This page documents exactly how tools are researched, tested, scored, and written about. If you want to understand why a verdict is what it is - or verify that a finding is grounded in evidence - this is where to start.
No step is skipped. No review is published before the process is complete.
No single source is treated as definitive. Cross-verification across multiple independent platforms is required before a claim is stated as fact.
Every score is Arooj's assessment based on direct testing. Scores reflect the tool's performance for its intended primary use case - not overall impressiveness.
Scores are assigned on a 1 - 10 scale and reflect how well the tool does the specific job it is positioned to do. A grammar checker is scored on grammar checking accuracy, not on whether it has a good generative AI feature. A plagiarism checker is scored on detection reliability and source coverage, not on how nice the interface looks.
The score reflects the primary use case. Secondary features that work well can lift the score. Secondary features that are overstated in marketing - and don't hold up to testing - can lower it. The Grammarly score of 7.5 reflects strong core proofreading and genuine integration value, offset by AI detection claims that independent testing does not support at the level marketed.
Scores are updated when tools improve significantly, when pricing changes affect value-for-money, or when independent research changes the evidence base for a specific claim.
| Score range | What it means | Label |
|---|---|---|
| 9.0 - 10.0 | Best in category. Does what it promises at the claimed level. Limitations are minor and clearly disclosed. | Excellent |
| 7.5 - 8.9 | Strong for its primary use case. Some secondary features may be overstated. Worth the cost for the right user. | Recommended |
| 6.0 - 7.4 | Useful within a narrower scope than marketed. Meaningful limitations exist. Best for a specific user type. | Conditional |
| 4.0 - 5.9 | Works in limited scenarios. Significant gaps between marketing claims and real-world performance. | Limited |
| Below 4.0 | Does not reliably do what it claims. Not recommended at current pricing or feature level. | Not Recommended |
The testing criteria differ by category because the use cases differ. What matters for a grammar checker is not what matters for an AI video generator.
| Category | Core feature test | Pricing verified | User reviews consulted | Independent studies | Free tier tested |
|---|---|---|---|---|---|
| Writing & Editing | Error detection accuracy vs Word and Google Docs | ✓ | ✓ | Where available | ✓ |
| SEO & Content | Content scoring consistency, keyword data accuracy | ✓ | ✓ | ✓ | ✓ |
| AI Detection | False positive rate on human text, detection on edited AI | ✓ | ✓ | ✓ Required | ✓ |
| AI Assistants & LLMs | Output quality, reasoning, context retention, voice fidelity | ✓ | ✓ | ✓ | ✓ |
| AI Image Generation | Prompt fidelity, output consistency, style control | ✓ | ✓ | Where available | ✓ |
| AI Video Generation | Output quality, coherence, lip sync (where applicable) | ✓ | ✓ | Where available | ✓ |
| AI Audio & Podcast | Voice naturalness, editing accuracy, export quality | ✓ | ✓ | Where available | ✓ |
| Plagiarism Checkers | Detection rate on known content, false positive rate | ✓ | ✓ | ✓ Required | ✓ |
Not every feature can be tested on every plan. This policy governs how untestable features are handled - transparently, not silently.
Stated explicitly: "I was not able to test [Feature X] during this review because [specific reason]. Based on available documentation and third-party user reports, here is what is reported about it - but treat this as unverified from my direct experience." No fabrication. No filler description from a marketing page.
When independent studies conflict - as they do on AI detector accuracy, where results range from 22% to 99% depending on the study - both findings are presented with methodology context. The most favorable study is not chosen to match a preferred conclusion. The conflict is the finding.
Freshness signals are tracked. Claims about tool performance that are time-sensitive - AI model accuracy, pricing, feature availability - are flagged for future verification and dated in the review. Static facts don't expire; dynamic data is tracked.
Features that behave differently by region, account type, or plan tier are noted explicitly. A feature available in the US Enterprise plan that isn't available in the UK free tier is not described as universally available. Access conditions are part of the review.
If you have found a factual error, a pricing discrepancy, or a finding that newer evidence contradicts - the contact page is where to send it. Every correction is taken seriously and reviewed against the source.