Talentopian Quarterly Trend Report — 2026 Q2, consolidated multi-lens edition

Scope: a self-published working draft. It has not been peer-reviewed. It describes signals visible in public labor-market data (BLS OEWS, FRED, Indeed Hiring Lab); it is descriptive, not causal, and it is not career-recommendation guidance. What this edition is: seven successive drafts — v0.1, v0.2, v0.3, v0.4, v0.4.1, v0.4.2 and v0.4.3 — consolidated into one report. Each draft added a lens on the same quarter; none replaced its predecessors, so all seven are presented here side by side rather than as a single "latest" version. Reporting window: 2026 Q2 (April 1 – June 30, 2026), covered to date 2026-04-01 → 2026-06-27, 88 of 91 days as stated at the v0.4 draft. The AI-role radar section (§R) uses three weekly snapshots dated 2026-06-19, 2026-06-22 and 2026-06-29. Author: Talentopian Research. Version: consolidated edition of v0.1 → v0.4.3 (original drafts 2026-06-26 → 2026-06-29). The v0.5 Q3 2026 baseline remains reserved for 2026-10-01.

How to read this report

The quarter was written up in three pillars — wage, demand and macro — that measure the state of the US labor market, and three later lenses that look at AI-adjacent and international signals at different granularities. The drafts are explicit that the lenses are not additional pillars: the pillars measure labor-market state, the lenses measure a secular trend, a wider geography and a finer role granularity.

Baseline — what the quarter does not contain (v0.1)

The first draft (2026-06-26) opened the quarter as a pre-data outline. It recorded measurement-infrastructure readiness, not user behaviour, and its own limitations section is the constraint that governs everything below it: the Q2 baseline of real play data is zero, so the draft is, in its own words, "infrastructure-and-pipeline-readiness state, NOT cohort-behavior state". It counted 2 self-published research artifacts for the quarter and 0 peer-reviewed ones.

Two consequences carry through this whole report. First, no finding here is an adoption, cohort or user-behaviour claim — every figure in §S through §R comes from public labor-market data, not from Talentopian usage data. Second, v0.1 also differs slightly on the window: it covered 87 of 91 days, through 2026-06-26, where the later drafts state 88 of 91 through 2026-06-27.

The remainder of v0.1 is internal infrastructure and partnership-preparation detail, which that draft scopes as an internal artifact; it is not reproduced here.

§S — Wage pillar: nominal raises masked real-wage erosion (v0.2)

S.1 — The citation-safe gate

The wage pillar reads BLS OEWS data for 2021 → 2025, inflation-adjusted to 2025 USD. The raw series covers 314 occupations, but it mixes well-covered series with sparse ones whose apparent growth is an artifact. Two filters produce the citable subset: requiring at least 4 years of coverage leaves 309 occupations, and excluding series flagged as likely artifacts (a nominal swing outside −30% to +80%) leaves 308 citation-safe occupations. The exclusion is material — the raw view contains entries such as Prosthodontists at +208% nominal over only 3 years covered, which would mislead any wage claim cited directly from it.

S.2 — Headline finding

Of the 308 occupations with reliable five-year wage data:

• 171 (55.5%) received a nominal pay raise yet lost real purchasing power to inflation
• 175 saw a real-wage decline overall (a nominal raise outpaced by inflation, or a nominal cut)
• Average inflation drag across the reliable set: 18.4%

S.3 — Where the losses and gains fell

Largest real-wage losses in the citation-safe set:

Largest real-wage gains in the same set:

The draft flags both patterns as non-obvious for a career-counselling audience: "academic/postsecondary teaching = worst real losers" — conventional wisdom positions higher-education roles as inflation-protected — and "service/skilled-trade roles = real gainers", against a default assumption that service work is the exposed category.

Bounds. 308 is the citation-safe subset of the 314 occupations in this pull, not the BLS occupational universe, which is much larger. The underlying view refreshes as BLS publishes, so these counts drift; the figures above are as read at the v0.2 measurement window (2026-06-27).

§D — Demand pillar: hiring split sharply by sector (v0.3)

D.1 — What the index is

The demand pillar reads the Indeed Hiring Lab job-postings index by US sector, baselined at Feb-2020 = 100 (above 100 means above pre-pandemic hiring). It runs weekly from 2020 to 2026-06-12 with roughly a two-week lag, and covers 53 sectors, of which 40 US sectors are populated — 15 above baseline and 19 below.

D.2 — Headline findings

Direction over the quarter (latest against the nearest date within the prior 90 days):

• Heating: Civil Engineering +17.7 · Media & Communications +11.0 (off a low base) · Banking & Finance +8.1 · Customer Service +6.2 · Security & Public Safety +5.7
• Cooling: Sales −22.1 · Mechanical Engineering −21.8 · Electrical Engineering −19.5 · Education −10.8 · Medical Technician −9.2

D.3 — Cross-pillar observation

The draft records the pairing between this pillar and the wage pillar in one line: "healthcare = high demand + … anti-copilot-resistant; knowledge/creative work soft on BOTH demand and real wages (AI-exposure consistent)". Healthcare-adjacent work reads consistently across demand and wages; knowledge and creative work reads soft on both — Data & Analytics, Media and Marketing all sit below the hiring baseline, and academic and postsecondary roles cluster among the worst real-wage losers in §S.3.

Bounds (verbatim caveats). US only; the index measures postings volume, a demand signal, not hires and not wages; roughly a two-week data lag; and Indeed's sector taxonomy is not the WEF/SOC taxonomy used elsewhere in our work, so the two must not be hard-joined — comparisons are made at the label level only.

§M — Macro pillar: a frozen labor market (v0.4)

M.1 — The backdrop

The macro pillar reads 12 US labor indicators from FRED (unemployment, U-6, participation, JOLTS openings / hires / quits, initial claims, payrolls, average hourly earnings, average weekly hours, the employment cost index and CPI), each as latest plus year-over-year and quarter-over-quarter.

The pattern the draft names: openings up, hires down, quits down and claims low together describe a low-hire, low-fire market — employers neither shedding nor absorbing workers, with mobility dried up. The flat 4.3% headline unemployment rate masks the softening that is visible only in U-6 and participation.

M.2 — Cross-validation of the wage finding

Average hourly earnings rose +3.4% year-over-year ($36.28 → $37.53) against CPI at +4.2% (320.6 → 334.0), giving real earnings of approximately −0.7% year-over-year. The employment cost index moved at the same sub-CPI pace, +3.4%.

This matters because it is an independent route to the §S finding. The sector-level result (171 of 308 occupations with a nominal raise and a real loss) comes from BLS occupational data with per-occupation inflation anchoring; the macro result comes from FRED national aggregates via an earnings-versus-CPI gap. Two distinct sources and two distinct methods reach the same directional conclusion, which makes the real-wage decline a more robust reading than either alone.

Bounds. US national aggregates, at mixed frequencies — JOLTS and the employment cost index lag by roughly two months, claims run weekly to 06-13 — and year-over-year and quarter-over-quarter comparisons use the nearest observation at or before the offset. CPI here is headline CPI-U, not core CPI and not PCE. The view self-refreshes, so these values are as read at the v0.4 measurement window (2026-06-27) and will move as the agencies publish.

§T — What the three pillars support together (v0.4)

With wage, demand and macro in place, the v0.4 draft consolidated what the triad does and does not license.

1. Real-wage decline is the robust finding. Sector-level (§S) and macro-level (§M.2) evidence converge by independent methods, as set out above.
2. The AI-exposure reading is consistent across pillars, with a retraction attached. Knowledge and creative work is soft on demand (§D) and academic and postsecondary roles are among the worst real-wage losers (§S.3). A companion analysis also reported a resilience-versus-creativity association — see the note below, which is load-bearing.
3. Healthcare converges at the positive end. High postings demand (§D), relative resistance to automation, and a low-hire / low-fire macro backdrop (§M.1) all point the same way for healthcare-adjacent work at the label level.

Retraction note (carried forward, do not read past it). The companion anti-copilot analysis had reported that its resilience score was negatively correlated with creativity at multiple-comparison-corrected significance. Those significance claims were retracted in that report's later revision: at the true per-pair sample sizes — 43 to 119, the result of pairwise deletion over columns with missing values, with the creativity dimension 75% missing — no pair survives Bonferroni correction. The directional pattern is still descriptively visible at a raw correlation of −0.30, but it rests on N = 43 and is not robust to correction. The convergent reading in point 2 above rests on the wage and demand data, not on the retracted significance claims.

T.1 — Scope bounds: convergent, not joined

§A — AI-adoption curve lens (v0.4.1)

The first lens beyond the triad reads the Indeed Hiring Lab AI Tracker: the share of job postings mentioning AI, across 9 countries, daily, from 2019-01 to 2026-05.

• United States: 1.7% of postings mentioned AI in the 2019 baseline, 5.7% at the latest reading (2026-05) — 3.3× since 2019, with a year-over-year change of +2.82pp
• International leader — Canada: approximately 17% of postings, 9.4× since 2019
• All 9 tracked countries are up, in a range of 1.6× to 9.4× growth since 2019

This is a lens, not a fourth pillar. The pillars measure what the US labor market is doing; this measures how fast AI is penetrating job-posting language across several countries — a secular trend rather than a labor-market state. Read against §T, it supplies the temporal component: the AI-exposure picture is not a static observation but a trend that is intensifying, and healthcare's relative resistance becomes more valuable, not less, as adoption accelerates.

Bounds (verbatim caveats). This is keyword share, not displacement — a posting mentioning AI may be augmented, replacing or merely adjacent. Country and language composition affects the shares. Only the ai metric is populated; the generative-AI sub-metric is not. Korea is not among the 9 tracked countries.

§I — International hiring-momentum lens (v0.4.2)

The second lens extends the US-only demand pillar to a 6-country Indeed Hiring Lab set, reporting the top-heating sector per country.

The draft describes a six-country set but names only these five; the sixth is not identified in the source, so it is not named here.

Two readings follow, both partial:

• Healthcare is a partial multi-country pattern. US Pharmacy and French Nursing are both top-heating in their own countries, which extends the §T healthcare convergence beyond the US — but the other named countries show non-healthcare top sectors, so the pattern is partial, not universal.
• Adoption rate and hiring momentum are partly decoupled. Canada leads the §A adoption ranking at roughly 17% and 9.4× since 2019, yet its top-heating sector is Electrical Engineering, not a knowledge-work or AI-front-line sector. A country's AI-adoption rank does not determine which of its sectors heats.

Bounds. The verbatim caveat is "small-sector volatility": a single top sector per country, in a multi-sector economy, is noisy at small sample sizes. The draft also rules out five readings explicitly: a top-heating sector is not a career recommendation; healthcare is not universally top-heating (2 of 6 countries); a heating sector is a short-window descriptor, not a structural economic forecast; 6 countries is not comprehensive global coverage; and adoption is not shown to cause hiring-momentum shifts — the two are co-observed at country level, and causation would need controlled analysis that small samples and endogeneity currently rule out.

§R — AI-role emergence radar lens (v0.4.3)

R.1 — What the radar does

The third lens works one granularity level below sector: it takes weekly snapshots of AI-adjacent job postings and classifies specific role labels by week-over-week momentum. Three snapshots exist — 2026-06-19, 2026-06-22 and 2026-06-29 — which the drafts record as sufficient for a first week-over-week computation. Automated freshness monitoring guards against a stale-snapshot condition silently freezing the surface.

R.2 — First-window findings (preliminary)

5 of the rising labels are job titles our production career taxonomy does not yet recognise, and are flagged as taxonomy candidates. The framing is deliberate and preserved from the source: the radar surfaces candidates for taxonomy expansion; integration is a human review decision, never automatic. That gate is what keeps taxonomy expansion inside the no-fabrication discipline.

R.3 — Reading against the adoption lens

The §A aggregate share growth (1.7% → 5.7%, 3.3× since 2019) and the §R label-level view meet at one point: some of that aggregate growth is compositional — new role labels appearing, not only growth within labels that already existed. The draft is careful to state this as a partial explanation rather than a cause, on a sample of three weekly snapshots.

The review gate is also the rate limiter. The radar produces a continuing queue of taxonomy-candidate roles whose integration depends on review decisions; the lens makes that queue visible and prescribes nothing about it.

R.4 — What this lens does not support

• Not "these are the most important emerging jobs" — the radar surfaces every rising label absent from the taxonomy; ranking importance would need salary, demand and sector-fit analysis the radar does not contain
• Not "add these to the taxonomy automatically" — the surface is human-review-gated with no auto-add
• Not "Talentopian recognises these roles now" — a candidate flag is not a recognised status, and production matching for these roles stays gated on review
• Not "week-over-week momentum predicts long-term role viability" — a three-snapshot window is too short for trend confidence, and first-window classifications will revise as history accumulates. The single "new (no prior snapshot)" label is a bootstrap artifact that will move into the rising or stable buckets as snapshots accumulate
• Not "these are new careers to pursue" — the radar is a taxonomy-input surface, describing what production matching should cover, not a career-recommendation surface

Scope and limitations

What this report is. A consolidation of seven self-published drafts covering 2026 Q2, describing signals visible in public labor-market data — BLS OEWS wages, Indeed Hiring Lab postings, FRED macro indicators, the Indeed AI Tracker and our own weekly snapshots of AI-adjacent postings.

What it is not.

• Not peer-reviewed. These are self-published working drafts. Peer review is a separate, planned path; nothing here has been through it.
• Not a causal analysis. Every relationship described here is an association observed in descriptive data. Nothing in this report identifies AI adoption, or anything else, as the cause of a wage or hiring movement.
• Not career-recommendation guidance. No section should be restated as advice to enter or avoid an occupation. The §S wage finding in particular must not be restated as "you will lose purchasing power in this job".
• Not global. The three pillars are US-only. The adoption lens covers 9 countries and the international hiring lens 6. Korea is in none of these datasets, so any Korean-market reading would be an inflation of the evidence.
• Not a joined dataset. The pillars use different taxonomies and are compared at the label level only; cross-pillar arithmetic and occupation-level joins are not supported (§T.1).
• Not a marketing report. There are no audited financial, user or revenue figures here.
• Not fixed in time. Every underlying view self-refreshes as the source agencies publish. Figures are as read at the measurement window stated in each section — 2026-06-27 for the pillars and the adoption lens, 2026-06-29 for the two later lenses — and will drift.

How to cite these figures

Version history, and where the drafts disagree

Consolidating seven drafts surfaced four places where they do not agree. None is resolved silently here:

1. The reporting window. v0.1 states 87 of 91 days, through 2026-06-26; v0.2 through v0.4 state 88 of 91, through 2026-06-27. This page uses the later figure and records the earlier one.
2. Quarterly heating and cooling figures. v0.3 reports these rounded (Sales −22, Mechanical Engineering −22, Electrical Engineering −20, Civil Engineering +18, Banking +8); the demand-pillar source it cites carries them unrounded (−22.1, −21.8, −19.5, +17.7, +8.1). §D.2 above uses the unrounded values.
3. Lens ordinals. The drafts number the lenses inconsistently — v0.4.1 calls itself the fourth lens; v0.4.2's header calls itself the fourth while its own body calls it the fifth; v0.4.3 also calls itself the fifth. This page therefore identifies each lens by draft version and section letter and does not use an ordinal.
4. The sixth country. v0.4.2 describes a six-country set but names five. The sixth is left unnamed here because the source does not name it.

Provenance

• Original drafts: v0.1 through v0.4.3, authored 2026-06-26 → 2026-06-29 by Talentopian Research.
• This consolidated edition: 2026-09-01. It merges the seven drafts into one report and reconciles the four disagreements listed above. It adds no new data and no new analysis.
• Next scheduled update: the v0.5 Q3 2026 baseline, reserved for 2026-10-01. The drafts note that any further distinct data lens before then — an age-demographic or education-flow lens, for example — would ship as a v0.4.x increment rather than as v0.5, and that same-axis refinements get inline notes instead of a version bump.
• Saturation, acknowledged: three lenses on AI-adjacent labor signals accumulated in quick succession. The v0.4.3 draft records the standing test for the next one — whether an additional lens genuinely crosses a granularity boundary, or whether refining an existing lens is the more truthful frame.

— Talentopian Research (consolidated multi-lens edition of v0.1 → v0.4.3, 2026-09-01)

3,588 words.  ·  All research  ·  Talentopian home