Sources and coverage
Which feeds are searched, which topics are tracked, when collection started, and where coverage is thin.
Where papers come from
- OpenAlex
- Open catalogue of scholarly work. The broadest of the three, and the main route in for published journal articles.
- arXiv
- Preprints, primarily economics and computer science. This is where the newest work shows up first, often months before publication, and also where the least filtered work lives.
- Semantic Scholar
- Adds citation context and helps resolve records that the other two describe inconsistently.
Topics tracked
Search is scoped to nine themes. A paper must land in at least one:
- AI and labour productivity — output, efficiency, task completion
- AI and labour markets — employment, wages, displacement, demand
- Human-AI collaboration — how people and systems divide work
- AI and skills or training — what becomes valuable, and reskilling
- AI and organisational design — firm structure, hierarchy, coordination
- AI and innovation — research productivity, discovery, invention
- AI and inequality — distributional effects across workers and firms
- AI adoption and diffusion — who adopts, how fast, and why
- AI governance and policy — regulation and institutional response
The corpus has a start date
Collection began in late 2025. Work published before then is largely absent, including most of the papers people think of as the canon of this field. This is the single most surprising thing about the corpus, and it is a property of when collection started, not a fault in search.
In practice that means searching for a well-known author in AI economics can return little or nothing by them. Acemoglu, Bloom and Dell'Acqua have no author record here at all; Autor and Brynjolfsson have very few. Their pre-2025 work is simply outside the window the pipeline has covered.
Two consequences worth holding on to:
- Search will still return a full page of results. Ranking is semantic, so a query always finds its nearest neighbours. A page of results for "Acemoglu" is not evidence that any of them are by Acemoglu. Read the author line, not the result count.
- This corpus is not the place to establish precedence. If you need the earlier literature, use the reference checker, which searches an economics corpus reaching back to 1995, or go to the original databases.
Where coverage is thin
Absence from this corpus is not evidence that no such research exists.
Known and structural gaps:
- English-language work only. Research published in other languages is largely invisible here.
- Journal publication lags. Preprints arrive quickly; published versions can be months behind, so recent periods over-represent preprints.
- Books, chapters and reports. Poorly covered by all three feeds. Policy and institutional research suffers most from this.
- Paywalled work without an accessible abstract. If nothing can be read, nothing can be assessed.
- Anything the relevance gate rejected. The gate is deliberately strict, and it will sometimes be wrong. See the intake pipeline.
The coverage view shows which themes and outcome categories are actually well populated, which is the fastest way to see whether an apparent consensus rests on a handful of papers.