The Resume-Skill Demand Signal Reference, One Row per Signal

You have a resume skill and a decision: feature it, keep it quietly, or retrain out of it. The "top skills employers want" articles will not settle it, because they contradict each other, they expire, and they never tell you how to read a skill they did not happen to list. This reference names each demand signal one at a time, states what it proves, states how it lies, and points to the public source that publishes it, so you can evaluate any skill on your own resume instead of memorizing this year's ranking.

It is for job seekers deciding what to put forward, and for anyone weighing a retrain or a move. Read it end to end once, then jump to the row you need.

Why one ranked list can never answer "are my resume skills still in demand"

The short answer: every "most in-demand skills" list is built from one source measuring one thing over one horizon, so two honest lists disagree without either being wrong. A skill can top a survey of what employers expect by 2030 and be flat in today's postings at the same time.

The three big public sources literally measure different quantities. The World Economic Forum's Future of Jobs Report surveys employers about where skills are heading; it drew on over 1,000 leading global employers representing more than 14 million workers across 22 industry clusters and 55 economies. The Bureau of Labor Statistics models numeric employment change over a ten-year window. Indeed's Hiring Lab counts live postings right now. Ask them the same question and you get three answers because they answer three different questions.

The stakes are not academic. WEF projects that 39% of existing skill sets will be transformed or become outdated between 2025 and 2030. That is why the durable move is to read signals, not rankings. A ranking tells you where a skill sat when the article was written. A signal tells you how to check today, and again next quarter.

A ranking tells you where a skill sat last year. A signal tells you how to check it yourself, today and next quarter.

The seven demand signals, and what each one proves

There are seven observable signals that recur across the primary sources. Each proves one narrow thing and lies in one predictable way. Learn the row, not the ranking.

Two of these are the ones people misread most, so hold them together. Posting frequency without velocity is the classic trap: business operations skills appeared in nearly three-quarters of all US job postings in one recent quarter, more than any other category. That does not make "business operations" a differentiator on your resume. It makes it a baseline the employer already assumes. Frequency proves the skill is expected. Velocity proves whether demand for it is still growing.

The public sources and the exact unit each publishes

Before you trust a number, know its unit. Each source reports in its own currency, and mixing them is how people build contradictory conclusions.

A few grain notes that save you from misreading. O*NET assesses all 35 skills for each of 968 occupations, with importance rated 1 to 5 where 1 means not important and 5 means extremely important; these are analyst ratings, not counts of jobs. Indeed's index is anchored so that February 1, 2020 equals 100, which keeps comparisons consistent across time and countries. When you read an Indeed figure of 112, that means 12% above the pre-pandemic baseline, not 112 jobs.

Refolk works in a different currency again: supply. The tables below use counts from Refolk's index of professional profiles, which tells you how many people already hold a skill, a number the posting sources cannot give you.

Supply is a signal too, and scarcity only pays when paired with velocity

Supply tells you how crowded the field of people holding a skill already is. On its own it proves nothing about demand, but paired with rising postings it turns a scarce skill into a genuine advantage.

Here is the supply picture directly from Refolk's index.

Read this correctly. The US Python pool is about 7.4 times the German Python pool, and Rust is roughly 0.58% the size of the US Python pool. That scarcity is real, but scarcity is a supply fact, not a demand signal. It becomes a demand signal only when posting velocity confirms employers actually want it. The confirming layer exists here: since late one recent February, software developer postings in the US rose almost 15% while postings overall declined 7%. Scarce supply plus rising postings is the combination that earns interviews. Scarce supply with flat postings is just a small field.

You can check the supply side of any skill directly against Refolk's index.

Velocity and level diverge by design, so read them together

Velocity and level pull in opposite directions across sectors, and reading one without the other picks the wrong winner. High growth almost always sits on a low base; high level almost always grows slowly.

The clearest illustration is AI adoption across sectors, where the growth rates and the current levels tell nearly opposite stories.

Education shows 200% generative-AI growth, but off the lowest base of any sector, so the absolute number of jobs is small. Human resources shows 66% growth. Data and analytics is a different regime entirely: nearly 45% of data and analytics postings now contain AI-related terms, compared with about 15% in marketing and 9% in human resources. So AI in data and analytics is already table stakes at a high level, while AI in education is a fast-rising sliver. If you list an AI skill, which sector you are applying into decides whether it reads as expected or as a rare edge.

Automation exposure is a fork, not a death sentence

Exposure tells you a skill's tasks are technically feasible to automate. It does not tell you the job goes away. Both Pew and the Treasury emphasize that these scores measure potential technical exposure, not actual displacement; exposure identifies feasibility, it does not predict whether jobs will be eliminated or transformed.

That distinction changes the verdict on the same score. In 2022, 19% of American workers were in jobs most exposed to AI, and those jobs cluster in white-collar, higher-education roles, while clerical jobs built on routine cognitive tasks are projected to keep shrinking. So the identical "high exposure" label means augmentation for an engineer and substitution for a data-entry clerk. WEF's own breakdown matches this: the fastest-growing roles are technology-related, including big data specialists, fintech engineers, and AI and machine learning specialists, while the largest absolute declines fall on clerical and secretarial workers, cashiers, postal service clerks, bank tellers, and data entry clerks.

BLS adds a timing caution worth internalizing before you retrain in a panic. The historical record shows technology impacts occupations gradually, not suddenly, and it can take time for employers to figure out how to incorporate new technology. Overall, BLS projects total employment to grow from 170.0 million in 2024 to 175.2 million in 2034, an increase of 3.1%. Exposure is a reason to check whether your skill is on the augmented side or the substituted side, not a reason to abandon it overnight.

Time-to-fill and geography, the two signals people skip

Time-to-fill measures how long a role sits open, and a long fill relative to the baseline usually means the underlying skill is scarce. But it also breaks in a specific way, so anchor it against a benchmark before you infer demand.

Average time to fill fell from 48 days in 2023 to 41 days in 2024, so the baseline itself moves; re-check it rather than memorizing a number. Tech roles take an average of 62 days to fill, nearly three weeks longer than the all-industry average, which points to genuine scarcity. But a long fill can also mean a broken hiring process, so compare against the 44-day baseline before you read scarcity into it.

Geography can flip a national signal on its head. A skill that looks in demand nationally can be oversupplied exactly where its workers cluster. Indeed's index has shown large tech metros running near or below their pre-pandemic baseline while mid-size metros averaged 112.0 and small metros 115.5. If your scarce skill is concentrated in the same handful of metros as everyone else who holds it, the national number is hiding your real local market. That is when featuring the skill may quietly imply relocation.

How to triage one resume skill

The job is to run each skill through the signals in order and end with a keep, drop, or retrain verdict. Budget one to two hours the first time; it goes faster once the sources are bookmarked.

The figure below shows the same flow as a pipeline, so you can see where a skill can drop out.

Once you have verdicts, the tailoring problem is real: the skills you feature should shift per posting to match what that employer actually names. This is exactly where Refolk removes friction, tailoring the resume to each posting from your own history and scoring how well you fit, so the featured skills track the signal you just read rather than a static master file.

Here is a compact rubric you can copy into your notes and apply to each skill.

Skill: ____________________  Occupation/category: ____________________

Posting frequency vs baseline:  above (2) / at (1) / below (0)
Year-over-year velocity:        rising (2) / flat (1) / declining (0)
Salary premium (and # of salaried postings): clear + N>=50 (2) / weak or N<50 (1) / none (0)
Survey projection (WEF direction): rising (2) / mixed (1) / declining (0)
Automation exposure:            augmented (2) / neutral (1) / substituted (0)
Geographic concentration:       national (2) / few metros, I am in one (1) / few metros, I am not (0)

Verdict rule:
- FEATURE if velocity=2 AND (frequency>=1 OR premium=2) AND exposure>=1
- KEEP    if frequency=2 but velocity<=1 (baseline, do not lead with it)
- RETRAIN if velocity=0 AND exposure=0

How this goes wrong: the eight failure modes

Most bad calls come from reading one signal in isolation or trusting a number that rests on thin data. These are the specific ways the signals lie.

1. Mistaking ubiquity for demand. A skill in about 75% of postings, like business operations, is table stakes, not a strength. Featuring it as a differentiator is the false positive. Check whether its share is rising or flat before you lead with it.
2. Velocity off a tiny base. The 200% generative-AI growth in education is a large percentage of almost nothing. Always pull absolute posting counts alongside the growth rate.
3. Salary premium from thin data. With only about 23% of postings advertising pay in some markets, a premium can be a sampling artifact. Check the count of salaried postings, not just the delta.
4. Reading exposure as demand. High AI exposure does not mean the skill is dying; exposure identifies feasibility, not whether jobs are eliminated or transformed. It may mean augmentation.
5. Treating a survey/live mismatch as an error. WEF's 2030 projections and today's postings measure different horizons. A skill rising in WEF can be flat in postings now. That is a horizon gap, not a contradiction.
6. Taxonomy drift mistaken for a trend. Vendors change how they extract skills, and Indeed cautions that its taxonomy evolution may contribute to differences in results over time alongside genuine labor market trends. A jump in share can be a measurement change.
7. National numbers hiding geographic concentration. A skill can be in demand only in a few metros. When large tech metros run near baseline while mid-size metros run at 112 and small metros at 115, the national average hides where your competition actually is.
8. Confusing time-to-fill with prestige. A long fill can mean a scarce skill or a broken process. Check against the 44-day baseline before inferring demand from a slow hire.

One more honest limit. There is no published universal cutoff for calling a skill "in demand." Practitioners read a skill's share of advertised roles relative to a category baseline, because baselines differ wildly by occupation and by vendor. A fixed threshold like "X% means in demand" is not established publicly and would break the moment you moved between occupations. When you want a number, always state the baseline it is measured against.

Before you finalize which skills to feature

Run this check across your whole skill list, not skill by skill, so the resume reads as a coherent set rather than a pile of individually defensible entries.

Keeping the read current

Demand signals move, so this is a re-check habit, not a one-time audit. Re-run the triage on your two or three lead skills each quarter, and any time you consider a new target role or metro. The mechanisms are stable even when the numbers are not: level versus velocity, premium against thin data, exposure as a fork, and geography inverting the national picture. Bookmark the primary sources and read them in their own units rather than through a listicle that repackages them. When a new ranked list crosses your feed, do not memorize it - use it as a prompt to check the one skill you actually care about against the signal that proves what you need to know.