Updated October 2026
How we rank
Three readers, three rubrics, five criteria each, with published weights. Every sub-score cites the facts behind it, every fact cites its source and the date we checked it, and a commission is never an input. Scoring version 0.1, last run 14 Oct 2026.
The short version
- We collect facts from primary sources (company pages, terms, filings) and reputable reporting. Each fact records its source URL, the publication date when known, the date we retrieved it, and whether it is the company's own claim.
- Analysts score each rubric criterion from 0 to 10 against the written anchors below, citing at least one fact. A criterion with no evidence is left unscored, never guessed.
- The overall score is the weighted mean of the scored criteria, renormalised to the weight they cover, to one decimal.
- A score is published only when criteria worth at least 35% of the weight are evidenced. Below that, the page says "Not enough public data" and shows how many criteria we could evidence.
- Companies are ranked by score; unscored companies follow, ordered by how much of the rubric we can evidence, then by name. The order is the same for every visitor.
Today the site tracks 16 companies and 827 sourced facts, of which 508 are the companies' own claims and are labelled that way wherever they appear.
Buyers of company data
Offer size and evidence30%
- 9-10: a published offer range reaching $1M+, backed by independent reporting of real deals or budget
- 7-8: a published offer range of $100K or more, stated by the company
- 5-6: a range under $100K, or a revenue share with no figures
- Not scored: no offer range published
Privacy and de-identification25%
- 8-9: an independent party removes personal data before transfer, and client-held data is excluded
- 7: a documented process, such as a SOC 2 Type II attestation
- 5-6: personal data removal described loosely
- Not scored: nothing published
Contract clarity20%
- 8: the seller keeps ownership, reviews samples, and the structure is a license
- 6-7: some terms published, such as fees or structure
- Not scored: no terms published
Speed to payment15%
- 9: payment within 30 days of signing, stated in writing
- 7: an offer within two weeks and payment on Net 30-60 terms
- Not scored: no timeline published
Track record10%
- 8-9: independent reporting of completed deals or a committed budget
- 6-7: independent reporting of bids or a smaller deal record
- Lower: reported incidents
- Not scored: no independent record
Human-data vendors for AI teams
Expert depth and vetting25%
- 9-10: documented expert pools in named fields with published vetting
- 7-8: documented expert pools, vetting described
- Not scored: no public evidence
Quality systems20%
- 9-10: published gold sets, agreement rates or third-party audits
- 7-8: quality process described in detail
- Not scored: no public evidence
Security and neutrality20%
- 8: SOC 2 Type II documented and no reported incidents
- 6: SOC 2 documented, but an incident reported in the last 12 months
- 5: part-owned by a company that competes with its customers
- Not scored: no public evidence
Scale and speed20%
- 9.5: reported run rate of $2B or more
- 9: $1B or more
- 8: $500M or more
- 7: $250M or more
- 6: under $250M
- Not scored: no reported figure
Transparency15%
- 8-9: public terms, pricing or rate cards
- 6-7: public program terms, no pricing
- Not scored: little published
AI training platforms for experts
Pay level and evidence30%
- Scored from the median pay in public job listings, once our pay index is published
- Not scored: pay index not yet published
Work availability20%
- Scored from listing volume over 90 days
- Not scored: not yet measured
Payment reliability20%
- Scored from published payment terms and reported late payments
- Not scored: not yet measured
Onboarding fairness15%
- Scored from whether screening is paid, how long it takes, and retake rules
- Not scored: not yet measured
Reputation and incidents15%
- Lower for reported breaches, lawsuits or pay cuts
- Not scored: not yet measured
Labels
Labels are earned by rule, only by companies with a published score. Nobody can buy one. Labels that need data we do not have yet stay switched off.
- Highest offers (sellers): Highest offer sub-score among ranked buyers.
- No fees (sellers): States in writing that sellers pay no fee.
- Fastest payment (sellers): Highest speed-to-payment sub-score.
- Strongest privacy controls (sellers): Highest privacy sub-score.
- SOC 2 (buyers): SOC 2 Type II stated publicly where we could check it.
- Lab-neutral (buyers): Documented as having no AI lab among its owners.
- Deepest expert pool (buyers): Largest documented expert pool. Waiting for comparable data. Not awarded yet.
- Top pay (experts): Highest median pay in our pay index. Waiting for the index. Not awarded yet.
- Most work available (experts): Most open roles over 90 days. Waiting for data. Not awarded yet.
- Pays on time (experts): Needs enough first-party reviews. Waiting for reviews. Not awarded yet.
Verified and warning status
Warning marks a confirmed breach, a lawsuit, reported pay cuts or an ownership conflict, always with the reason and the source on the profile. Verified means we have checked the business registration, real customers and the terms ourselves. No company is marked verified yet, because we have not finished those checks for any of them.
Evidence rules
- Primary sources first: company pages, terms, filings and press releases, then reputable reporting.
- When sources disagree we show both figures and say so, for example "Surge 2024 revenue reported as $1B to $1.2B".
- Gross and net figures are labelled. Estimates (such as Sacra's) are labelled as estimates.
- Claims we could not source, or that rest on one weak source, are left off the site until they can be confirmed.
- Negative facts (breaches, lawsuits, ownership conflicts) are shown on the profiles they belong to and never removed for a commercial reason.
Reviews
Reader reviews are shown beside our score and never blended into it or into the ranking. Only reviews we collect and publish ourselves count toward an average or appear in structured data. Every review is moderated, negative reviews are never hidden, companies may reply but cannot remove a review, and reviewers must disclose any connection to the company.
Updates and corrections
Every fact and score is rechecked at least monthly, and immediately when a company changes its terms or a material event is reported. Each profile carries a change log. To report an error, use "Report an inaccuracy" on any profile or email info@trainerlist.ai. Fixes are dated and logged.
Reusing our data
You may quote and reuse our rankings and facts with a link back to the page you used. The full ranking with every source is at /data/companies.json, and a plain-text summary for AI systems is at /llms.txt.
Who scores
Scores are set by the editorial team. The named lead analyst and reviewing editor are still to be announced, and their names will appear on every profile when they are. See About.
Version history
- v0.1, 14 Oct 2026. First published. Three rubrics, 35% evidence threshold, labels by rule.
FAQ
About the method
Each audience has its own published rubric of five weighted criteria, and every sub-score cites the facts behind it. A score appears only when we can evidence at least 35% of the rubric's weight; otherwise the card says there is not enough public data. Commissions never enter the calculation.
No. No company can pay to be listed, to move up, or to change a score. Some companies run referral programs that may pay us when a reader we send signs a deal, and we disclose each one by name. The ranking is calculated from sourced facts before any commission is considered, and commissions are never an input.