By Tristan Tully and Oriane Casale
Summary: This article looks at mentions of AI keywords in national job postings over two periods spanning six months, September 2025 and April 2026. Overall, 8% of April 2026 postings had AI-related keywords compared to 7% of postings six months earlier. In April 2026, the majority of these mentions, 68%, were for generic AI-related terms, while 32% were for specific AI tools, skills or knowledge. By occupation, the highest numbers and shares of AI-mentions were in Computer & Mathematical, Architecture and Engineering, Legal, Management, and Business & Financial Operations occupations. By education level, job postings requiring a Bachelor’s degree or higher had both the highest numbers and highest shares of AI-related key words. Based on these finds, we expect the role of AI in technical, information-technology, analytical and managerial occupations to continue to expand.
How is AI impacting jobs? Are employers asking for AI skills and how is that showing up in job postings? Do I need to know how to use AI to get a job in my field? These questions are on the minds of job seekers’, career counselors, students, educators and others across the country. This article presents findings on AI-related skills or knowledge mentions in nationwide job postings in the National Labor Exchange (NLx) spanning a six-month period: September 2025 and April 2026. The analysis looks at AI-related keyword in job postings by type of AI knowledge required as well as occupation and education requirements of these postings.
Overall, this analysis finds that 8% of April 2026 job postings had AI-related skills or knowledge mentions based on our list of 141 AI-related keywords (excluding header/footer references to Human Resource processes or generic job expectations). This was an increase from 7% in September 2025. The majority of these mentions, 60%, were for just two key words: AI and artificial intelligence – likely indicating a general commitment to the use of AI on the job rather than a specific use or skill/knowledge. Occupation groups with the greatest shares of AI-related keyword mentions included military-specific, legal and computer sciences. In line with this, the percent of job postings with AI mentions increased as years of education required for the position increased. In the methodology section we will discuss the impact of duplicate job postings and how we handled them in this analysis.
This analysis does not attempt to predict how AI will impact the availability of jobs or the overall demand for labor in the future. Instead, we focus on the status quo – what job postings from September 2025 and April 2026 can tell us about how and which jobs are being impacted now and how employers are thinking about AI. We have included a reading list at the end of this article that includes research with a broader focus for those who are interested.
AI-Related Keywords
This analysis divides the list of 141 keywords that we use to identify AI-related job postings into two categories. Category 1 keywords refer to the generic, unspecified use of AI tools to improve existing job tasks. Keywords in this category include, for example, AI, machine learning and ChatGPT. Category 2 keywords are specific AI tools or knowledge, representing a deeper level of integration of AI into the job. Because these tools are relatively new, this category represents jobs that have been transformed or potentially newly created as a result of this technology. Examples of these keywords are tensorflow, PineCone and Azure OpenAI.
The total number of AI keywords identified in job postings increased from September 2025 to April 2026. There was growth in both categories of AI-related keywords.
Figure 1: Count of Job Postings with AI-Related Keywords by Category of Keyword, September 30, 2025 and April 21, 2026
Occupation
This part of the analysis looks at which occupational groups are most and least likely to have job postings with AI-related key words. By absolute size, the top five occupation groups with the highest number of AI-related job postings are, in order: Computer and Mathematical; Management; Architecture and Engineering; Business and Financial Operations; and Sales and Related. The top five by total job postings remained the same between September 2025 and April 2026, except Healthcare Practitioners and Technical replacing Sales and Related. While Healthcare job postings with AI mentions grew over the 6-month period, Sales and Related grew much faster. Every occupational category increased in AI-related mentions over the period except Food Preparation and Serving.
By share of job postings with AI-related key words (see Figure 2), the top occupational groups were Computer and Mathematical, at 13% ; Architecture and Engineering at 6%; Legal at 6%; Business and Financial Operations at 5%; and Management at 5% (also note Military was excluded for having less than 17 jobs listed in total). The top five occupational categories remained the same by share between the two time periods, with each category increasing in AI-related share.
Occupations with both large numbers and a relatively high share of AI-related key words are Computer & Mathematical, Architecture and Engineering, Legal, Management, and Business & Financial Operations occupations. These findings illustrate AI’s growing role in supporting technical, information technology, analytical and managerial white-collar jobs. In addition, t here is secondary evidence that at least Healthcare Practitioner occupations use AI enhanced tools extensively (Faiyazuddin, M.D. et al. 2025), but these tools are either not frequently mentioned in job postings or the mentions are so specific that we did not capture them in our list of terms on which we’re searching.
Looked at from bottom up, the five occupation groups with the lowest shares of AI-related jobs are Food Preparation & Serving (0%); Building and Grounds Cleaning and Maintenance (0%); Farming, Fishing, and Forestry (1%); Protective Service (1%); and Healthcare Support (1%.) These results reinforce the finding that AI is increasingly incorporated into white-collar jobs, whereas blue-collar occupations are less likely to be impacted by AI.
Figure 2: Percent of Job Postings with AI Mentions by Occupation
Education
Just as the findings by occupation indicate that white-collar job postings are most likely to require AI-related skills or knowledge, the findings by education indicate the same. The share of job postings mentioning AI-related skills/knowledge increases as years of education required increases.
Job postings requiring a graduate degree had the highest share of AI mentions, at 19% (16% in September 2025). Job postings requiring a Bachelor’s degree had the second highest, at 6% (5% in September 2025), Associates degree was 1% (1% in September 2025) and high school diploma was 1% (0% in September 2025) (see Figure 3). The number job postings with AI mentions increased in all educational categories between September 2025 and April 2026. Even though graduate-level jobs have the greatest share of AI mentions, the majority of job postings involving AI require or prefer a Bachelor’s degree (see Figure 4).
Figure 3: Percent of AI-Related Job postings by Education Level, September 2025 and April 2026
Figure 4: Count of AI-Related Job postings by Education Level, April 2026
Methodology
This analysis examines national job postings data available on NLx, which were accessed through the NLx Research Hub. We developed a Python program written with AI-assistance to scan the openings using fuzzy keyword matching. A total of 1.5 million job postings from two periods, September 2025 and April 2026, were scanned. Fuzzy matching (Aho-Corasick in our case) is a statistical matching method that scans inputs for matches at a given percent of matching. For our research, we used a 90% threshold for keywords, meaning 90% of letters in any given set had to match keywords.
Our keyword list, developed in collaboration with the New Jersey Office of Research and Information at the New Jersey Department of Labor and Workforce Development was a combination of internal review of keywords related to AI and program-specific keywords provided by Yifan Wang, Ph.D. To avoid false-positives, our team iterated a filter to remove AI-mentions not directly related to job responsibilities. In particular, we were careful to filter out AI-related disclaimers for human resources-related processes. Because we noticed an abundance of job postings mentioning AI only in passing, we required job postings to mention AI at least twice to be included as AI-related.
Moving onto the dimensions we used for findings, our approach to each of these differed slightly. For the occupation-specific results, NLx applies O*NET/SOC codes to each job posting using an autocoder. Our analysis grouped job postings at the 2-digit SOC level. This was to simplify occupation categorization for discussing results and to ensure there was a sufficient sample for all categories for statistically valid results. The Military-related SOC had no more than a few hundred job postings. For this reason, this category is largely ignored when discussing the results.
Education-related results were achieved by cross-examining the AI-keyword scan with education keywords, using the same fuzzy-match method. We used multiple keywords for each level of education to ensure we captured all common means of referring to levels of education without being too vague to have false positives.
For our geographic analysis, we used structured data fields for city and state locations for posted jobs. Geography raised the issue of Ghost Jobs. Ghost Jobs refer to a category of job postings that are not necessarily in the location they are coded to, may not exist, are remote and/or may be duplicate (for example ghost jobs are duplicated 50 times, once for each state). Direct Employers, the supplier of these job postings, tags job postings with these various characteristics as Ghost Job.
Due to the inconsistent, and sometimes nonexistent status of Ghost Jobs, we opted to exclude them from examination of AI-related job postings. It is critical to note that the share of AI-related job postings when excluding Ghost Jobs is 8%. When including Ghost Jobs, the share is as high as 13%. When we examined only Ghost Jobs, we found half of them were AI-related job postings. Because ghost jobs differ so strongly from other jobs in terms of AI-mentions, we think it is important to exclude them from AI-related analysis.
Because our methodology involved scanning job postings, it does not look at job losses due to the introduction of AI. Instead, it focuses on the increasing prevalence of AI mentions in job postings across the country. However, the later period of job postings (April 2026) had both more total job postings and a higher share of AI-related jobs.
Conclusion
White-collar and college graduate job seekers can expect AI to be involved in some aspects of their occupation. Even if not all individuals within the occupation interact with AI directly, members of their teams, supervisors and managers are already likely using it or will be soon. Based on our findings, we expect the role of AI in technical, information technology, analytical and managerial roles to continue to expand.
By comparison, blue-collar and high school graduate job seekers can expect to see far less impact from AI in their occupations. To the extent that AI is involved in blue-collar occupations, we expect it is mostly used at the supervisory and managerial levels for white-collar tasks. Based on this analysis, the occupational groups most affected include Computer & Mathematical, Architecture and Engineering, Legal, Management, and Business & Financial Operations occupations. Job seekers in these fields, and those seeking jobs with high educational requirements, may benefit from familiarizing themselves with AI tools and their uses.
References
Keywords Used
| Keyword | Category |
|---|---|
| AI Personalization | 1 |
| AI Security | 1 |
| Artificial General Intelligence | 1 |
| Artificial Intelligence | 1 |
| Artificial Intelligence Development | 1 |
| Artificial Intelligence Systems | 1 |
| Artificial Neural Networks | 1 |
| Autonomous Vehicles | 1 |
| ChatGPT | 1 |
| Conversational AI | 1 |
| Convolutional Neural Networks (CNN) | 1 |
| Deep Learning | 1 |
| Face Detection | 1 |
| Facial Recognition | 1 |
| Google Gemini | 1 |
| GPT-4 | 1 |
| Image Generation | 1 |
| Image Recognition | 1 |
| Image Segmentation | 1 |
| Instance Segmentation | 1 |
| Intelligent Virtual Assistant | 1 |
| Language Models | 1 |
| Machine Translation | 1 |
| Machine Vision | 1 |
| Microsoft Copilot | 1 |
| Object Recognition | 1 |
| Object Tracking | 1 |
| PineCone | 1 |
| Prompt Engineering | 1 |
| Semantic Kernel | 1 |
| Semantic Search | 1 |
| Sentiment Analysis | 1 |
| AI | 1 |
| LLM | 1 |
| AdaBoost (Adaptive Boosting) | 2 |
| Adversarial Machine Learning | 2 |
| Agentic AI | 2 |
| AI/ML Inference | 2 |
| AI Agents | 2 |
| AI Copywriting | 2 |
| AIOps (Artificial Intelligence For IT Operations) | 2 |
| Amazon Lex | 2 |
| Amazon Textract | 2 |
| Apache MXNet | 2 |
| Apache OpenNLP | 2 |
| Applications Of Artificial Intelligence | 2 |
| Attention Mechanisms | 2 |
| Autoencoders | 2 |
| AutoGen | 2 |
| Automated Machine Learning | 2 |
| AWS Certified Machine Learning Specialty | 2 |
| AWS SageMaker | 2 |
| Azure AI Language Understanding (LUIS) | 2 |
| Azure Cognitive Services | 2 |
| Azure Machine Learning | 2 |
| Azure OpenAI | 2 |
| BERT (NLP Model) | 2 |
| CatBoost (Machine Learning Library) | 2 |
| Collaborative Filtering | 2 |
| Computer Vision | 2 |
| Contextual Image Classification | 2 |
| DALL-E 3 | 2 |
| Deep Learning Methods | 2 |
| Deep Reinforcement Learning (DRL) | 2 |
| Distributed Machine Learning | 2 |
| Explainable AI (XAI) | 2 |
| Fast.ai | 2 |
| Federated Learning | 2 |
| Few Shot Learning | 2 |
| Generative Adversarial Networks | 2 |
| Generative AI Agents | 2 |
| Generative Artificial Intelligence | 2 |
| Google AutoML | 2 |
| Google Cloud Dialogflow | 2 |
| Graph Neural Networks (GNNs) | 2 |
| H2O.ai | 2 |
| Hugging Face (NLP Framework) | 2 |
| Hugging Face Transformers | 2 |
| Keras (Neural Network Library) | 2 |
| LangChain | 2 |
| Large Language Modeling | 2 |
| LightGBM | 2 |
| LLaMA (Language Model) | 2 |
| Long Short-Term Memory (LSTM) | 2 |
| Machine Learning | 2 |
| Machine Learning Algorithms | 2 |
| Machine Learning Methods | 2 |
| Machine Learning Model Monitoring And Evaluation | 2 |
| Machine Learning Model Training | 2 |
| Meta-Learning | 2 |
| Meta-Reinforcement Learning | 2 |
| Microsoft Certified: Azure AI Engineer Associate | 2 |
| Microsoft Certified: Azure AI Fundamentals | 2 |
| MLOps (Machine Learning Operations) | 2 |
| Multimodal Learning | 2 |
| Multimodal Models | 2 |
| Natural Language Generation | 2 |
| Natural Language Generation (NLG) | 2 |
| Natural Language Processing (NLP) | 2 |
| Natural Language Understanding | 2 |
| Natural Language Understanding (NLU) | 2 |
| Neural Architecture Search (NAS) | 2 |
| Neuro-Symbolic AI | 2 |
| OpenAI Gym | 2 |
| OpenAI Gym Environments | 2 |
| Operationalizing AI | 2 |
| Pose Estimation | 2 |
| PyTorch (Machine Learning Library) | 2 |
| Qdrant | 2 |
| Recommender Systems | 2 |
| Recurrent Neural Network (RNN) | 2 |
| Recurrent Neural Networks (RNNs) | 2 |
| Reinforcement Learning | 2 |
| Reinforcement Learning (RL) | 2 |
| Reinforcement Learning from Human Feedback (RLHF) | 2 |
| Residual Networks (ResNet) | 2 |
| Retrieval Augmented Generation | 2 |
| SAS Certified Professional: AI & Machine Learning | 2 |
| SAS Certified Specialist: Machine Learning | 2 |
| SAS Certified Specialist: Natural Language Processing And Computer Vision | 2 |
| Scikit-Learn (Python Package) | 2 |
| Semi-Supervised Learning | 2 |
| Sentence Transformers | 2 |
| Seq2Seq | 2 |
| Sequence-to-Sequence Models (Seq2Seq) | 2 |
| Small Language Model | 2 |
| SpaCy (NLP Software) | 2 |
| Speech Recognition | 2 |
| Stable Diffusion | 2 |
| Supervised Learning | 2 |
| Synthetic Data Generation | 2 |
| TensorFlow | 2 |
| Transfer Learning | 2 |
| Transformer (Machine Learning Model) | 2 |
| Variational Autoencoders | 2 |
| Vertex AI | 2 |
| Weaviate | 2 |
| Word2Vec Models | 2 |
| Word Embedding | 2 |
| Xgboost | 2 |
| Zero Shot Learning | 2 |