The Skill-Demand Source Reference, One Row per Report and What It Backs

This reference is for job seekers who keep meeting "top in-demand skills" lists and need to know whether any of them justifies adding, moving up, or cutting a skill on the resume for the roles they target. It gives one row per source type: the population it measures, the window it covers, whether it counts employer postings, worker self-reports, or hires, and the resume move it can and cannot back. Find the row for the claim you're looking at, then use the procedure to turn it into a decision.

What each skill-demand source actually counts

Skill-demand sources count four different things: employer postings, worker profile behaviour, employer opinions, and matched hiring outcomes. Only the first measures what a screener reads in a posting, so only posting-based counts should choose the keywords in your skills section.

The four kinds of basis, each defined once:

• Posting-based: counts how often a skill is written into job postings. This is the closest thing to what a screener actually checks.
• Self-report: counts what workers list about themselves on profiles. It measures supply, not demand.
• Survey: counts what employers say when asked. It measures stated preference or expectation.
• Outcome: counts what happened to people whose resumes listed a skill. It measures correlation with getting hired.

The pages that rank for "skills to put on a resume" mostly blend all four into one top-ten list. A list like that can't tell you which skill belongs on your resume, because its items were never measured in the same unit.

Posting-based rows: O*NET flags, job-board badges, and your own tally

Your own tally and ONET's In Demand flag are the only sources here that count skills written into employer postings, and the In Demand flag is the only one cut to a single occupation. ONET's Hot Technologies flag is economy-wide, and I couldn't verify any published method behind a job-board skill badge.

O*NET In Demand (per occupation)

ONET defines In Demand skills as requirements "frequently included in the employer job postings for a particular occupation." The threshold is fixed: a skill is In Demand for an occupation if it appears in 5% or more of that occupation's employer postings. The list is built from unique US nationwide employer postings in a designated 12-month period, which ONET chose to minimise seasonal variation. An ONET-SOC code is the occupation code ONET uses to organise all its data, and the flag only means something against one code.

• Population: US employer postings, all seniorities, one occupation.
• Window: one designated 12-month period per release. I couldn't find a published refresh cadence, but successive releases re-run the designations.
• Basis: postings.
• Proves: a tool shows up in a meaningful minority of the occupation's postings nationally.
• Lies when: you read it as local, or as required. Five percent means up to 95% of postings may leave the skill out.
• Can back: moving up a tool that your own tally also shows.
• Cannot back: adding a tool your target postings don't mention.

The C language shows why the occupation code matters. C is In Demand for Software Developers at 6% of postings. For Data Scientists it sits at 3%, below the threshold, so it isn't flagged. Same skill, two different answers.

Coverage has also changed. Release 30.3 lists 2,400+ In Demand skill linkages across 285 occupations. Earlier releases spread a similar number of links across 495 occupations, so each flagged list got denser and many occupations now carry no In Demand flags at all. O*NET hasn't said why. If your occupation has no flags, that tells you nothing about whether its postings value tools. It just means you need your own tally. The In Demand column was added in release 27.1.

O*NET Hot Technologies (economy-wide)

Hot Technologies are the "software and technology requirements most frequently included across all employer job postings." Release 30.3 lists 176 employer-based Hot Technologies, which produce 11,500+ occupation linkages. A tool can be Hot across the economy without mattering to your occupation at all.

• Population: US employer postings across all occupations.
• Basis: postings, pooled economy-wide.
• Proves: a tool is common somewhere in the labour market.
• Lies when: your SOC code is a catch-all. On one example occupation page, the flagged list runs from SAP software to Autodesk AutoCAD, and no single job needs both.
• Can back: a pointer to which tools to look for in your own postings.
• Cannot back: any resume move on its own.

The 30.3 Software Skills file has 31,821 rows, against 32,773 in the 30.2 Technology Skills file. If an old tutorial sends you looking for "Technology Skills", the renamed file is the one you want.

Job-board skill badge

A job-board skill badge is a label a job site attaches to a posting or a candidate to suggest that a skill matches or is in demand. I found no published method for how these badges are computed, so their population, window, and basis aren't publicly established.

• Proves: nothing on its own until you check it.
• Lies when: you treat it as an employer requirement.
• Can back: a prompt to re-read the posting. Find the exact term in the posting text. If it's there, your tally already counts it. If it isn't, the badge has no standing.
• Cannot back: any add, move up, or cut.

Your own posting-frequency count

Your own tally counts how many of your target postings mention each term at least once, written as "k of N postings." This is document frequency. It differs from term repetition. One practitioner guide says terms that appear "3 or more times are almost certainly core requirements," but that counts repeats inside a single posting. It helps you read one posting. It's the wrong measure for comparing across postings.

• Population: the postings you will actually apply to, at your seniority, in your market.
• Window: the weeks you spend collecting.
• Basis: postings.
• Proves: what your actual market writes down.
• Lies when: the sample is small, one employer dominates, or recruiter reposts count the same role twice.
• Can back: add, move up, and cut.
• Cannot back: anything about a market or level you didn't sample.

Published advice on sample size disagrees: 3 to 5 job descriptions, 5 to 10 listings of similar seniority and company size, or at least 6. Nobody has published a statistically justified minimum. One read of three live US postings found between 10 hard skills (a sales role) and 17 (a finance role and a marketing role) per posting. With a sample that small, three postings produce dozens of terms and one employer's stack can swamp the counts. I use at least 10 postings across at least 5 employers.

Profile and growth rows: rising-skills lists and profile skill stock

Rising-skills lists and profile skill counts measure what workers do with their own profiles, not what employers write into postings. They're supply-side signals: useful for seeing where a field is moving, weak as evidence for any single keyword.

Professional-network rising-skills list (LinkedIn Skills on the Rise)

The current edition of LinkedIn's Skills on the Rise ranks year-over-year growth in two areas: skill acquisition (skills added to member profiles) and hiring success (skills held by members who were hired over the past year). The window runs from December 1 to November 30 and is compared with the same period a year earlier. Language skills, basic digital literacy skills, and overly broad skills are excluded. The list covers 12 markets, and a skill needs "sufficient representation and activity volume" to be ranked at all.

The previous edition worked differently. It was a single list of the 15 fastest-growing skills in the US, and it had a third pillar, emerging demand, which measured "the increased presence of a given skill in paid job postings." The current edition dropped that pillar. Ranking pages still describe the list as employer demand.

• Population: members in 12 markets.
• Window: 12 months against the prior 12.
• Basis: profile additions plus hires.
• Proves: a skill is growing fast among members, and growing among people who got hired.
• Lies when: you read growth as volume. A small skill that doubles can outrank a huge skill that grows slowly.
• Can back: breaking a tie between two skills you can prove, or setting a retraining direction.
• Cannot back: adding a keyword.

Profile skill stock (Refolk's index)

Profile skill stock is the number of workers who list a skill on their own profiles. In Refolk's index of professional profiles, US profiles titled Data Analyst list Tableau far more often than Power BI, and the US Tableau cohort is several times the UK one.

• Population: workers' self-reported profile skills, by title and country.
• Basis: self-report.
• Proves: how crowded a skill claim is among the people you compete with.
• Lies when: you read it as demand. Profiles pile up skills over years, so they may lag current tooling.
• Can back: judging whether a skill will set you apart or just match the crowd.
• Cannot back: whether employers want it.

There's a tension here I can't resolve with public data. In one resume-outcome study, Tableau and Power BI appear on nearly equal shares of resumes (3.2% and 3.6%), while profile stock differs 10.6x. My hypothesis is that profile lists lag what people currently put on resumes. That isn't established, so check it against your own posting tally.

Survey and outcome rows: employer outlooks, graduate surveys, resume-outcome studies

Employer surveys report what employers say they want or expect. Resume-outcome studies report what happened to people who listed a skill. Neither counts posting text, so neither should pick a keyword, but outcome data is the best available evidence for trimming generic labels.

Global employer outlook (WEF Future of Jobs)

The World Economic Forum's Future of Jobs report surveys 1,043 major employers representing more than 14.1 million workers, across 22 industry clusters and 55 economies, about the next five years. Its skill rankings are employer expectations of future change, not counts of postings. Its headline findings: just under 40% of workers' core skills are expected to change over the period, and 63% of employers named skill gaps as a major barrier.

• Population: large employers, globally.
• Window: a five-year projection made at survey time.
• Basis: survey.
• Proves: which broad skill families employers expect to matter more.
• Lies when: you copy its labels, such as "analytical thinking", into a skills section. They're families, not posting text.
• Can back: a retraining direction.
• Cannot back: keyword wording, unless a target posting uses the same phrase.

Graduate-employer survey (NACE Job Outlook)

NACE's Job Outlook reports the share of responding employers who say they look for an attribute on college graduates' resumes. One edition had 237 respondents: 162 NACE employer members plus 75 nonmembers. A spring update drew 142 member respondents, 19.9% of eligible members. Its total respondent count isn't verified. Data collection for one edition ran for about six weeks.

• Population: employers who chose to respond, skewed toward NACE members.
• Window: the survey weeks.
• Basis: survey.
• Proves: what hiring employers say they value in entry-level candidates.
• Lies when: you read a percentage as posting demand. More than half of respondents cited 10 or more skills, so the percentages measure how broad the wish lists are, not what employers screen on.
• Can back: an emphasis or retraining direction for entry-level candidates.
• Cannot back: keyword choice.

The spring update PDF appears to show 35% of entry-level jobs requiring AI skills against 16.5% of entry-level job descriptions including them. The PDF layout is garbled, so treat that pairing as unverified until you check it in the document itself. If it holds, it shows exactly how far stated requirements can drift from what's written into postings.

Resume-outcome study (Careerminds)

The Careerminds study matched a sample of 58,433 resumes against a LinkedIn-derived employment dataset. 9,700 of them had enough follow-up history to observe an outcome, and of those, 15.8% led to a new job, roughly one in six. Skill hire rates are compared against that baseline.

How common a skill is on resumes and how well it pays off run in opposite directions. Communication is the most-listed skill here and sits below baseline. Tableau is rare and sits highest. The likely mechanism is that common labels don't separate candidates, while specific tools signal fit for a particular occupation. The study also reports that holders of technical or analytical skills were 42% more likely to land a job than those listing only soft skills (20.1% vs 14.2%).

• Population: resumes with observable follow-up, occupation mix unpublished.
• Window: not specified.
• Basis: matched outcomes.
• Proves: a correlation between listing a skill and getting hired.
• Lies when: you read it as cause. Occupation mix, seniority, and selection into observable follow-up aren't reported as controlled.
• Can back: trimming generic soft-skill labels your tally doesn't use.
• Cannot back: adding a tool because its hire rate is high.

Which resume move each source can back

Adding or moving up a skill needs posting evidence plus proof you hold the skill. Cutting a generic label can lean on outcome data, as long as your tally agrees. Growth lists and surveys only break ties or set a retraining direction.

The matrix below is the decision once you've done the tally. The horizontal axis is how many of your postings mention the skill. The vertical axis is whether you can prove it with a project, a role, or a result.

How to check if a skill is in demand for your target role

To check whether a skill is in demand for your role, map the role to one occupation code, read its O*NET flags, tally at least 10 live postings, and only then consult growth and outcome sources. The full run takes about three hours, and the output is a skills section you can defend term by term.

Step 4 is where most people stall, because an hour of copying posting text into a sheet feels like busywork. The tally is what makes every later step safe. Use this sheet so the counts and the evidence sit in one place.

Target: [title] | [seniority] | [market] | SOC [code]
Postings: N = [count] | Employers = [count] | Reposts dropped = [count]

Term | k of N postings | O*NET In Demand for this SOC? | Proof (project, role, result) | Move
SQL | _ of _ | yes / no | [bullet or link] | add / move up / cut / hold
Tableau | _ of _ | yes / no | [bullet or link] | add / move up / cut / hold
Communication | _ of _ (exact phrase only) | n/a | [bullet] | keep only if a posting uses it

Once the tally exists, write a one-line source note for each change so you can re-check it later.

Skill: [term] | Move: add / move up / cut
Primary evidence: [k] of [N] target postings, collected [week]
Supporting: O*NET In Demand for SOC [code], release [number]: yes / no
Not used as evidence: rising-skills rank, employer survey percentage
Proof on resume: [which experience bullet demonstrates it]

After the base skills section is set, each application still needs its own cut, because one posting's terms differ from the tally's average. Refolk tailors the resume to every posting you apply to and scores how well you actually fit it, so the per-posting version of this tally happens without retyping the posting text.

How skill-demand sources mislead a resume

Every source in this reference misleads in a predictable way, and almost every failure comes from reading one unit as another: growth as volume, opinion as requirement, correlation as cause, or a national figure as local. Here are the eight failure modes, each with what it looks like and the check that catches it.

1. An O*NET flag read as a local requirement. The data is national and covers all seniorities, so a flagged tool can be missing from your city's or level's postings, and the 5% threshold means up to 95% of postings may leave it out. Check: compare against your own tally.
2. Hot Technology confused with In Demand. Hot is economy-wide, and catch-all SOC codes produce flagged lists running from SAP software to Autodesk AutoCAD. Check: use only the In Demand column for your own SOC code.
3. Skills on the Rise read as volume. The list measures growth, so a small skill can rank high, and the current edition contains no posting data. Check: confirm a posting count before adding the skill.
4. A WEF ranking used as a resume keyword. It's a five-year employer projection of broad skill families, not posting text. Check: does any of your target postings use that phrase?
5. NACE percentages read as posting demand. A high share of employers saying they look for teamwork doesn't make teamwork a screening term, especially when more than half of respondents tick 10 or more skills. The possible 35% vs 16.5% AI gap, still unverified, points the same way. Check: tally the postings.
6. A resume-outcome rate read as causal. Tableau holders may cluster in hiring-heavy occupations, and only about 16.6% of matched resumes (9,700 of 58,433) had observable outcomes. Check: compare within your occupation. The study doesn't publish that breakdown, so treat any gap as a lead, not a verdict.
7. A self-tally with too few postings. With three postings, one employer's stack dominates the counts. Check: use at least 10 postings across at least 5 employers, and drop recruiter reposts of the same role.
8. Profile stock read as demand. The 10.6x Tableau to Power BI ratio among US Data Analysts in Refolk's index reflects what workers list, which may lag current tooling. Check: compare against posting counts.

Two of these cause most of the damage. Failure 3 puts unproven, trend-driven skills on resumes, which interviewers then probe. Failure 6 makes people add a tool because its hire rate looks high, when the tool may just mark a different occupation from theirs. Both are caught by the same check: a k of N count from your own postings.

Keeping the skills section current

The skills section goes stale when your target changes or when a source changes its method, so re-check the method, not just the ranking. Each source here changes in a way you can see if you know where to look.

• O*NET: read the release notes for each new release. Release 30.3 renamed the skills file and its columns, and it stopped updating the "Tools Used" and "Work Values" files. Confirm the In Demand threshold and the 12-month posting window still apply, and re-pull your SOC code's flags.
• Rising-skills lists: read the methodology note in each edition before reading the ranking. One edition change already removed posting demand entirely, and another could add it back.
• Employer surveys: check the respondent count and how it splits between members and nonmembers. A smaller or more member-heavy pool changes what the percentages mean.
• Outcome studies: check whether a new study publishes results by occupation. That's the breakdown that would turn a correlation into usable evidence for your role.
• Your tally: redo it whenever your title, seniority, or market changes, and refresh it if your applications stop converting. Postings shift faster than any annual report.

Refolk's fit score per posting is a quick early warning here. If your scores drop across a batch of similar postings, the market's terms have moved away from your base skills section, and it's time to re-tally.