window size
  • Provider and model family
  • Open-source vs proprietary license
  • India-affordable pricing tags

Data Sources

  • Arena (arena.ai) text and WebDev leaderboards
  • Artificial Analysis Intelligence Index
  • OpenRouter and official provider list prices
  • Model cards, Hugging Face, and release notes

LLM data is updated as providers release new models or change pricing. Visit our LLM Explorer to compare all 250+ models.

How We Ingest Research Papers

The research hub tracks 8,300+ AI papers from 34,000+ authors. New work is ingested daily from arXiv and cross-referenced with HuggingFace daily papers.

What We Track

  • Title, abstract, authors, and arXiv categories
  • Publication date and PDF / arXiv links
  • HuggingFace upvotes and paper URLs
  • AI-generated summaries and topic tags
  • Beginner / intermediate / advanced difficulty
  • Linked code repositories when available

Data Sources

  • arXiv API for cs.AI, cs.CL, cs.LG, and cs.IR
  • HuggingFace daily papers for upvotes and trending
  • Author records built from each ingested paper
  • AI enrichment for summaries, tags, and difficulty

Papers are added as they appear on arXiv. Visit the research hub to browse all 8,300+ papers.

Questions About Our Methodology?

We believe in transparency. If you have questions about how we rate tools or want to report an error in our assessment, reach out to us.

Contact Us