This page is how UnfoldWithAI evaluate AI tools—the shared methodology behind directory scores, reviews, and shortlists like writing and coding assistants. Use it when you run your own POC or when you read our verdicts.
It complements the buyer journey in Complete Guide to Choosing AI Tools.
Principles
- Evidence over demos: real tasks beat vendor slides.
- Comparability: same rubric across tools in a category.
- Honesty about limits: hallucinations, privacy, and lock-in called out.
- Freshness: scores show last-reviewed intent; major changes trigger updates.
- Disclosure: affiliate relationships never silently inflate scores.
Scoring criteria (0–10)
We score on a 0.0–10.0 scale and weight categories as follows (sums to 100%):
| Criterion | Weight | What we look for |
|---|---|---|
| Ease of use | 15% | Onboarding, clarity, learning curve |
| Features | 15% | Depth vs stated use cases |
| Performance | 15% | Speed, reliability, output quality |
| Value | 15% | Price vs capability for the target user |
| Support | 10% | Channels, docs community, responsiveness |
| Integrations | 10% | API, connectors, ecosystem |
| Documentation | 10% | Examples, clarity, maintenance |
| Business value | 10% | ROI potential, team fit, compliance posture |
Overall bands: 9.0+ exceptional · 8.0–8.9 excellent · 7.0–7.9 good · 5.0–6.9 mixed · below 5 not recommended. See tool pages such as ChatGPT, Claude, GitHub Copilot, and Notion AI for applied scores when published.
Evidence standards
A score needs notes: tasks run, date, account type (free/paid), and limitations. We prefer primary sources for pricing and security claims. Model behavior changes—treat any single session as a sample, not eternal truth.
Conflicts and disclosures
If a recommendation uses affiliate links, we disclose on the page. Affiliate status does not change the rubric; it changes how CTAs are labeled. Editorial overrides of weighted scores must be noted internally.
Update cadence
- Quarterly pass on actively scored tools.
- Within 14 days of major pricing, model, or ToS shifts that affect buyers.
- Immediate correction if we publish a factual error.
Automation and prompting quality affect outcomes as much as vendors—keep prompt engineering and the automation playbook in the loop.
How this shows up on UnfoldWithAI
Directory entries and future review posts should reference these weights. Shortlists such as writing and coding assistants cite the same bands so a “8.2” means the same thing across categories. When we change weights, we note it here and bump the “Last updated” signal on affected pages.
Buyers can print the criteria table into POC notes today; printable checklists will also land in Resources as packs ship.
FAQ
Why not use only public benchmark leaderboards?
Leaderboards rarely match your workflow, data, or review standards. We use them as signals, not verdicts.
Can vendors pay for a higher score?
No. Commercial relationships may fund the publication; they do not rewrite the rubric.
How should buyers reuse this methodology?
Copy the criteria table into your POC notes, keep weights stable within a category, and document tasks. That is enough to evaluate AI tools consistently inside your company.
Next steps
Apply the rubric while you choose AI tools, browse the directory, and read industry context under Business Solutions. Resources will host printable checklists as they ship. Subscribe for methodology updates.