Research AI & skills
Teamed Research · October 2026The AI skills L&D employers actually want
AI appears in nearly half of L&D job postings. Almost none of them ask you to build anything. We read 9,135 job postings and 36,463 candidate documents to find out what employers are actually asking for, how candidates describe the same skills differently, and which of the shortage stories the data does not support.
9,135 job postings36,463 candidate documents2023 Q4 – 2026 Q3
Six percent to forty-six in eight quarters
AI mentions sat flat through 2024 at 8–10% of postings, then turned sharply in 2025 and have not stopped. The inflection is 2025 Q1.
The second line matters just as much. Postings naming generative AI specifically — a model, a chatbot, a named product — reached only 11.8%. Most of the growth is in the generic word “AI”, not in specific generative capability.
Only 12% of AI postings ask you to build anything
This is the number that should change how you plan your learning. For every posting mentioning AI we took the sentence the employer actually wrote and sorted it onto three rungs.
Build work is rare. Agents, custom assistants, retrieval systems, fine-tuning and model work together account for 11.8% of 2026 postings with AI language. The overwhelming majority want a practitioner who uses AI fluently inside existing L&D work — building courses, making content, reviewing output, teaching colleagues.
If you have been putting off AI because you think it means becoming an engineer — that is not what this market is asking for. What it asks for is an instructional designer who works fluently with AI in the room.
The shape of the ask: create something, and check your work
If 82% of postings want AI used rather than built, the obvious next question is: used to do what? We sorted every 2026 posting with AI language by whether its AI sentence asks the candidate to produce something, to use AI responsibly, or both.
Creation is close to universal. Responsible use is attached to more than half. And nearly one in two asks for both in the same breath — make something, and show you checked it.
Read together, the modal AI ask in L&D is not “can you use the tools”. It is closer to “can you make something with them, and can you tell me how you checked it”. Those are two different skills, and the counts above say employers are asking about both.
Four in five AI postings never name a tool
Only 17% of postings with AI language name a specific product — 19% in 2026. The ask is nearly always a capability, not a license. Learning a named tool to match a job post is answering a question most employers are not asking.
And when they do name something, the most-named AI product in L&D job posts is not a chatbot. It is AI video and avatar generation. L&D’s AI demand is shaped like L&D work: producing learning media at volume.
Claude went from zero mentions in 2024 to 46 in 2026, drawing level with ChatGPT once each vendor’s own job posts are excluded. Gemini tripled. Copilot rides Microsoft licensing rather than preference. The assistant market in L&D is two years old and already competitive.
Candidates name prompt engineering far more often than employers ask for it
Among 2026 documents and postings that mention AI at all, prompt engineering appears on 15.9% of candidate documents (230 of 1,445) and in 6.7% of job postings (68 of 1,021). Measured against every 2026 document and posting rather than only those mentioning AI, it is 8.0% against 2.9%. Either way the ratio is about 2.4–2.7×.
The same pattern shows up in the theme chart below: prompt engineering is the smallest AI theme in job postings and the one that has moved least — 0.4% of all postings before 2025, 2.9% in 2026 — while every other theme grew faster from a higher base.
This is a difference in vocabulary, not evidence that the skill is worthless or that anyone’s effort is wasted. Employers may assume prompting rather than specify it, and a practitioner who prompts well will do better work whether or not a job post names it. What the gap does say is that writing “prompt engineering” on a CV is unlikely to match words an employer is searching for.
The sleeper skill nobody is teaching
While prompt engineering stays flat, one AI theme has grown roughly tenfold in three years and almost nobody is talking about it: governance and responsible use — policy, bias, privacy, intellectual property, and the question of who checks the output. It went from 1.5% of all postings before 2025 to 15.6% in 2026.
Year-matched, there is no AI mention gap
Here is the trap, and it is how most “AI skills shortage” headlines get made. Across an entire multi-year candidate archive, only 10.3% of documents mention AI. Against a 45.8% posting rate last quarter that looks like a four-fold shortage.
It is an artifact. Archives are weighted to older material, so the aggregate measures when a document was written, not what anybody can do. Year-match it and the gap closes: 46.7% of documents written in 2026 mention AI, against 42.8% of postings published in 2026. Candidate documents mention AI slightly more often than postings do.
And where employers are vague, candidates are concrete: 55.6% of 2026 documents mentioning AI name a specific product, against 19% of 2026 postings. Candidates are roughly three times more specific than the job posts they are answering.
Both sides of this comparison are mention rates in text. Equal mention rates do not establish equal proficiency, equal availability, or that any particular employer can fill any particular role. What they do rule out is the specific claim that practitioners are not writing about AI while employers are asking for it. On the evidence here, they are writing about it at least as much.
AI language tracks decision scope more than technical specialism
AI language is not concentrated in the technical corners of L&D. The lowest rates are at intern and entry level and the highest are at manager level. But it is not a clean ladder: Director+ postings carry AI language less often than Senior ones, and the highest rate of all sits on the smallest band in the chart.
By specialism the spread is narrower than by seniority: strategy and leadership 42.4%, instructional design 34.4%, eLearning technology 33.0%.
Employers have settled on what they want, not on how you prove it
Put the findings together. Nearly nine in ten postings with AI language ask you to make something with AI. More than half ask how you use it responsibly. Nearly half ask for both.
Now look at what employers ask you to show. Portfolio and work-sample language appears in 39.3% of AI postings against 34.8% of all postings — a difference of 4.5 percentage points, or 1.13×. It is a real difference and a small one: employers asking for generative, responsibly-used AI work are only slightly more likely to ask for evidence of it than employers asking for anything else.
That is the state of the market as we measure it in October 2026: a clear and fast-growing request, and no settled convention for answering it.
What this means
For practitioners. The build rung is rare and the create-and-check pattern is common, so the highest-value thing to be able to show is not a tool certificate but a piece of work with your reasoning attached. The vocabulary gap on prompt engineering suggests describing what you produced and how you checked it, rather than naming the technique.
For employers. Four in five of your AI postings name no product, and more than half ask about responsible use without saying how a candidate should evidence it. Both are cheap to fix in the job post, and both are currently costing you signal from the candidates you want.
For the market. Governance language has grown roughly tenfold while the convention for demonstrating it has not emerged at all. That gap is the most likely thing to change in the next year.
What this measures, and what it does not
This is keyword presence in text, not verified capability — it measures what people wrote. “Mentions AI” is not the same as “requires AI”: a third of postings with AI language mention it exactly once, and 11% only in opening company boilerplate. Demand is the jobs listed on the Teamed board, which skews US corporate L&D, higher education and healthcare, and toward instructional design and eLearning roles — not the whole market. Supply is a candidate document archive, about 97% of it résumés. A third source, 4,176 candidate accounts carrying self-described skills and experience, was scanned for context and no finding in this report rests on it — which is why it is named here and not in the headline figures.
Themes and rungs are keyword rules applied to a ±250-character window around each AI mention, so what gets classified is the employer’s AI sentence rather than the whole document. Categories overlap except where a chart says rungs are exclusive. A product counts only on an unambiguous name. Short acronyms are matched on word boundaries, never as substrings.
Denominators differ between charts, so read the subtitle on each. Three counts of “2026 postings with AI language” appear in this analysis and they are not interchangeable: 1,021 from the pass that classifies rungs, themes and named tools; 1,006 from the create-and-check pass, which requires a classifiable AI sentence; and 993 from a strict word-boundary count over the full posting text. Percentages are always stated against the denominator printed on the chart they come from.
The counterweight to the caveat above: where AI does appear, it is increasingly written as a requirement. In 2026, 49.1% of postings with AI language put it in requirement language, against 7.7% that frame it as preferred-only. AI has moved from the nice-to-have list into the qualifications list.
Cite this report
Teamed runs a specialist marketplace for L&D hiring, which means we hold two things side by side that are rarely seen together: what employers write when they are hiring, and what practitioners write about themselves.
Teamed Research (2026). The AI skills L&D employers actually want. Teamed. www.teamedforlearning.com/research/ai-skills-ld-employers-want-2026/
Figures are Teamed's. Please keep the method note attached — every number is a floor on what the text says, not a measure of what people can do.