Home › Methodology
Methodology
Subscriber count tells you who found an audience. It does not tell you who teaches well. Topranks ranks on evidence instead, using one of three methods per topic — and every topic page says which one.
How a list is ranked
- Research consensus — three independent research reviews and practitioner recommendations each rank their own picks; creators chosen by more of them, and ranked higher, come first. Every cited video is checked to be real and on the creator's own channel.
- Evidence tiers — used where public discussion is thin: practitioner recommendations first, then creators picked by several independent research reviews, each screened for teaching quality from transcripts where available. Labelled “limited evidence” on the page.
- Practitioner endorsement — who people doing the work recommend learning from. The strongest signal we have; used where there is enough public discussion.
- Transcript grade — how good the teaching is inside each creator's recent videos, scored from the transcripts.
- Editorial review — topic focus and authority, judged by hand. Where grades exist on these lists they are shown alongside the rank and can disagree with it; we show that rather than hide it.
Every list has at least 20 creators. Where strong evidence runs out before 20, the remaining places go to the next-best evidence and are labelled as such on the page. Exception for now: Turo & Car Rental (17), where fewer channels qualified; it will be rebuilt.
1. Discover
Each topic starts with a wide query matrix run against YouTube search in up to 17 languages. Some topics also draw on Google, Reddit, podcasts and “best X YouTubers” roundups, and expand through who creators collaborate with and cite.
2. Validate
Recent uploads are pulled from the YouTube Data API and checked against the topic. The bar is ~70% of the last ~40 uploads on topic. A creator can sit below it only if they are clearly well known for the topic and still post about it regularly — those entries are marked medium confidence and say so.
3. Verify
Topranks ranks independent practitioner-educators. Platforms, SaaS tools, agencies-as-brands and corporate channels go to a vendor appendix, never the ranked list. Channels must be active — except on endorsement-ranked lists, where a channel practitioners still recommend stays on, marked ARCHIVE with the date of its last upload.
4. Rank
Practitioner endorsement
A transcript can show whether someone teaches clearly. It cannot show whether they are among the best. People who do the work already know that. We read Reddit threads where creators ask who to learn from, plus threads that discuss each candidate by name. A model lists which creators each commenter recommends or warns against; every one of those is then re-checked one comment at a time, and only counts if the comment itself names the creator. A creator's score is the number of distinct people who recommend them, minus the number who warn against them. Every quote on a creator page links to its comment.
- Endorsed by 2 or more, with more recommendations than warnings, ranked by that net count
- Endorsed by one practitioner with no warnings, ranked next
- No community data yet: only used to reach 20, chosen on transcript grade with no warnings
Transcript grade (rubric v2)
Grading is done by a model. Eight models from six labs graded the same 57 transcripts and were compared against a panel of two frontier models from different labs. Claude Haiku 5.5 was chosen for agreeing most closely with the panel on non-English content and on topics other than Meta Ads. It was not the steadiest on re-runs — a creator's grade can move by a few points between runs — which is one reason grades are a quality check on endorsement-ranked lists rather than the ranking.
Each creator's five most recent solo, on-topic videos are read in full and scored 1–5 on four criteria:
- Specificity — real numbers, settings, named tools and worked examples
- Explains why — the mechanism behind the advice, not just the instruction
- Actionability — whether a viewer could act on it right after watching
- Signal vs. noise — teaching, as opposed to hype, story or self-promotion
The grade is the average, out of 100. Bands: S 85+, A 75–84, B 65–74, C 55–64, D 45–54, F below 45. Interviews, podcasts and off-topic videos are detected in the same pass and excluded; titles are not trusted, because many solo-looking titles turn out to be interviews. A creator needs at least three qualifying videos to be graded; anyone short of that is listed with the reason rather than scored low.
The original grade (Meta Ads)
Meta Ads was graded first, by hand and without a written rubric: the median score of up to ten recent confirmed solo videos, each sampled at several points through its length. Bands: S 68+, A 62–67, B 55–61, C 47–54, D 38–46, F below 38. These grades are not comparable with rubric v2 grades until Meta Ads is re-graded.
Coverage
| Topic | Ranked by | Listed | Languages | Graded | Researched |
|---|---|---|---|---|---|
| Meta Ads | editorial review | 25 | 7 | 43 | 19 Sep 2026 |
| Google Ads | editorial review | 25 | 7 | – | 14 Oct 2026 |
| Email Marketing | evidence tiers (limited evidence) | 25 | 1 | – | 23 Oct 2026 |
| SEO | editorial review | 25 | 9 | – | 11 Oct 2026 |
| Instagram Growth | editorial review | 25 | 8 | – | 11 Oct 2026 |
| YouTube Growth | research consensus | 25 | 1 | 4 | 23 Oct 2026 |
| Etsy | editorial review | 25 | 5 | – | 11 Oct 2026 |
| Dropshipping | editorial review | 24 | 6 | – | 14 Oct 2026 |
| Info Business | transcript grade | 25 | 6 | 25 | 11 Oct 2026 |
| Sales | transcript grade | 22 | 8 | 22 | 14 Oct 2026 |
| Airbnb & STR | editorial review | 25 | 6 | – | 10 Oct 2026 |
| Real Estate Investing | transcript grade | 20 | 3 | 20 | 14 Oct 2026 |
| Web Design | transcript grade | 20 | 7 | 20 | 11 Oct 2026 |
| UI/UX Design | editorial review | 25 | 7 | – | 14 Oct 2026 |
| CRO | editorial review | 20 | 6 | – | 14 Oct 2026 |
| AI Automation | transcript grade | 22 | 7 | 22 | 14 Oct 2026 |
| Turo & Car Rental | editorial review | 17 | 1 | – | 10 Oct 2026 |
What this does not measure
- Whether a creator's advice actually made anyone money. Nobody can measure that from the outside, and any site claiming to is guessing.
- Paid courses, communities or coaching. Only free public content is assessed.
- Creators without usable captions. On Meta Ads, 16 otherwise-qualified channels could not be graded because no transcript was retrievable. They are recorded as unresolved, not as low quality.
Honest limitations
- Transcript grading covers 7 of 17 topics (5 on rubric v2, Meta Ads on the original scale).
- Endorsement data comes mostly from English-language Reddit, so non-English creators are rarely mentioned. They reach endorsement-ranked lists mainly through the transcript-grade places.
- Endorsement counts are small: a handful of strong names per topic have real backing, and the rest are labelled by how much evidence they have.
- Meta Ads discovery stopped on a daily API quota limit, not on saturation — it was still finding new channels.
- Subscriber counts are live from the API and move. Everything else is as of the research date shown per topic.
- A high rank or grade means a creator is recommended or teaches clearly and stays on topic. It is not an endorsement of every claim they make.