Methodology

How we evaluate vendors

Best Data Annotation is an independent editorial research platform covering the AI data industry, based in New York City. It researches the companies that label, evaluate and build training data for AI models — publishing vendor research, comparisons and market data with public methodology, cited sources, and every commercial relationship disclosed.

This page is the contract between this platform and its readers: the criteria below are the only criteria behind our rankings, and this page changes only in public, with the date updated.

The evaluation criteria

1. Model fit. What the vendor actually sells — managed service, expert network, platform, or data lab — and whether its marketing matches that reality.
2. Evidence of delivery. Named customers, case studies, funding, filings or reporting that a third party produced. Self-published claims are labeled as such.
3. Pricing transparency. Whether a buyer can learn what the work costs before a sales process. Published pricing is scored as a positive; it is rare enough in this market to matter.
4. Workforce transparency. What is publicly known about who does the work and how they are vetted and paid.
5. Provenance practice. For data labs and expert data: whether artifacts ship with rubrics, reviewer notes and rights documentation.
6. Fit boundaries. Every profile must answer "who should not buy this" as specifically as "who should" — a vendor with no stated non-fit is a vendor whose profile is an advertisement.

The evidence standard

• Every factual claim about a vendor traces to a public source, cited on the page. Our source hierarchy: regulatory filings and the vendor's own published terms first; reputable reporting second; third-party review sites last, and labeled.
• Company-reported figures (crowd sizes, participant counts, valuations from company pages) are always attributed as company-reported.
• We do not claim to have tested, used or purchased any vendor's service unless a page explicitly describes a documented engagement — and at launch, no page does.
• Pricing we could not verify is written as "Not published". We never estimate a vendor's price and present it as their price.

We sell nothing we rank

Read page one of the search results for this category carefully: nearly every other “best data annotation” list is published by a vendor that ranks its own product first. This publication is not a vendor. We sell no annotation services, run no labeling workforce, and rank no house product — there is no house product. The vendors in our directory include our competitors for this page’s attention, not our own offerings. Where a commercial relationship does exist — a referral link — it is disclosed adjacent to the link itself, as the next section details.

How the platform makes money — in full

• Referral commissions. A small number of outbound links are referral links. Each is labeled with a disclosure adjacent to the link, stating that we may earn a commission at no cost to the reader. At launch there are exactly two such relationships, both with Micro1 (company registration and expert application), and they appear only where Micro1 is a genuinely fitting recommendation under the criteria above — including pages where Micro1 does not win the top pick.
• Vendor partnerships. Vendors may pay for enhanced listings (expanded profiles, verified-data reviews) through the partner program. Partnership buys presentation, never position: ranking inputs are the criteria above, applied identically to partners and non-partners, and partner status is disclosed on the vendor's profile.
• Nothing else. No sponsored rankings, no paid placements disguised as editorial, no sale of reader data.

Corrections

Vendor facts change. If a page is wrong, email [email protected] with the correction and a source. We fix verified errors, note material changes on the page, and update the date. Vendors receive the same corrections process as readers — accuracy is not a partnership perk.

Methodology version 1.0 · Last updated: November 8, 2026