# The Demand-Sourced Resume, From Posting Frequency to a Finished Draft

*You can turn demand data for one target role into a finished resume layer where every featured skill is backed by a posting-frequency signal and evidence you can defend.*

- Canonical URL: https://www.refolk.ai/candidates/guides/demand-sourced-resume-skills
- Pillar: Reading the market
- Format: Playbook
- Published: 2026-10-01
- Last reviewed: 2026-10-01
- Reading time: 15 min
- Keywords: how to find in-demand skills for my resume, build resume from job posting data, most requested skills in job postings, data-driven resume skills section, which skills to put on resume by demand

## Key takeaways

- Business operations skills appeared in more than 70% of all US job postings at the end of 2025, while communication and technology skills each appeared in less than half, so a near-universal skill is one covering a majority to supermajority of postings.
- In Refolk's index, US Data Analysts list SQL on 12,939 profiles but Python on only 2,140 - Python is held by 16.5% of the SQL population, making it a genuine differentiator while SQL is table stakes.
- Penetration and growth measure different things: AI literacy grew more than 70% year over year yet still appeared in only about 1 in 500 postings, so featuring it as core work signals you misread the market.
- Up to 80% of raw collected postings are duplicates, so deduplication on a 60-day title-plus-company-plus-location window is load-bearing for honesty, not just tidiness.
- 59% of employers want skills demonstrated with concrete examples rather than merely listed, and 57% have caught someone embellishing a skill set, so split demanded skills into bullet-with-metric versus keyword-only.
- Posting data refreshes monthly and tracked skill lists shift fortnightly while BLS projections are annual, so refresh the demand layer quarterly over an annual structural anchor.

This guide rebuilds your resume's skills, keywords, and bullets from what employers are actually posting for one target role, not from what you guess matters. It is for a job seeker aiming at a specific occupation who wants a repeatable method, in order, from aggregated posting demand to a finished resume layer. By the end you will have a process that turns demand data into a draft where every featured skill is backed by both a posting-frequency signal and evidence you can defend in an interview.

Most pages on this topic do one of two things: funnel you toward a vendor tool, or hand you a generic "top skills" list that tells you what is hot but not how to get it onto your own page defensibly. Neither gives you a neutral, ordered procedure with keep-or-cut thresholds and placement rules. That is what this is.

## What "demand-sourced" means, and why guessing fails

A demand-sourced resume features skills because they appear in postings at a measurable rate, not because they feel important. The core move is to replace intuition with two numbers per skill: how common it is (penetration) and whether it is rising (growth).

Guessing fails because real demand stratifies more sharply than most people expect. Within a single role and one market, some skills are near-universal and some are rare, and the gap is large. In Refolk's index, US Data Analysts list SQL on 12,939 profiles but Python on only 2,140. Python is held by 16.5% of the SQL population. If you treat both as equally central, you both bury the skill that every screen expects and overstate one that differentiates you only if you can prove it.

**12,939 - US Data Analyst profiles listing SQL in Refolk's index**

Against 2,140 listing Python, which makes SQL table stakes and Python a differentiator held by only 16.5% of the SQL population.

The method also forces a distinction that intuition blurs. Penetration and growth measure different things. Business operations skills were required in more than 70% of all US job postings at the end of 2025 yet barely grew, while AI literacy grew more than 70% year over year but still appeared in only about 1 in 500 postings. A resume that leads with a fast-but-rare skill as if it were core signals that you misread the market. A resume that ignores a near-universal skill fails the first filter. You need both numbers before you decide anything.

## Which demand signals to pull, and in what order

Pull three tiers of data in a fixed order: the annual structural baseline first, then real-time posting demand, then your own hand-coded sample if you need it. Leading with the structural sources anchors growth and stops you chasing a spiking-but-rare skill.

The three tiers each answer a different question, and they have different refresh rates and blind spots.

| Tier | What it gives you | Cadence | Blind spot |
| --- | --- | --- | --- |
| BLS Employment Projections | Skill-importance scores, 10-year growth | Annual | Importance, not posting counts; no tool names |
| BLS OEWS | Wages and employment, ~830 occupations | Annual (May reference) | No skills data at all |
| Real-time posting aggregates | Live keyword and tool frequency | Monthly/fortnightly | Scraped; needs benchmarking against official stats |

The Employment Projections skills data were released alongside the 2024-34 projections and cover 832 detailed occupations, structured into skills categories derived from O*NET. Treat these scores as importance ratings, not posting frequency. They tell you a skill matters to the occupation; they do not tell you how often it shows up in a live ad. OEWS covers roughly 830 occupations with a May reference date but carries no skills at all, so it anchors wage and growth, nothing else.

Real-time posting aggregates fill the gap the official sources leave: specific tool names and live keyword demand, where postings are scraped, deduplicated, and added daily. That currency is also their weakness. A single vendor's scraped list can diverge from official labor indicators, so anything only one source calls "hot" needs a second look.

> **Rule:** Anchor before you chase
>
> Pull the annual baseline before the real-time feed. Growth off a tiny base is seductive, and leading with posting data makes a rare, spiking skill look central before you have the penetration number to contradict it.

If you have no access to an aggregator, you can reproduce the real-time tier yourself. Hand-code 20 to 30 live postings for your exact role and market: paste each into a sheet, tally which skills appear, divide by the posting count. It is coarser than a vendor index but it is honest, and it is yours.

## Deduplicate before you count anything

Collapse repeat listings before you compute a single frequency, because duplicate inflation is the fastest way to make a niche skill look mandatory. The documented standard is a 60-day original-versus-duplicate window on normalized title, company, and location.

Here is the mechanism. If a Marketing Specialist job at one company is first posted on March 1, that is the original posting, and for the next 60 days any copies found elsewhere are treated as duplicates. If the same employer reposts the identical ad every day for a year across different boards, a naive count treats it as hundreds of jobs; the collapse treats it as a handful. This matters because up to 80% of raw collected postings are removed as duplicates. Skip the step and a single reposting employer can single-handedly make a skill look universal.

#### From raw postings to countable demand

| Stage | Figure | Note |
| --- | --- | --- |
| Raw collected postings | 100% | Everything scraped or pasted in |
| After 60-day dedup | ~20% | Originals only, repeats removed |
| Role-and-market matched | subset | Your exact occupation and location |
| Skill-tagged | per skill | Each original coded for the skills it names |

*Up to 80% of collected postings are duplicates, so counting before you collapse inflates every frequency.*

Deduplication is load-bearing for honesty, not just tidiness. The whole method depends on frequencies you can trust, and a single inflated skill pushes you to over-claim it on your page. When you count, check whether one firm drives a skill's apparent demand. If it does, discount it.

## Classify each skill: near-universal or differentiating

Tag every skill as either near-universal (table stakes you must show) or differentiating (what separates you), using penetration as the primary test and growth as a secondary one. A near-universal skill appears in a majority to supermajority of postings; a differentiator shows rapid growth off a small base.

The external benchmarks make the thresholds concrete. Business operations skills sat above 70% of US postings. By contrast, HR skills appeared in 27.3% of postings and AI literacy in about 0.2%, the latter despite more than 70% year-over-year growth. A skill above roughly 50% penetration is baseline; a skill with strong growth but low penetration is a differentiator you feature carefully, with proof.

| Skill/category | Posting penetration | Growth signal | Read |
| --- | --- | --- | --- |
| Business operations | Above 70% of US postings | Flat | Near-universal, must appear |
| HR skills | 27.3% of postings | n/a | Role-dependent baseline |
| AI literacy | ~1 in 500 (~0.2%) | +70% YoY | Differentiator, evidence it |

Your own index data does the same job inside a single role. In Refolk's index of US Data Analyst profiles, SQL is the baseline every screen expects and Python is the genuine differentiator.

| Skill | US Data Analyst profiles | Share of SQL population | Classification |
| --- | --- | --- | --- |
| SQL | 12,939 | 100% (baseline) | Near-universal |
| Tableau | 7,972 | 61.6% | Common |
| Python | 2,140 | 16.5% | Differentiating |

Geography changes the baseline, not the category. The same title and skill, measured across two countries, shows how market size inflates raw counts.

| Country | Data Analyst profiles listing SQL | Multiple vs smaller market |
| --- | --- | --- |
| United States | 12,939 | 3.6x |
| United Kingdom | 3,588 | 1.0x (base) |

SQL appears 3.6 times more often in the US index than the UK index. That does not mean SQL is 3.6 times more important in the US; it means the US market is larger. Always normalize to share-of-role, not absolute count, or a US-tuned layer will over-weight everything when you aim it at a smaller market.

> Penetration tells you what every screen expects; growth tells you what is coming. Confuse them and you lead with the wrong skill.

## Assign placement: bullet or keyword

Decide, for each demanded skill, whether it becomes a resume bullet or a skills-block keyword, and let evidence make the call. A demanded skill you can back with a quantified outcome becomes a bullet; a demanded skill you hold but cannot yet evidence goes in the skills block as a keyword for ATS matching.

This rule exists because skills-based hiring is now majority practice and employers want proof. NACE's Job Outlook 2026 found 70% of employers use skills-based hiring, up from 65% a year earlier, and one 2025 report puts adoption at 85%. Decisively, 59% of employers want to see skills demonstrated with concrete examples on a resume rather than merely listed. A bare keyword list reads as unevidenced; a bullet with a number reads as proof.

#### Placement by demand and evidence

Horizontal axis runs from Low demand to High demand. Vertical axis runs from Weak evidence to Strong evidence.

| Quadrant | What it means |
| --- | --- |
| Weak evidence, low demand | Cut it; it earns no space |
| Weak evidence, high demand | Keyword only; name it, do not claim an outcome |
| Strong evidence, low demand | One supporting bullet if space allows |
| Strong evidence, high demand | Lead bullet with a metric |

*Where a skill sits on demand and on your ability to prove it decides whether it earns a bullet, a keyword, or nothing.*

The high-demand, strong-evidence quadrant is where your featured bullets come from. The high-demand, weak-evidence quadrant is exactly where ATS keyword-stuffing tempts people, and it is a trap: the keyword that gets you past the filter is the precise claim tested next. Name it as a keyword, speak to it honestly, but do not build a bullet on ground you cannot hold.

**Demand-sourced bullet skeleton**

```
[Demanded skill], used to [action] [object], producing [quantified outcome] over [timeframe].

Example:
SQL and Tableau, used to rebuild the weekly revenue dashboard, cutting report turnaround from 3 days to same-day across 6 teams.
```

*Replace each slot with your own numbers. Keep the demanded skill first so a skills-first screen credits it immediately.*

[Refolk](/candidates) writes your resume from your own history and then tailors it to each posting you apply to, which is where this placement decision becomes repeatable rather than a one-off sort. Once you have classified your skills, the tailoring step reorders bullets and keywords to the posting in front of you without you re-running the whole method each time.

## The procedure, start to finish

Run these eight steps in order for one target role. The total is roughly a day of focused work the first time, and far less on each quarterly refresh. The one order choice that matters is pulling the structural baseline before the real-time feed, so growth anchors against importance and you do not chase a rare skill.

#### Demand to draft, in order

1. **Define the target role and market** - Fix one SOC-level occupation and one location filter you hold constant across every later step. Done when you have a single role and market you will not change mid-process.
2. **Pull the structural baseline** - Look up the role in BLS Employment Projections skills tables and OEWS for importance, wage, and projected growth. Done when you have the category importance profile plus a growth number, treating these as importance not posting counts.
3. **Pull real-time posting demand** - Gather skill frequencies from an aggregator or hand-code 20 to 30 live postings for the exact role. Done when each candidate skill carries a share-of-postings figure.
4. **Deduplicate and normalize** - Collapse repeats using a 60-day window on title, company, and location before counting. Done when no single employer's reposted ad inflates a skill's frequency.
5. **Classify near-universal vs differentiating** - Tag each skill baseline (above roughly 50 to 70% penetration) or differentiator (growth off a low base). Done when every skill carries one tag.
6. **Assign bullet vs keyword** - Choose a bullet where you have a quantified example, a keyword where you hold the skill but cannot evidence it. Done when each featured skill maps to one line or one keyword slot.
7. **Run the evidence and over-claim check** - Confirm every bullet survives a behavioral probe or timed task; downgrade or cut what you cannot defend. Done when no claim leans on a hedge like "familiar with".
8. **Set the refresh schedule** - Calendar a quarterly posting-demand re-run and an annual re-anchor to projections and wages. Done when both recurring reminders exist.

Refolk scores how well you actually fit a posting, which is a fast way to see whether your newly placed skills register the way you intended before you commit the layer to every application.

## How this goes wrong: failure modes and false positives

The method breaks in predictable ways, and each failure has a cheap check. This section is the most valuable part of the standard, because a resume built on a flawed count is worse than one built on honest intuition.

- **Treating importance scores as posting frequency.** A skill scores high in the projections data but barely appears in live ads. The projections measure importance, not frequency. Check: cross-reference against an actual posting count before you feature it.
- **Duplicate inflation.** A single employer reposting across boards makes a niche skill look universal. Check: apply the 60-day title-plus-company-plus-location collapse, and if one firm drives a skill's count, discount it.
- **Chasing growth off a tiny base.** A skill at +70% year over year that sits in 0.2% of postings is still rare, and featuring it as core reads as naive. Check: require both a growth number and a penetration number before promoting anything to a bullet.
- **Keyword-stuffing the skills block.** It lifts your ATS match, then collapses at interview. 57% of employers have caught someone embellishing a skill set, and 55% caught exaggerated accomplishments. Check: ask whether you could survive a timed task or behavioral probe on the skill.
- **"Familiar with" hedging.** Vague qualifiers signal an unevidenced skill and read as cover for a gap. Check: rewrite as a quantified outcome or move the skill to keyword-only.
- **Single-source demand.** One vendor's scraped list can diverge from official indicators. Check: benchmark against the projections and wage data, and flag any skill only one source calls hot.
- **Stale layer.** Posting demand shifts monthly and tracked skill lists every fortnight, so a year-old section misrepresents the market. Check: date-stamp the layer and re-run quarterly.
- **Over-reducing a fragmented role.** For roles like Business Operations Manager no single skill tops about 27% of postings, so a tidy "core stack" is fiction. Check: if the top skill is under roughly 50% penetration, present a spread rather than a single baseline.

> **Watch out:** The filter is not the finish line
>
> Keyword-stuffing to pass the ATS backfires at the next stage. The skills assessment or behavioral probe is built to expose exactly the claim you could not back, and 57% of employers have already caught someone doing it.

The fragmentation failure deserves its own note. Business operations skills top 70% across postings as a category, but that is a category, not a single tool. Specific roles stratify differently: project management roles required a business operations skill 97.7% of the time, while customer service and administrative skills appeared in only 37.1% and 35.8% of all postings. Read your own role's distribution before you declare a "core stack."

## See your own differentiator in the index

Before you commit a skill to a bullet, it helps to see how rare or common it actually is among people who hold your target title. That comparison is what turns a guess into a classification.

Ask me this: `US-based data analysts who list SQL and a BI tool like Tableau or Power BI and have shipped dashboards in production.` - [run the search](https://www.refolk.ai/start?q=US-based%20data%20analysts%20who%20list%20SQL%20and%20a%20BI%20tool%20like%20Tableau%20or%20Power%20BI%20and%20have%20shipped%20dashboards%20in%20production.).

*Returns profiles matching the exact skill combination so you can gauge how common your stack is and spot where a differentiator like Python separates you.*

Seeing how many people in your market hold the same baseline makes the differentiating skill obvious, and it keeps you from treating a near-universal skill as if it were a selling point.

## Keep the layer current and verify before you ship

Refresh the demand layer quarterly and re-anchor to the annual sources yearly, matching the cadence of the slowest data you rely on. Posting data updates monthly and tracked skill lists shift fortnightly, so a quarterly refresh catches real movement; the projections and wage data are annual, so the structural baseline only needs a yearly review. Over-churning the resume on fortnightly noise is as wrong as letting it go stale.

Before you call the layer finished, run this check against the draft.

#### Before the layer ships

- [ ] One role and one market held constant across the whole layer
- [ ] Every featured skill has a share-of-postings figure, not just an importance score
- [ ] Counts are deduplicated on the 60-day title-company-location window
- [ ] Each skill is tagged near-universal or differentiating with both a penetration and a growth read
- [ ] Every bullet maps to a quantified, defensible outcome; unevidenced skills sit in keywords only
- [ ] No featured claim leans on "familiar with" or a similar hedge
- [ ] Each skill is corroborated by more than one source, or flagged if only one calls it hot
- [ ] The layer is date-stamped with a quarterly refresh and annual re-anchor on the calendar

The discipline that keeps this honest over time is the same one that built it: never feature a skill you cannot defend, and always re-check the number before you trust it. When you refresh, date the new layer, re-run the deduplication, and re-classify anything whose penetration or growth has moved. A resume built this way does not just pass a skills-first screen; it holds up when the person reading it asks you to prove the line.

## Frequently asked questions

### How do I find in-demand skills for my resume without a paid data tool?

Hand-code 20 to 30 live postings for your exact role and market. Paste each into a sheet, tally which skills appear, and divide by the number of postings to get a share figure. This reproduces what an aggregator does at smaller scale. Anchor it against the free BLS Employment Projections skills tables and OEWS so you are not chasing one noisy sample, and apply the 60-day duplicate collapse so one employer's reposts do not distort your counts.

### What counts as a near-universal skill versus a differentiating one?

A near-universal skill appears in a majority to supermajority of postings for your role and market. Business operations skills, for example, were required in more than 70% of all US postings at the end of 2025. A differentiating skill shows rapid growth off a small base: AI literacy grew more than 70% year over year yet still appeared in only about 1 in 500 postings. Near-universal skills are table stakes you must show; differentiators separate you but only if they are genuinely penetrating, not just fast-growing.

### Should a demanded skill go in a bullet or the skills block?

It depends on evidence. A demanded skill you can back with a quantified outcome becomes a resume bullet; a demanded skill you hold but cannot yet evidence goes in the skills block as a keyword for ATS matching. This split exists because 59% of employers want skills demonstrated with concrete examples rather than merely listed, and 57% have caught someone embellishing a skill set. The keyword that passes the filter is the exact claim tested at interview.

### How often should I rebuild the demand layer?

Refresh the posting-demand layer quarterly and re-anchor to the annual structural sources once a year. Real-time posting data updates monthly and tracked skill lists shift roughly every fortnight, so a year-old skills section misrepresents the market. The BLS projections and wage data are annual, so the structural baseline only needs yearly review. Date-stamp the layer each time you touch it so you know when it last matched reality.

### Why not just stuff the skills block with every hot keyword to beat the ATS?

It backfires at the next stage. Keyword-stuffing can lift your ATS match, but 57% of employers have caught skill embellishment and the usual flag is a skills test or behavioral probe. If you claim Python proficiency, a timed coding challenge exposes a keyword you cannot defend. Only feature a skill in a bullet if you could survive a practical task on it; everything else stays a keyword and is spoken about honestly at interview.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/candidates/guides/demand-sourced-resume-skills*