A new attempt to observe AI’s labour-market impact

Revelio Labs launched its AI Labor Market Tracker on July 28, 2026, presenting it as a monthly measure of how artificial intelligence is changing the US workforce. The launch matters because the public debate has often moved faster than the data. Predictions of widespread displacement sit alongside claims that AI will create more jobs than it removes, yet broad employment totals are poorly suited to identifying where task automation, changing hiring practices and new skill requirements are emerging.

The tracker is designed to look beneath those aggregate figures. It draws on Revelio Labs’ workforce and job-posting data to follow labour supply and demand, employment, advertised pay, job content and aspects of the hiring process. Its timeline begins with the public release of ChatGPT in November 2022, allowing users to compare developments in more- and less-AI-exposed work over the same period.

That approach is useful precisely because AI is unlikely to affect every occupation in the same way. A technology that can draft text, write code or summarise information may reduce the time required for particular tasks without eliminating the broader role. Conversely, it can change which skills employers seek, which jobs they post and how teams are organised well before any change becomes obvious in economy-wide payroll data.

The headline findings point in different directions

The tracker’s initial results underline that distinction. Revelio Labs says hiring demand in occupations it classifies as most vulnerable to AI, including data engineers and financial analysts, has fallen by 36% relative to the least exposed occupations since November 2022. It also reports that the most significant recent changes in the mix of work activities have taken place within occupations rather than through a wholesale shift between occupations.

Read narrowly, that is evidence of adjustment rather than a definitive measure of jobs destroyed. A fall in job-posting demand can mean fewer new openings, slower expansion, greater productivity from existing staff, a cyclical downturn, or a mix of those forces. It does not on its own establish that AI caused the change, nor does it show how existing workers’ hours, duties or earnings have changed.

The tracker’s other headline result complicates a simple displacement narrative. It says companies that have successfully adopted AI have recorded a 27% increase in headcount as they reorganise around new requirements. That result is directionally consistent with earlier research by Revelio Labs and Ramp, which linked observed AI vendor spending to employment records. Their analysis found that high-intensity adopters had higher total and entry-level headcount over the two years after adoption, while low-intensity adopters showed no statistically significant employment increase.

The two findings can coexist. AI can substitute for some tasks and reduce recruitment in particular roles while helping productive firms expand elsewhere. The central question is therefore not whether AI is universally job-creating or job-destroying, but who captures the productivity gains, which tasks are automated, and whether workers can move into complementary work quickly enough.

Measurement is the tracker’s strength — and its constraint

A recurring monthly dashboard can make labour-market changes more visible, especially when it separates occupations, companies, skills and wages. Public employment statistics are comprehensive but typically lag developments and have limited detail on job tasks. Job postings can reveal employers’ intentions earlier, while professional-profile data can help track movements in headcount and occupational composition.

However, those advantages do not remove the underlying identification problem. AI adoption is not random. Firms that invest heavily in AI may already be larger, more technical, better capitalised or growing faster than their peers. The Ramp-Revelio study explicitly found that adopters differed from non-adopters before adoption. Its design sought to address this by comparing adopters with firms that adopted later, but the authors still frame the evidence as an early look rather than a final verdict on long-term effects.

The data source also affects what can be inferred. Online profiles and job advertisements give rich, timely signals, but they are not a complete census of employment. They can be more representative of digitally visible and white-collar work than of the entire labour market. The tracker will be most valuable if it clearly documents its exposure measures, company classifications, occupational coverage, revisions and uncertainty alongside its charts.

Why correlation requires caution

External work offers a useful counterweight. Federal Reserve Bank of New York researchers examined US job-posting data using an occupational AI-exposure measure and found little evidence of a distinct AI-driven decline in labour demand to date. Although postings in more-exposed occupations were relatively weaker, the divergence had begun before ChatGPT’s release and did not show a clear further break afterwards. The researchers likewise found no clear post-2022 divergence between junior and senior postings within highly exposed occupations.

This does not contradict Revelio Labs’ results so much as it shows why methodology and definitions matter. One analysis may compare a specific group of highly exposed occupations against less-exposed ones; another may focus on changes around a defined event date and test whether earlier trends undermine a causal interpretation. Both can identify real patterns while reaching different conclusions about causation.

Academic research using firm-occupation data has similarly suggested a mixed mechanism: AI can substitute for labour at the task level, but productivity gains and reallocation towards complementary tasks can offset the reduction in demand at the job or firm level. The net effect can therefore be muted in aggregate even as disruption is substantial for particular workers.

A practical tool, not a final answer

For employers, the tracker could help distinguish a genuine skills transition from a broad hiring slowdown. For policymakers, it may offer earlier warnings of uneven impacts across occupations, wages and career stages. For workers, its most useful contribution may be a clearer view of how job content changes, rather than a binary assessment of whether a profession is “safe”.

Its credibility will depend on disciplined interpretation. A monthly tracker should be treated as a system for monitoring signals and testing hypotheses, not as a machine for assigning every hiring decline or wage shift to AI. The effects of a general-purpose technology will unfold alongside interest rates, business cycles, restructuring, outsourcing and changing consumer demand.

Revelio Labs has created a timely framework for observing those changes at a more detailed level. The next test is whether subsequent editions maintain methodological transparency and reveal durable patterns, including where AI’s gains and costs are distributed across the workforce.

Sources