Ask for a talent map, not a macro index
Metix AI maps a company’s or a field’s talent structure from public professional profiles. This page only states the method.
1. Where the records come from
Talent maps read public professional profiles and open job postings. The report library states an 860M+ global pool (metix.ai/reports). The maps do not cite surveys or government labor statistics.
2. What one record is
A visible record is current employment at a named employer in an identifiable function. A visible sample is not official headcount and is not used to estimate a global employee total.
3. How function and industry tags add up
Some maps count function or industry as tags. One posting can carry more than one tag, so tag shares can sum past the posting count. See the overseas jobs structure map. Do not add tag shares up as if they were distinct postings.
4. Movement is not net hiring
Movement between companies is the prior employer visible on a current profile. It is not a complete record of net hires, exits, or retention. The three-lab map states that limit: OpenAI, Anthropic and xAI.
5. The postings-to-talent proxy
Where a map sets open jobs next to identifiable current talent, the proxy is job postings divided by identifiable current professionals. It is not an official vacancy rate. See the AI infrastructure talent race map.
6. Ethnic-Chinese identification
Maps that report an ethnic-Chinese share use five signals together: a name-ethnicity model, Chinese characters, stated Chinese-language ability, education in China, and a multi-spelling surname list. The primary cut is high confidence plus medium confidence. The ethnic-Chinese chapter of the published frontier-lab map is a separate full-population cut. This page does not recompute it.
7. Names
Public pages mask personal names by default. A full name list follows a form request. It is not embedded in the HTML.
8. How to cite
Metix AI. (2026). How Metix talent maps are built. Metix AI Talent Intelligence. https://metix.ai/reports/methodology
9. Maps and the sibling page
Series reports count with a different rule — exclusive company homepage, Senior titles included, open jobs only when measured — on the series methodology page. That page does not explain talent-map samples. This page does not replace that rule.
- Talent map library
- Series counting rules (a sibling page, not a redirect)
- Example: OpenAI, Anthropic and xAI talent structure
- Example: what share of xAI is Human Data & Evaluation
FAQ
Is this official company headcount?
No. A talent map is a visible sample of public professional profiles. Example: the three-lab map’s role, source, and movement comparison covers 9,421 U.S. profiles (https://metix.ai/reports/mapping/openai-anthropic-xai-talent-structure). That denominator belongs to that map. It is not a census of the whole pool.
Why can function shares on the overseas-jobs map sum past 100%?
That map counts tags. One posting can carry more than one tag. See https://metix.ai/reports/mapping/job-distribution-ex-china-2026.
Is the postings-to-talent proxy an official vacancy rate?
No. It is job postings divided by identifiable current professionals. See https://metix.ai/reports/mapping/nvidia-ai-infrastructure-talent-race-2026.
How is ethnic-Chinese identified?
Five signals together: a name-ethnicity model, Chinese characters, stated Chinese-language ability, education in China, and a multi-spelling surname list. The primary cut is high confidence plus medium confidence. The frontier-lab map’s ethnic-Chinese chapter is a separate cut (https://metix.ai/reports/mapping/frontier-ai-labs-talent-2026). This page does not recompute it.
How do I cite one number?
Cite the number from that report or question page, with its date and Metix AI Talent Intelligence. Cite the method from this page. A visible sample is not official headcount.
Ask how a figure was counted
Method questions come to Metix AI. The figure itself still cites the report page.
Start a Mira trial