Top 5 Rankings & Hands-on Testing
What this skill is
A method for producing ranked tool comparisons that earn their order: explicit criteria, verified evidence per tool, a stated method, and an honest hands-on vs desk-researched label. The reader can see why each tool sits where it does — and what was never tested.
Who it is for: anyone writing "best X for Y" posts, tool roundups, or comparison carousels who wants the ranking to survive a skeptical reader checking the claims.
When to use
- Any ranked or comparison post.
- Any "best X for Y" claim anywhere in a post.
Rules (hard)
- Never create a Top 5 by copying another creator, repeating affiliate rankings, or picking familiar brand names without checking them.
- Never fabricate a test. If no test was run, the post is desk-researched and says so on the asset.
- If five credible products cannot be verified, publish a smaller shortlist ("3 tools worth testing") or retitle to "5 options to compare" — never invent a ranking or an entry.
- Sponsorship/affiliates never determine ranking. Disclose any relationship.
- Add a "last checked" date to every ranking, as a
[DATE]placeholder filled in on the day you publish (see Dates). Never ship a stale date from an earlier draft. - Compare on the same task and, where practical, the same prompt/input. Save test notes and example outputs when a test is run.
Weighted ranking criteria (the author's method)
These weights are the author's fixed method for this skill, not a researched fact — they are stated here so the reader can judge the ranking, and they do not change between rankings.
| Criterion | Weight |
|---|---|
| Task performance (does it do the stated task well) | 30% |
| Ease of use (learning curve, setup friction) | 15% |
| Value / pricing (free tier generosity, paid step-up) | 15% |
| Features / workflow fit (integrations, fit to the task) | 15% |
| Reliability (uptime, consistency, support) | 10% |
| Privacy / rights / trust (data handling, commercial-use terms) | 10% |
| Availability / support (regions, plans, docs, help) | 5% |
Score each tool 0–10 per criterion, multiply by weight, total /100. Publish the score table (or a readable summary of it) so the reader can see why each tool sits where it does.
Desk-research cap: where evidence is thin (desk research only), the task-performance score is capped at 7/10 and marked "desk estimate". The cap exists because a ranking that never measured performance cannot award full marks for it. Say so on the asset.
Claim ledger (mini-ledger, inline — no external dependency)
Before scoring, build a claim ledger: one row per factual claim you will publish. No claim reaches the post without a ledger row.
| # | Claim | Source (official page, help doc, or test note) | Date checked | Status |
|---|---|---|---|---|
| C1 | [e.g. "Tool X free plan gives 150 tokens/day"] | [official pricing page URL] | [DATE] | CONFIRMED / UNVERIFIED |
| C2 | … | … | … | … |
Status rules: CONFIRMED = read on the official page or measured in a
test you ran. UNVERIFIED = second-hand, outdated, or not yet checked —
do not publish it, or label it as unverified in the post. A template
ledger is in references/template.md.
Required research per tool
- Official product name + verified official URL (domain checked, no lookalikes).
- Best-fit task and intended user — what it is best at, not that it is universally best.
- Current free tier, trial, pricing, credit limits, regional availability — currency + date checked.
- Key features + relevant integrations (no filler features).
- Limitations, output quality, learning curve, privacy/data handling, usage restrictions, known gaps.
- Hands-on tested vs desk-researched — labelled.
- Why it earned its position + which audience member should choose it.
Hands-on test protocol (use when a test IS run)
- Fix one task and one input (same prompt, same source material) for every tool.
- Run each tool once, same day, save the raw output (screenshot/file) with timestamps.
- Score the outputs blind where possible against the 7 criteria; note failures, hallucinations, and time taken.
- Record test notes: date, tool version/model, exact prompt, input, output excerpt, score.
- Label the post "hands-on tested — [date]" and link or show the evidence. A test that was not run is never described as one.
Desk-research protocol (use when no test is run)
- Build the claim ledger first (above).
- Score from documented evidence only; mark task-performance scores as "desk estimate" and cap them at 7/10.
- State on the asset: "Desk-researched — last checked [DATE]. No hands-on test yet."
- Retitle honestly if the evidence cannot support a strict ranking: "5 options to compare" is acceptable; a fabricated #1–#5 is not.
Dates
Every ranking carries a "last checked" date as a [DATE] placeholder in
drafts. Fill it with the actual check date on the day you publish, and
re-check every price and free-tier claim on that day — pricing changes
often, and a ranking with a stale date misleads the reader.
Top 5 carousel card spec
Each card: consistent ranking badge (#1–#5) · tool name · "best for" line · key benefit · limitation · access-link cue. Slide 1 = hook + task + "last checked" date. Final slide = how to choose (which tool for which reader) + one CTA. 7–10 slides, 1080×1350, slide 2 = second hook.
Template
Use references/template.md for every Top 5 build: claim ledger,
per-tool research sheet, score table, and card spec checklist.
Worked example — 5 AI image generators for social post visuals
TEMPLATE — replace the topic, tools, and scores with your own. The tools and prices below are public data used to show the format, not content to copy.
Task (fixed for all five): generate a 1080×1350 social post visual with a readable headline. Method: desk-researched only — no hands-on test was run. Last checked: 16 November 2026. Prices below were verified against the vendors' pricing pages via search on that date; re-check them before reuse, because pricing changes.
| Tool (official site) | Entry point verified 16 Nov 2026 |
|---|---|
| Midjourney (midjourney.com) | No free plan. Basic $10/month. |
| Adobe Firefly (adobe.com) | Free plan with limited generations. Standard $9.99/month (2,000 credits). |
| Ideogram (ideogram.ai) | Free plan (weekly credits; free images public). Plus $20/month, or $15/month billed annually. |
| Leonardo AI (leonardo.ai) | Free plan, 150 tokens/day. Essential $12/month. |
| ChatGPT — GPT Image (chatgpt.com) | Free with limits. Plus $20/month. |
Score table (0–10 per criterion; task-performance scores are desk estimates, capped at 7/10 per the desk-research rule):
| Tool | Task perf. (30%) | Ease (15%) | Value (15%) | Features (15%) | Reliability (10%) | Privacy (10%) | Availability (5%) | Total /100 |
|---|---|---|---|---|---|---|---|---|
| #1 Adobe Firefly | 6 | 8 | 8 | 7 | 8 | 9 | 8 | 74 |
| #2 ChatGPT (GPT Image) | 7 | 9 | 6 | 7 | 8 | 6 | 8 | 72 |
| #3 Leonardo AI | 6 | 7 | 8 | 8 | 7 | 6 | 7 | 69 |
| #4 Ideogram | 7 | 7 | 7 | 6 | 7 | 6 | 7 | 68 |
| #5 Midjourney | 7 | 6 | 5 | 7 | 8 | 6 | 6 | 65 |
Worked total, Firefly: 6×0.30 + 8×0.15 + 8×0.15 + 7×0.15 + 8×0.10 + 9×0.10 + 8×0.05 = 1.80 + 1.20 + 1.20 + 1.05 + 0.80 + 0.90 + 0.40 = 7.35 → 74/100.
Best-for lines (example):
- #1 Adobe Firefly — best for commercial-safe brand work (licensed training data; strongest free-to-paid step-up in this set).
- #2 ChatGPT — best for text-in-image and chat-based edits.
- #3 Leonardo AI — best for high-volume creation and consistent characters (most generous free daily allowance here).
- #4 Ideogram — best for posters and images with accurate text.
- #5 Midjourney — best for artistic, polished-looking images; the only tool here with no free plan.
What the asset would say: "Desk-researched — last checked 16 November 2026. No hands-on test yet." Scores are illustrative example scores for this worked example, not measured test results — they show the method, and a real ranking must re-verify every price and re-score from its own evidence.
Troubleshooting
| Problem | Fix |
|---|---|
| Only 3–4 credible tools verify | Publish a smaller shortlist or retitle to "options to compare". Never pad the list. |
| A price changed since the last check | Re-check the official pricing page, update the ledger row date, and re-score Value before publishing. |
| Two tools tie | Break the tie on the highest-weighted criterion (task performance), and say so. If still tied, list them as tied — do not invent a gap. |
| A tool's free tier disappeared | Re-score Value and Availability from the new facts; if the tool no longer fits the task, drop it and say why. |
| Evidence is thin for one criterion | Score conservatively, mark it "desk estimate", and note the gap on the asset. |
Known limits
- The weights are the author's method, not a researched fact. A different weighting would produce a different order — that is why the score table is published with the ranking.
- Desk-researched scores are estimates, not measurements. The 7/10 task-performance cap exists for exactly this reason.
- Prices and free-tier limits change frequently. A ranking is only as current as its "last checked" date.