If your resume lists "ChatGPT" or "prompt engineering" as a standalone skill, you're leading with the skill employers now ask for less. Workday found that demand for basic prompting fell 25% from its January 2026 peak, while demand for building AI tools and automating workflows rose 51%.
This piece shows how to turn that line into evidence, with before/after bullets for engineers, analysts, marketers, and operators.
What did Workday's October 2026 report actually find?
Workday's data shows employers shifting from people who use AI to people who build with it. Prompting demand peaked in January 2026 and then fell 25%. Demand for building AI tools, automating workflows, and AI engineering climbed 51% between September 2025 and July 2026.
Workday based this on job postings from about 550 employers that run its recruiting software. Its own summary is blunt: "Companies have moved on from wanting people who can just use AI to wanting people who can build with it."
The competition numbers explain why this matters for your resume in particular:
- 84% of job seekers said they used AI during their search (Workday AI@Work Pulse survey).
- 69 applicants per filled role at the median, up from 58 a year earlier.
- ~60 days median time to fill, roughly flat. Employers get more applicants and don't hire any faster.
- +27% applicants per filled role year over year in financial services.
- +40% in technology and media.
Phil Willburn, VP of people systems at Workday, framed the broader shift this way: "Employees may not be changing jobs, but their jobs are changing around them."
One caveat. The skills trend covers about 550 enterprise employers over ten months. That's enough to show which way demand is moving, but it's a short window, so treat 25% and 51% as a direction, not a precise forecast.
Workday's October 2026 report, with 84% of job seekers saying they use AI in their search.
Why does a "prompt engineering" line no longer stand out?
A "prompt engineering" line no longer stands out because far more people list it than list any hands-on building skill. In Refolk's index of professional profiles, about three US professionals list "Prompt Engineering" for every one who lists LangChain, a framework for building AI applications.
| Skill listed | US profiles | Multiple of LangChain |
|---|---|---|
| Prompt Engineering | 66,312 | 3.01x |
| ChatGPT | 37,558 | 1.71x |
| LangChain (build signal) | 22,009 | 1.00x |
Source: Refolk's index, US only, one query per skill. Profiles overlap, so the rows don't add up to a total.
The mechanism is basic supply and demand applied to a keyword. Employers want prompting less. Meanwhile 66,312 people still put it forward as a differentiator. A label that's common and losing demand can't separate you from the other 68 applicants. A build signal is rarer and gaining demand, so it can.
The second finding is more awkward. More people list "Prompt Engineering" than list "ChatGPT", by 1.77x. The inflated label beats the name of the tool many of them probably use. On a lot of profiles, "prompt engineering" likely just means "I use ChatGPT," and a recruiter has no way to tell which profiles it doesn't.
Refolk's index, US professionals. The label is common, the hands-on proof is rare.
Builders list "prompt engineering" too, so the word proves nothing on its own
The term "prompt engineering" isn't the problem. Listing it with nothing around it is. In a sample of 25 profiles from the "Prompt Engineering" group in Refolk's index, titles include "Principal AI Systems Engineer" and "Engineering Director, Agent Stack." Those titles point to people building systems, and they list the same skill as someone who once rewrote an email in ChatGPT.
Companies such as Vercel, Airbyte, Dell Technologies, and Infoblox show up in these samples. Vercel appears in both the Prompt Engineering and ChatGPT groups. These are sample appearances, not hiring signals. The point is that the same two words cover everyone from casual users to people who work on agent infrastructure.
So don't delete "prompt engineering." Give it context. Put it inside a bullet that names:
- The system you built or changed.
- The input that fed it (tickets, CRM records, query logs, briefs).
- The output it produced.
- The result in time, volume, errors, or revenue.
That's how a screener tells you apart from the casual user who listed the same skill.
Why do recruiters discount AI skill lines?
Recruiters discount AI skill lines because many of them are padded. Greenhouse found that 62% of US respondents tried at least one tactic to get past screening software, and 40% of those added skills or keywords they don't have.
Tom Hyde of Greenhouse APAC called candidate AI use a "doom loop of candidates feel like they need to do it to stand out." Everyone adds the keyword, the keyword stops meaning anything, and so everyone adds more keywords.
Employers make it worse from their side. Reporting on Lightcast's pay research notes that many employers sprinkle "artificial intelligence" into postings to attract applicants, which blurs the signal in both directions. The posting says AI, the resume says AI, and neither tells the other anything. So before you rewrite for a posting, work out whether it actually wants a builder or a user. Refolk scores how well your actual history fits each posting, so you can see whether your automation work is the main qualification or a side point, and decide how much space to give it.
An AI skill with no named tool, no output, and no number is exactly the claim screeners have learned to skip.
What gets past that skepticism is something the reader can check: a link, a metric, or how many people or teams adopted what you built.
How to list AI skills on a resume so they read as proof
Write every AI claim as something you built or automated, using a named tool, with a measured result and an adoption count. Specific AI skills carry more weight than vague ones. That's true in pay data and it's true on a resume.
Lightcast's research found that jobs advertising at least one AI skill carried salaries 28% above similar roles with none. When a posting names two or more AI skills, the skills get more specific and the premium climbs as high as 43%. (The source notes the 43% figure may be carried over from 2025.) The reason is the same one that applies to your resume. Specific mentions carry information, and vague ones get averaged into noise.
Use this template:
[Built / Automated] [workflow] using [tool] → [time, volume, or error result], adopted by [N people or teams].
Check each AI bullet against this list before it goes on the page:
- Verb is build-side: built, automated, shipped, integrated, evaluated, replaced. Not "leveraged," "utilized," or "proficient in."
- Tool is named: LangChain, Zapier, a specific GPT model. Not "AI tools."
- Workflow is concrete: ticket triage, lead scoring, report generation, contract review.
- Result has a number: hours saved, items processed, error rate, turnaround time.
- Adoption is stated: who uses it besides you.
If you can't fill in the result, you probably experimented with something and didn't build it. That's fine, but it belongs in the skills section, not a bullet.
Writing this once is manageable. Rewriting it for every posting, because one employer wants "LLM integration" and the next wants "workflow automation," is where most people go back to the generic line. That's the work Refolk takes off your hands: it builds your resume from your own history, then rewrites it for each posting you paste in, keeping your actual tools and numbers and phrasing them the way that employer does.
Before and after: AI skill bullets for four roles
The fix is the same for every role: replace the claim of familiarity with a record of something you built. The numbers below are placeholders. Use your real ones, and if you don't have one, measure it before you claim it.
| Role | Before | After (template) |
|---|---|---|
| Engineer | Prompt engineering, ChatGPT | Built a LangChain retrieval service over internal docs, cutting support escalations by [X]%, used by [N] teams |
| Analyst | Proficient in ChatGPT for analysis | Automated weekly KPI commentary by feeding query output to a GPT model, reducing report prep from [X] hours to [Y] |
| Marketer | AI tools for content creation | Built a GPT-based brief-to-draft workflow in Zapier, raising output from [X] to [Y] assets/month with [Z]% first-pass approval |
| Operator | Experienced with AI productivity tools | Automated vendor invoice intake with Zapier + GPT, processing [X] invoices/week and cutting entry errors by [Y]% |
Engineers
Engineers have the easiest time and still get it wrong most often, because listing "LangChain" feels like enough. It isn't. Name the architecture (retrieval, agents, evals), the data source, and the production signal: latency, cost per call, accuracy on an eval set, or the number of internal users. If you wrote evals or guardrails, say so. That work is the scarce part.
Analysts
The ChatGPT skills resume example most analysts write is "used ChatGPT to accelerate analysis." Swap it for the pipeline: which query or dataset fed the model, what it produced (anomaly summaries, narrative commentary, categorized feedback), and who reads the output. Applicants per filled role rose 27% in financial services, so a vague line has more competition to get lost in.
Marketers
Marketers should drop "AI tools for content" entirely. Name the workflow you built around the model: brief intake, brand-voice checks, variant generation, routing for approval. Then report throughput and quality, not just volume. Volume alone invites the question of how much cleanup it needed.
Operators
Operators often have the best AI automation resume bullet available and leave it off because "it was just Zapier." Workday counts automating workflows as a hands-on, in-demand skill. A no-code flow that runs every day and that a team depends on is a build. Write it like one.
Does "building with AI" require writing code?
Building with AI does not require writing code. Workday explicitly groups "automating workflows" with building AI tools, and Lightcast found that as of 2024, 51% of job postings requiring AI skills were outside IT and computer science.
Cole Napper, VP of research and insights at Lightcast, has pointed to AI demand moving beyond tech roles. That's the opening for operators, marketers, recruiters, and finance teams. The AI skills employers want in 2026 include connecting a model to a real process so that it runs without you.
Things that honestly count as building, without code:
- A Zapier flow that calls a model and writes the result somewhere useful.
- A shared GPT assistant with a defined input, instructions, and a team that uses it.
- A no-code automation that classifies, routes, or summarizes incoming records.
- A documented prompt library, versioned and adopted as standard by a team, with before/after quality checks.
Things that don't count: using ChatGPT to write your own emails, or having "explored" a tool. Those are fine to mention in an interview, but they aren't bullets.
FAQ
Should I remove "prompt engineering" from my resume?
No. Keep it once in your skills section, and only if a bullet in your experience shows you building something with it. In Refolk's index, 66,312 US professionals list the term, including principal-level AI engineers, so the word alone doesn't hurt you. It just doesn't help. It's the context around it that separates you from the casual user.
What if I've only used ChatGPT for writing and research?
Then list "ChatGPT" in skills by its real name, not "prompt engineering," and leave it out of your bullets. Better still, spend a week automating one repetitive task at work, measure it, and write that up. One real automation beats a skills line that tens of thousands of other profiles share.
How many AI skills should I list?
List only the ones you can back up with a bullet. Lightcast found pay premiums rise when postings get more specific about AI (28% for at least one AI skill, up to 43% for two or more), which shows how much specificity is worth. But Greenhouse found 40% of US candidates who gamed screeners added skills they don't have, so recruiters are watching for padding. A few named tools you can back up beat a long list.