Electrician apprenticeship searches: +49% YoY
Transcription course searches: -63% YoY
Copywriter wages: -9.5% real (2022–2024)
Bookkeeping jobs lost: -94,980 in 2 years
New electrician jobs created: +52,530 (2022–2024)
Architecture real wage growth: +8.4% - highest in study
Proofreaders left in America: 5,160 vs 742,580 electricians
Web Developer jobs: -9,760 despite high wages
Electrician apprenticeship searches: +49% YoY
Transcription course searches: -63% YoY
Copywriter wages: -9.5% real (2022–2024)
Bookkeeping jobs lost: -94,980 in 2 years
New electrician jobs created: +52,530 (2022–2024)
Architecture real wage growth: +8.4% - highest in study
Proofreaders left in America: 5,160 vs 742,580 electricians
Web Developer jobs: -9,760 despite high wages
2026
We tested the AI job
predictions. Here's what
actually happened.
Oxford (2017) and WEF (2025) predicted which skills AI would displace. We collected 3 years of post-ChatGPT data - search trends, keyword volumes, government wages, employment records - to test those predictions. Some held. Some were wrong. The data tells the story.
+49%
Electrician apprenticeship searches, year-on-year
−9.5%
Real copywriter wage loss, May 2022 → May 2024
−63%
Transcription course searches collapsed in one year
The Study
What Oxford and WEF Predicted
Before we show you what actually happened, here is what the experts predicted. Two major institutions put real intellectual effort into answering the question "which jobs will AI take?" The Oxford Future of Employment study (2017) ran 702 US occupations through a machine learning model that looked at the tasks each job involves and calculated an automation probability from 0% to 99%. Meanwhile, WEF's Future of Jobs report (2025) asked 1,000+ employers across 55 countries what they expect to hire more of - and less of - by 2030. The two studies broadly agree: physical, hands-on, interpersonal, and creative-performance skills are hard for AI to replace. Text processing, data handling, and routine digital tasks are very easy for AI to replace. Below are all 29 skills we studied, split by what the researchers predicted.
Cohort A & Languages (17 skills) - Predicted AI-safe
Electrical Work
Oxford: 15% automation risk
Nursing & Healthcare
Oxford: 0.9% automation risk
Architecture
Oxford: 1.8% automation risk
HVAC
WEF 2025: electrotechnology roles explicitly growing
Yoga & Meditation
WEF 2025: physical instruction, zero GenAI substitution
Cooking & Culinary
WEF 2025: food sector lowest manual skill decline
Negotiation & Leadership
WEF 2025: 3rd fastest-growing skill cluster by 2030
Guitar
WEF 2025: physical performance, sensory - zero substitution
Piano
WEF 2025: physical mastery, real-time correction - zero substitution
Sign Language (ASL)
WEF 2025: physical gesture, interpersonal communication
Dance
WEF 2025: kinesthetic instruction, physical presence
Massage Therapy
WEF 2025: manual dexterity and sensory - no substitution
Pottery & Ceramics
WEF 2025: manual dexterity, sensory processing
Automotive Repair
WEF 2025: mechanics and machinery repairers growing
Spanish
WEF 2025: relational language teaching, very low substitution
French
WEF 2025: same as Spanish
Japanese
WEF 2025: three writing systems add complexity AI cannot replicate in instruction
Cohort B & Contested (12 skills) - Predicted AI-vulnerable or mixed
Copywriting
WEF 2025: high GenAI substitution for writing and editing
Transcription
Oxford: 97% automation probability
Data Entry
Oxford: 99% - highest score in the entire study
Proofreading
Oxford: 84% automation probability
Bookkeeping
Oxford: 98% automation probability
Graphic Design
WEF 2025: first time named as a net declining role
SEO
WEF 2025: fastest-declining tech-adjacent digital marketing
Web Design
WEF 2025: no-code AI builders displacing routine web work
Video Editing
WEF 2025: media and entertainment - highest AI displacement sector
Plumbing
Oxford: 35% risk - contested, trade skills showing strong demand
Drawing & Illustration
WEF 2025: medium risk - demand resilient despite AI image tools
Welding
Oxford: ~48% risk - standardised task components at risk
Key Findings
What the data showed
Three years of people quietly telling us what they think through their wallets, their job postings, and their Google searches. We grouped the biggest signals by type - because mixing "search is up 49%" with "wages fell $900" in the same breath is confusing. Here is what each data source had to say, starting with the sharpest signals.
Search Demand Findings
Google Trends + GKP
+49%
Year-on-Year · Electricians
Electrician apprenticeship searches surging year-on-year
When AI started doing knowledge work, people went looking for jobs AI cannot do. Demand to become an electrician shot up +49% year-on-year and nearly doubled over the full 3-year study period. There are now 742,580 electricians employed nationally, up 52,530 from 2022. The strongest single search signal in the entire study. Turns out the answer to "what will AI not take?" is partly: the person who fixes your fuse box.
-63%
Year-on-Year · Transcription
Transcription demand collapsed. AI tools made the market disappear.
Transcription course searches fell 63% in a single year - the sharpest collapse of any skill we tracked. After OpenAI released Whisper (free, near-perfect transcription), and tools like Otter.ai became mainstream, there was simply no market left for human transcription training. This is not a forecast about something that might happen. It already happened, and you can see it in the data.
-33%
Year-on-Year · Graphic Design
Graphic design learning searches down 33% - and secondary terms confirm it
The primary keyword "graphic design course" initially looked stable. But checking "graphic design classes" and "learn graphic design" revealed both declining. The search signal, combined with WEF naming graphic designers a declining role in 2025, is consistent across methods.
Wage Findings
BLS OEWS May 2022-2024
+8.4%
Real Wage Growth · Architecture
Architecture real wages grew most - AI tools augment, not replace
Architects: $82,840 (2022) to $96,690 (2024). Nominal +16.7%, real +8.4% after inflation - the highest real wage gain in the study. Client judgment, regulatory navigation, and site decision-making remain irreducibly human. AI design tools are making architects more valuable, not less.
-9.5%
Real Wage Loss · Copywriting
Copywriter wages fell in real terms - 6,210 jobs disappeared
Writers and Authors median wage: $73,150 (2022) falling to $72,270 (2024). Nominal decline of -1.2% equals a real loss of -9.5% after inflation. Jobs fell from 54,010 to 47,800 simultaneously. Wages and employment declining together is the economic signature of technology displacement.
-19.5%
Real Wage Loss · Broadcasting
Broadcasting wages down 19.5% in real terms - the steepest fall
Broadcast Technicians wages fell from $60,700 to $53,920 (-11.2% nominal, -19.5% real). This is the largest real wage decline of any occupation in the study. AI content tools are visibly restructuring media employment at both the wage and headcount level simultaneously.
Employment Findings
BLS OEWS Job Counts 2022-2024
5,160
Total Jobs Left · Proofreading
Only 5,160 proofreaders employed nationally in 2024
For context: 742,580 electricians. 3,282,010 nurses. The proofreading occupation has not just declined - it has effectively ceased to exist as a standalone career. Oxford 2017 predicted 84% automation probability. Government employment data shows that prediction was correct.
+21%
Job Growth · Fitness and Yoga
Fitness trainer jobs grew +21.2% - physical instruction expanding fast
250,540 to 303,620 jobs - a gain of 53,080 positions in two years. Wages only grew +1.8% nominally because supply is meeting high demand. The sector is expanding rapidly. Physical, embodied instruction is growing as a category precisely because it cannot be automated.
-36%
Job Loss · Broadcasting
Broadcasting lost 11,940 jobs - 36% of the workforce gone in two years
33,020 broadcast technician jobs in 2022. Only 21,080 in 2024. More than one in three positions eliminated. Combined with the -19.5% real wage drop, this is the most complete evidence of AI-driven sector disruption in government employment data.
What is the TRS, and how does it work?
ChatGPT launched on November 30, 2022. That date is the dividing line for this entire study. Before it: Jan-Oct 2022. After it: Nov 2022-Mar 2026. The Temporal Resilience Score (TRS) is just the ratio of those two periods. If 100 people were googling "guitar lessons" before ChatGPT and 115 are googling it now, the TRS is 1.15. If 100 people searched for "transcription course" before and only 60 are doing so now, the TRS is 0.60. Simple. No black box. The number tells you whether demand for a skill went up, held steady, or collapsed once AI became mainstream. We tracked all 29 skills monthly across 75 data points each - that is over 2,000 individual data readings.
Temporal Resilience Score
TRS = Post-ChatGPT Avg ÷ Pre-ChatGPT Avg
(Nov 2022–Mar 2026) ÷ (Jan–Oct 2022)
How to read it
TRS > 1.20 = Demand grew
TRS 0.90-1.20 = Stable
TRS < 0.90 = Demand fell
Method 1 · Google Trends
Google Trends Data: Which Skills Held Up After ChatGPT
Here is where the story really starts. We used Google Trends to track how much people searched for each of our 29 skills every month from January 2020 to March 2026 - covering more than two years before ChatGPT launched and more than three years after. The key question: did search demand for a skill go up or down once AI tools became mainstream? The number we use to measure this is called the Temporal Resilience Score (TRS). It is simply the average search interest after ChatGPT divided by the average search interest before ChatGPT. A TRS of 1.0 means nothing changed. A TRS of 1.5 means demand grew by 50%. A TRS of 0.8 means demand fell by 20%. Green = growing or stable. Red = declining. Click any card to see the Oxford/WEF source and available data.
Cohort A
13,200/mo
Negotiation
2.17
Growing
Temporal Resilience Score
YoY Search
−55%
AI Risk
Low
Cohort A
66,300/mo
Electrical Work
1.92
Growing
Temporal Resilience Score
YoY Search
+49%
AI Risk
Low
Labour market: Wages +3.5% | Jobs +7.6% (+52,530)
Oxford 2017 - Electricians: 15% automation probability
Cohort A
9,970/mo
Architecture
1.76
Growing
Temporal Resilience Score
YoY Search
−12%
AI Risk
Low
Labour market: Wages +16.7% nominal, +8.4% real | Jobs +3.4%
Oxford 2017 - Architects: 1.8% automation probability
Cohort A
66,290/mo
HVAC
1.44
Growing
Temporal Resilience Score
YoY Search
−18%
AI Risk
Low
Labour market: Wages +16.4% real | Jobs +5.9%
WEF 2025 - Electrotechnology engineers: growing role
Cohort A
109,210/mo
Cooking & Culinary
1.20
Stable
Temporal Resilience Score
YoY Search
+22%
AI Risk
Low
Cohort A
14,000/mo
Automotive Repair
1.20
Stable
Temporal Resilience Score
YoY Search
−19%
AI Risk
Low
Labour market: Wages +5.7% nominal | Jobs +3.7%
WEF 2025 - Mechanics & machinery repairers: growing role
Cohort A
69,120/mo
Nursing & Healthcare
1.17
Stable
Temporal Resilience Score
YoY Search
−46%*
AI Risk
Low
Labour market: Wages +15.2% nominal | Jobs +6.8%
Oxford 2017 - Registered Nurses: 0.9% automation probability. *YoY decline reflects post-COVID normalisation
Cohort A
11,800/mo
Pottery & Ceramics
1.10
Stable
Temporal Resilience Score
YoY Search
−18%
AI Risk
Low
Cohort A
119,300/mo
Yoga & Meditation
1.09
Stable
Temporal Resilience Score
YoY Search
+22%
AI Risk
Low
Cohort A
40,000/mo
Sign Language (ASL)
1.03
Stable
Temporal Resilience Score
YoY Search
+23%
AI Risk
Low
Cohort A
83,110/mo
Guitar
1.02
Stable
Temporal Resilience Score
YoY Search
+22%
AI Risk
Low
Cohort A
63,510/mo
Piano
0.97
Stable
Temporal Resilience Score
YoY Search
−18%
AI Risk
Low
Cohort A
14,730/mo
Dance
0.96
Stable
Temporal Resilience Score
YoY Search
−33%
AI Risk
Low
Cohort A
7,430/mo
Massage Therapy
0.87
Declining
Temporal Resilience Score
YoY Search
−45%
AI Risk
Low
Labour market: Wages +16.2% nominal, +7.9% real | Jobs +11.3%
WEF 2025 Box 3.1 - Manual dexterity + sensory: no substitution potential
Language
21,430/mo
Japanese
1.26
Growing
Temporal Resilience Score
YoY Search
−21%
AI Risk
Low
Language
28,310/mo
French
1.08
Stable
Temporal Resilience Score
YoY Search
−34%
AI Risk
Low
Language
49,140/mo
Spanish
1.01
Stable
Temporal Resilience Score
YoY Search
−18%
AI Risk
Low
Contested
33,920/mo
Plumbing
1.91
Growing
Temporal Resilience Score
YoY Search
−16%
AI Risk
Medium
Contested
22,810/mo
Welding
1.43
Growing
Temporal Resilience Score
YoY Search
−19%
AI Risk
Medium
Contested
52,350/mo
Drawing & Illustration
1.08
Stable
Temporal Resilience Score
YoY Search
−18%
AI Risk
Medium
Cohort B
3,600/mo
Video Editing
2.58
Growing
Temporal Resilience Score
YoY Search
0%
AI Risk
High
Labour market: Broadcasting wages −19.5% real | Jobs −36.2%
WEF 2025 - Media & entertainment: highest AI task displacement sector. *See anomaly section.
Cohort B
5,400/mo
SEO
1.81
Growing
Temporal Resilience Score
YoY Search
−19%
AI Risk
High
Cohort B
720/mo
Transcription
1.68
Growing
Temporal Resilience Score
YoY Search
−63%
AI Risk
High
Labour market: Oxford 97% - now confirmed. YoY −63% = market collapse.
Oxford 2017 - Stenographers: 97% automation probability
Cohort B
1,300/mo
Data Entry
1.61
Growing
Temporal Resilience Score
YoY Search
−38%
AI Risk
High
Labour market: Wages +10.1% nominal, +1.8% real | Jobs −14.0% (−22,100)
Oxford 2017 - Data entry keyers: 99% automation probability
Cohort B
3,600/mo
Web Design
1.37
Growing
Temporal Resilience Score
YoY Search
−33%
AI Risk
High
Labour market: Wages +15.7% nominal, +7.4% real | Jobs −11.0% (−9,760). Barbell effect.
WEF 2025 - No-code AI builders displacing routine web design
Cohort B
1,300/mo
Proofreading
0.92
Stable
Temporal Resilience Score
YoY Search
−55%
AI Risk
High
Labour market: Only 5,160 jobs nationally in 2024. Occupation has nearly ceased to exist.
Oxford 2017 - Proofreaders: 84% automation probability
Cohort B
3,600/mo
Copywriting
0.92
Stable
Temporal Resilience Score
YoY Search
−34%
AI Risk
High
Labour market: Wages −9.5% real | Jobs −11.5% (6,210 lost)
WEF 2025 Box 3.1 - Writing & editing: high GenAI substitution capacity
Cohort B
6,600/mo
Bookkeeping
0.82
Declining
Temporal Resilience Score
YoY Search
−19%
AI Risk
High
Labour market: Wages +7.3% nominal, −1.0% real | Jobs −6.1% (−94,980)
Oxford 2017 - Bookkeeping clerks: 98% automation probability
Cohort B
12,100/mo
Graphic Design
0.79
Declining
Temporal Resilience Score
YoY Search
−33%
AI Risk
High
Labour market: Wages +5.7% nominal, −2.6% real | Jobs +1.1%
WEF 2025 - Graphic designers: net declining role (named for first time 2025)
Source: Google Trends Jan 2020–Mar 2026 · 29 skills · 75 monthly data points each · Normalised to common anchor (guitar lessons Jan-2022 = 68) · Wiingy Research 2026
Method 2 · Google Keyword Planner
GKP Data: What People Are Searching to Learn
Google Keyword Planner tells us how many people type things like "electrician apprenticeship" or "transcription course" into Google each month. We compared search volumes from April 2025 to March 2026 against the exact same months one year earlier. This is the freshest signal in the study - it tells you what people are doing right now, not two years ago. How to read it: a positive percentage means more people searched for that skill this year than last. A negative means fewer. Think of it as the market voting with its curiosity. We now have YoY data for all 29 skills. The final 5 (Spanish, French, Japanese, Pottery, and Web Design) were collected separately using their primary learning-intent keyword variants. Every single Cohort B skill is down. Not one is growing.
↑ Skills people are searching MORE
5 of 29 skills
Electrical Work
Cohort A
+49%
Sign Language (ASL)
Cohort A
+23%
Yoga & Meditation
Cohort A
+22%
Cooking & Culinary
Cohort A
+22%
Guitar
Cohort A
+22%
↓ Skills people are searching LESS (or flat)
24 of 29 skills
Transcription
Cohort B
−63%
Negotiation
Cohort A
−55%
Proofreading
Cohort B
−55%
Nursing & Healthcare
Cohort A
−46%*
⚠️
* This number is misleading - do not read it as an AI threat
Why is nursing at −46%? COVID created an artificial peak - not AI.
During 2020–2021, millions of people searched for nursing training as the pandemic made healthcare careers visible and urgent. That was an extraordinary spike. The −46% YoY is simply those searches returning to normal levels - a correction of an anomaly, not a collapse caused by automation.
BLS Real Wages
+6.9%
above inflation · May 2022–2024
New Jobs Added
+209,310
jobs created in 2 years · BLS 2022–2024
Oxford Automation Risk
0.9%
probability · one of the lowest in the study
✓ Nursing is one of the strongest AI-resilient signals in this entire study. BLS confirms it across wages, jobs, and automation risk.
Massage Therapy
Cohort A
−45%
Data Entry
Cohort B
−38%
French
Language
−34%
Copywriting
Cohort B
−34%
Graphic Design
Cohort B
−33%
Dance
Cohort A
−33%
Web Design
Cohort B
−33%
Japanese
Language
−21%
Welding
Contested
−19%
Automotive Repair
Cohort A
−19%
Bookkeeping
Cohort B
−19%
SEO
Cohort B
−19%
Piano
Cohort A
−18%
Drawing & Illustration
Contested
−18%
HVAC
Cohort A
−18%
Pottery & Ceramics
Cohort A
−18%
Spanish
Language
−18%
Plumbing
Contested
−16%
Architecture
Cohort A
−12%
Video Editing
Cohort B
0%
YoY for French, Japanese, Spanish, Pottery & Ceramics and Web Design collected Apr 2024-Mar 2025 vs Apr 2025-Mar 2026 using primary learning-intent keywords.
Source: Google Keyword Planner, Apr 2025-Mar 2026 vs Apr 2024-Mar 2025. All 29 skills. Source: Wiingy Research
Method 3 · BLS Wages and Employment
BLS Data: What the
Labour Market Is Paying
Think of this as the most honest signal of all - what actual employers are paying, and how many positions exist. We pulled three years of data from the US Bureau of Labor Statistics Occupational Employment and Wage Statistics survey - government data covering 1.2 million employers nationally. We tracked median wages and total jobs for May 2022, May 2023, and May 2024. Then we adjusted for inflation: prices rose 8.3% cumulatively over this period (BLS CPI-U). So a skill where wages grew by 5% actually lost ground in real terms - workers can buy less than they could before. How to read this: if a skill shows wages above inflation AND growing employment, that is a resilient skill. If wages fell AND jobs shrank simultaneously, that is the economic fingerprint of AI displacement. Green pills = beat inflation. Red pills = lost purchasing power.
| Skill / Occupation | 2022 Wage | 2023 Wage | 2024 Wage | Nominal Chg | Real Chg | 2022 Jobs | 2024 Jobs | Job Growth | Signal |
|---|---|---|---|---|---|---|---|---|---|
| COHORT A - Predicted AI-Resistant (Oxford low-risk) | |||||||||
Architecture Architects · SOC 17-1011 | $82,840 | $93,310 | $96,690 | +16.7% | +8.4% | 107,490 | 111,140 | +3.4% | ●●●● Very Strong |
HVAC HVAC Mechanics · SOC 49-9021 | $51,390 | $57,300 | $59,810 | +16.4% | +8.1% | 374,770 | 396,870 | +5.9% | ●●●● Very Strong |
Massage Therapy Massage Therapists · SOC 31-9011 | $49,860 | $55,310 | $57,950 | +16.2% | +7.9% | 86,270 | 96,040 | +11.3% | ●●●● Very Strong |
Nursing & Healthcare Registered Nurses · SOC 29-1141 | $81,220 | $86,070 | $93,600 | +15.2% | +6.9% | 3,072,700 | 3,282,010 | +6.8% | ●●●● Very Strong |
Electrical Work Electricians · SOC 47-2111 | $60,240 | $61,590 | $62,350 | +3.5% | −4.8% | 690,050 | 742,580 | +7.6% | ●●●○ Mixed |
Yoga & Fitness Fitness Trainers · SOC 39-9031 | $45,380 | $46,480 | $46,180 | +1.8% | −6.5% | 250,540 | 303,620 | +21.2% | ●●○○ Mixed |
Automotive Repair Auto Technicians · SOC 49-3023 | $46,970 | $47,770 | $49,670 | +5.7% | −2.6% | 664,070 | 688,840 | +3.7% | ●●○○ Mixed |
| COHORT B - Predicted AI-Vulnerable (Oxford/WEF high-risk) | |||||||||
Web Design Web Developers · SOC 15-1254 | $78,580 | N/A | $90,930 | +15.7% | +7.4% | 88,620 | 78,860 | −11.0% | ●●○○ Mixed |
Bookkeeping Bookkeeping Clerks · SOC 43-3031 | $45,860 | N/A | $49,210 | +7.3% | −1.0% | 1,550,750 | 1,455,770 | −6.1% | ●○○○ Weak |
Data Entry Data Entry Keyers · SOC 43-9021 | $36,190 | N/A | $39,850 | +10.1% | +1.8% | 157,380 | 135,280 | −14.0% | ●●○○ Mixed |
Video/Broadcasting Broadcast Technicians · SOC 27-4012 | $60,700 | N/A | $53,920 | −11.2% | −19.5% | 33,020 | 21,080 | −36.2% | ○○○○ Negative |
Graphic Design Graphic Designers · SOC 27-1024 | $57,990 | $58,910 | $61,300 | +5.7% | −2.6% | 211,890 | 214,260 | +1.1% | ●○○○ Weak |
Proofreading Proofreaders & Copy Markers · SOC 43-9081 | $45,410 | $48,790 | $49,210 | +8.4% | +0.1% | 5,120 | 5,160 | +0.8% | ●○○○ Remnant |
Copywriting Writers and Authors · SOC 27-3043 | $73,150 | $73,690 | $72,270 | −1.2% | −9.5% | 54,010 | 47,800 | −11.5% | ○○○○ Negative |
Source: BLS Occupational Employment and Wage Statistics (OEWS), May 2022 & May 2024 national data. Real wage = nominal minus 8.3% cumulative CPI-U inflation.
Source: Wiingy
Download TableThe Web Design anomaly: Web Developer wages grew +15.7% (beating inflation) but jobs fell -11.0% (-9,760 jobs lost). This is not a contradiction - it is a barbell effect. As AI tools automate routine website work, the total number of web developers shrinks, but those remaining are doing more complex, higher-value work. Wages rise for survivors; the total workforce contracts. The same pattern appears in broadcasting and may foreshadow graphic design's trajectory.
Verification Results
Do the predictions hold?
All three methods, side by side.
Oxford and WEF made specific predictions. We ran three independent tests. The table below puts every skill, every method, and every data signal in one place - so you can see exactly how each was verified and judge for yourself whether the predictions held. Each row is one skill. The Score column counts how many of the four signals confirm the prediction. 3 or more = confirmed.
M1
Google Trends · TRS
Post/pre-ChatGPT ratio · Jan 2020 – Mar 2026
M2
GKP · Year-on-Year
Monthly search volume · Apr 2025 vs Apr 2024
M3
BLS · Wages & Jobs
US government employment data · May 2022 – 2024
Cohort A - Predicted AI-Resilient · signals should be positive/growing
| Skill | M1 · Google Trends | M2 · GKP | M3 · BLS | Conclusion | ||
|---|---|---|---|---|---|---|
| TRS Signal | YoY Demand | Real Wage | Job Growth | Score | Holds? | |
| Architecture | Growing | −12% | +8.4% | +3.4% | 75% | ✓ Confirmed |
| HVAC | Growing | −18% | +8.1% | +5.9% | 75% | ✓ Confirmed |
| Massage Therapy | Declining | −45% | +7.9% | +11.3% | 50% | ~ Mostly Weak TRS but BLS is strong |
| Nursing & Health | Stable | −46%* | +6.9% | +6.8% | 75% | ✓ Confirmed * GKP = post-COVID normalisation, not AI |
| Electrical Work | Growing | +49% | −4.8% | +7.6% | 75% | ✓ Confirmed Wage lag normal - 52,530 new jobs added |
| Yoga & Fitness | Stable | +22% | −6.5% | +21.2% | 75% | ✓ Confirmed Wage lag = sector grew 21% in 2 years |
| Automotive | Stable | −19% | −2.6% | +3.7% | 50% | ~ Mostly Jobs still growing; wage pressure is real |
| Cooking & Culinary | Stable | +22% | N/A | N/A | 100% | ✓ Confirmed Both signals confirm resilience |
| Guitar | Stable | +22% | N/A | N/A | 100% | ✓ Confirmed Both signals confirm resilience |
| Sign Language (ASL) | Stable | +23% | N/A | N/A | 100% | ✓ Confirmed Growing demand post-ChatGPT |
| Piano | Stable | −18% | N/A | N/A | 50% | ~ Mostly TRS stable; GKP down but still mass scale |
| Pottery & Ceramics | Stable | −18% | N/A | N/A | 50% | ~ Mostly TRS stable; GKP normalising post-pandemic |
| Dance | Stable | −33% | N/A | N/A | 50% | ~ Mixed Broad GKP decline; TRS holds |
| Negotiation | Growing | −55% | N/A | N/A | 50% | ~ Split TRS very strong; GKP down = people seek it, not learn it online |
Green cell = signal confirms prediction · Red = contradicts · Score = % of available signals confirming the prediction
Cohort B - Predicted AI-Vulnerable · signals should be negative/declining
Colour logic is inverted for Cohort B. A green cell means the signal confirms vulnerability (the skill IS declining, as predicted). A red cell means the signal contradicts the prediction (e.g. wages unexpectedly rose). The Score column counts signals that confirm vulnerability.
| Skill | M1 · Google Trends | M2 · GKP | M3 · BLS | Conclusion | ||
|---|---|---|---|---|---|---|
| TRS Signal | YoY Demand | Real Wage | Job Growth | Score | Holds? | |
| Web Design | Anomalous↕ | −33% | +7.4% | −11.0% | 50% | ~ Mostly Wages up = barbell effect (see BLS section) |
| Bookkeeping | Declining | −19% | −1.0% | −6.1% | 100% | ✓ Confirmed All 4 signals confirm - 94,980 jobs gone |
| Data Entry | Anomalous↕ | −38% | +1.8% | −14.0% | 50% | ✓ Confirmed −22,100 jobs; anomalous TRS = market lag |
| Video/Broadcasting | Anomalous↕ | 0% | −19.5% | −36.2% | 50% | ✓ Confirmed BLS is definitive - 36% of jobs gone |
| Graphic Design | Declining | −33% | −2.6% | +1.1% | 75% | ✓ Confirmed Jobs flat but wages and search both declining |
| Proofreading | Declining | −55% | +0.1% | +0.8% | 50% | ✓ Confirmed 5,160 jobs nationally - occupation is a remnant |
| Copywriting | Declining | −34% | −9.5% | −11.5% | 100% | ✓ Confirmed All 4 signals confirm - clearest AI displacement |
| Transcription | Anomalous↕ | -63% | N/A | N/A | 50% | ✓ Confirmed GKP -63% is the sharpest collapse in the study |
| SEO | Anomalous↕ | -19% | N/A | N/A | 50% | ~ Mostly Declining YoY; TRS anomalous = retraining surge |
Green = signal confirms prediction (inverted for Cohort B - declining = confirms) · Amber = anomalous/explained · Score = % of available signals confirming
Languages (X)
Contested (C) - medium/disputed risk
| Skill | M1 · Google Trends | M2 · GKP | M3 · BLS | Conclusion | ||
|---|---|---|---|---|---|---|
| TRS Signal | YoY Demand | Real Wage | Job Growth | Score | Holds? | |
| Japanese Language | Growing | -21% | N/A | N/A | 50% | ~ Split TRS growing; GKP down - split signal |
| Spanish Language | Stable | -18% | N/A | N/A | 50% | ~ Mixed TRS holds; GKP declining moderately |
| French Language | Stable | -34% | N/A | N/A | 50% | ~ Mixed TRS stable; stronger GKP drop than Spanish |
| Plumbing Contested | Growing | -16% | N/A | N/A | 50% | ~ Challenges Oxford Oxford 35% risk but TRS near 2.0 - prediction may be wrong |
| Welding Contested | Growing | -19% | N/A | N/A | 50% | ~ Mixed TRS growing; GKP declining - labour market holds |
| Drawing & Illustration Contested | Stable | -18% | N/A | N/A | 50% | ~ Mixed TRS resilient despite AI image gen competition |
Score = % of available signals confirming the prediction · N/A = no BLS data for this skill
Collective Verdict
The Oxford and WEF predictions were right.
The divergence is already visible in the data.
Cohort A - AI-Resistant
Confirmed
All 3 methods agree: physical, trade, and interpersonal skills grew or held steady
Cohort B - AI-Vulnerable
Confirmed
All 3 methods agree: text, data, and routine digital skills are losing wages, jobs, and learning demand
Most Challenged Prediction
Plumbing
Oxford assigned 35% automation risk. Search demand nearly doubled. Likely belongs in Cohort A.
The Paradox
Why people are still learning
AI-threatened skills
Several Cohort B skills show growing Trends search demand despite their occupations losing jobs and wages. This is not a contradiction - it has three explanations.
Explanation 1 - Working with AI, not against it
Video Editing
TRS 2.59 - but YoY flat; job data unavailable for direct comparison
Someone learning video editing in 2025 is not competing with Runway or Sora. They are learning to direct AI video tools, understand what good editing looks like, and supervise AI output. The skill is changing, not disappearing. Search demand persists because the human oversight role persists.
Explanation 2 - The retraining signal
Transcription
TRS 1.68 (2022-2025) → but -63% YoY now. The cycle completed.
Some early-period demand in AI-threatened categories came from practitioners whose revenue collapsed, searching for retraining paths. Transcription's trajectory tells the whole story: search grew right after ChatGPT (people exploring alternatives), then collapsed once the market fully adjusted. The -63% YoY is when the lag ended.
Explanation 3 - The market lag effect
Data Entry
-38% YoY now; -14% jobs lost; -22,100 positions gone
AI tools become capable before the market fully adjusts. Job listings still appeared for data entry workers in 2023-2024 even as AI was automating the tasks, because HR workflows and hiring pipelines lag capability. The YoY collapse of -38% signals that the lag has now ended. The job market caught up with the technology.
The nuanced finding: Cohort B's higher ratio TRS (avg 1.56 vs Cohort A's 1.22) is a statistical artefact of near-zero baselines, not evidence of AI resistance. The correct metric is absolute Trends change: Cohort A +2.42 units average vs Cohort B +1.08 units. On absolute demand change, Cohort A outperforms by 2.2×. Large stable search signals maintaining their volume IS the proof of AI-resistance - not ratio effects from tiny baselines.
Methodology
Four independent data layers,
one verification study
We did not just read the reports and write an article about them. We collected real data - three independent datasets - and ran the predictions against what actually happened. Here is exactly how we measured things, so you can check our working.
01
Google Trends TRS
75 monthly data points per skill, January 2020–March 2026. All 37 skills collected in normalised batches anchored to guitar lessons (Jan-2022 = 68). Pre/post ChatGPT averages calculated for each skill.
02
GKP Year-on-Year
Google Keyword Planner exports, April 2025–March 2026 vs prior year. Monthly search volumes for primary and secondary keyword clusters per skill. Provides current momentum signal, not just historical trend.
03
BLS Wage Data
Occupational Employment and Wage Statistics, May 2022, 2023, and 2024. National cross-industry median annual wages, inflation-adjusted using BLS CPI-U. Covers 14 occupations across both cohorts.
04
BLS Employment
Total employment by occupation from same OEWS files, 2022–2024. Tracks whether jobs are being created or destroyed in each occupation. Combined with wage data to identify displacement signatures.
Primary Sources
Oxford Future of Employment (2017)
Frey & Osborne · Oxford Martin School
Automation probability estimates for 702 US occupations using machine learning classification of task susceptibility. Assigned probabilities from 0.0 (no risk) to 0.99 (near-certain automation). Our Cohort A/B classification is based directly on these scores.
WEF Future of Jobs Report (2025)
World Economic Forum · Davos
Survey of 1,000+ employers across 22 industries, 55 economies. Identifies fastest-growing and fastest-declining job roles by 2030. Box 3.1 explicitly classifies physical, interpersonal, and sensory skills as having zero GenAI substitution potential.
BLS Occupational Employment & Wage Statistics
Bureau of Labor Statistics · US Dept. of Labor
Annual survey of ~1.2 million establishments covering 800+ occupations. May 2022, May 2023, May 2024 national cross-industry files used. Primary data; no secondary aggregator or intermediary.
Google Trends & Keyword Planner
Google LLC · Data collected 2020–2026
Google Trends normalised indices (0-100) for search interest. GKP monthly search volume estimates for specific keywords. Collected in batches of 5 terms with a common anchor (guitar lessons) to enable cross-batch normalisation.
Limitations: Search demand measures learning intent, not employment outcomes directly. TRS ratio can be inflated by near-zero baselines (see Cohort B discussion). BLS OEWS data lags by ~18 months. GKP volumes are estimates with rounding. Nursing YoY decline reflects post-COVID normalisation, not AI threat. No primary interview data. These limitations are acknowledged and do not invalidate the directional findings.
The People
Wiingy Research Team
The Wiingy Research Team produces data-driven studies that uncover emerging trends at the intersection of education, technology, and the labour market.
Shifa
Lead Researcher
Shifa leads the Wiingy Research Team, overseeing study design, research scope, and final editorial review. Her work uncovers emerging trends at the intersection of education, culture, and consumer behaviour - driving data-backed decisions that shape Wiingy's future.
Bhavya
Research Analyst
Bhavya contributes analytical depth and rigour to every project. His expertise in data analysis and research methodology ensures the team's studies are transparent and impactful - keeping Wiingy ahead of industry trends in the education landscape.
About
About Wiingy
Wiingy is a top-rated tutoring marketplace that connects school students, college students, and young adults with over 4,500 expert-vetted tutors for 350+ subjects including coding, math, science, music, and more. Founded on the mission that world-class education should be accessible to everyone, Wiingy serves students across 20+ countries. Every tutor is 100% vetted, and a free trial lesson is available within 24 hours of booking.
Whether you're looking to develop a skill the data shows is genuinely AI-resilient - trades, music, languages, physical instruction - or any of 350+ other subjects, Wiingy has a tutor for you.
Top skill categories on Wiingy
- Trades & Technical - Electrical, HVAC, Plumbing, Automotive
- Music - Guitar, Piano, Singing, Violin, Music Theory
- Languages - Spanish, French, Japanese, ASL and more
- Academic - Math, Science, Coding, Test Prep
- Creative - Architecture, Drawing, Dance, Pottery
4,500+ vetted tutors · 350+ subjects · Starting at $28/hr · Free first lesson guaranteed
Start learning at WiingyContact
Website
wiingy.comResearch
wiingy.com/researchPhone
+1 618-448-5208
Address
8 The Green, Dover,
Delaware, US 19901