Ranking methodology

Rankings can't be bought. Sponsorship and affiliate links never change scores or order.

The 100-point rubric

Part	Points	What counts
Hands-on test	40	The same standard test for every tool in a category (below). Notes and output files are kept as evidence.
Value for money	20	What you get per dollar, with the free tier counted. Pricing page URL and check date recorded.
User reviews	20	Aggregated from G2, Capterra, the app stores and Reddit. Every source is linked with its review count and date.
Ease of use	10	Time to first useful result, setup friction, clarity.
Trust	10	Privacy policy and data use, uptime/status page, support channels.

Review weighting

A review source counts only with at least 20 reviews. Each source's average rating is weighted by log₁₀(review count):

reviews score = 20 × Σ( log10(countᵢ) × ratingᵢ / 5 ) / Σ log10(countᵢ)

So 1,000 reviews weigh 3× as much as 10 would (if 10 counted), not 100×. Reddit threads are summarized as positive/negative counts and linked.

Standard tests by category

AI video

Prompt (fixed): "10-second product clip: a matte black water bottle rotating on a wooden table, soft morning light, slow push-in, text-free."
Generate once at the tool's default settings, then one retry allowed; record which one is kept.
Export at the best free resolution in 16:9, and in 9:16 / 1:1 / 4:5 if offered.
Score: prompt adherence, motion quality and artifacts, product consistency across frames, resolution, watermark, time to result.
Check commercial-use terms and watermark on the free tier.

AI characters & avatars

Create one character from the same reference description ("woman in her 30s, short red hair, green jacket").
Produce the same character in 3 poses: standing front, seated three-quarter, walking side view (or 3 talking-avatar shots).
Score: identity consistency across the 3 outputs, hands/face artifacts, control over pose, export quality, watermark.

AI images

Prompt A: a flat vector logo for "Harbor Coffee" with a wave mark, readable text.
Prompt B: a studio product photo of a ceramic mug on white, soft shadow.
Score: text rendering, prompt adherence, photorealism, edit/inpaint ability, resolution, license and watermark.

AI voice & music

Voice: a fixed 75-word script read as a 30-second voiceover in a neutral US English voice; export WAV/MP3.
Music: a 15-second upbeat jingle for "Harbor Coffee" from one prompt.
Score: naturalness, pronunciation errors, timing, audio artifacts, commercial license on free tier.

AI writing

Brief (fixed): a 600-word blog post "How small restaurants can cut no-shows" with 3 subheads and a CTA.
Then a 120-word follow-up email to past customers from the same brief.
Score: factual errors (each claim checked), structure, tone match, edits needed, plagiarism check, word-count accuracy.

AI coding & app builders

Spec (fixed): build a to-do web app with email/password sign-up and login, per-user to-dos (add, complete, delete), persisted in a database.
Run the result; test sign-up, login, logout, a second user cannot see the first user's items.
Score: works end to end, number of prompts/fixes needed, code quality, deploy path, time to working app.

AI agents & automation

Automate 5 steps: new web-form lead → enrich (company from email domain) → score → create CRM contact (free CRM sandbox) → send a Slack/email notification.
Run it with 3 test leads, including one malformed.
Score: steps completed unaided, error handling, setup time, transparency/logs, free-tier run limits.

AI for business (contractors, restaurants)

Contractor: from a fixed job description (re-roof 2,000 sq ft, tear-off, architectural shingles) produce an itemized estimate the customer can accept.
Restaurant: publish a menu from a fixed 12-item list and take a test reservation/order for 4 people at 7 PM.
Score: completeness, accuracy against the input, customer-facing quality, setup time, free-tier limits.

Rules

No invented scores or reviews. A tool missing any sub-score or its evidence is shown as Unscored in "Unranked: testing"; the site build refuses to rank it.
Order is by total score, ties broken by hands-on score, then name. Sponsor status is not an input; the build recomputes the order with sponsor flags removed and fails if anything differs.
Free tiers are tested first; we don't pay vendors for testing access and vendors don't pay for scores.
A "Robb tested" badge appears only where Robb used the tool himself, with his write-up.
Every category is re-scored monthly; each page shows its last-updated date.

Disclosure

Tools & Agents is run by AI Corp Inc. The site sells Sponsored listings and uses affiliate links (for example NordVPN, NordPass and GearUP); every such placement is labeled Sponsored. A sponsored tool can appear in a ranking only at the position its score earns, with the label shown. Links to vendor sites are a Tools & Agents Pro feature; scores are free.

Corrections

Spot an error or out-of-date price? Email [email protected]. We review corrections at each monthly re-score or sooner for factual errors.

Last updated Oct 19, 2026.