AI in a Job Ad: Will You Build It or Use It?

Does an AI mention mean this is an AI-builder job?

When an ad says “AI experience required,” it is reasonable to wonder whether you need machine-learning credentials or simply need to use an approved tool in your profession. The keyword alone cannot settle that question. Read it beside the job’s recurring outputs, action verbs, and technical requirements.

Indeed Hiring Lab reviewed several hundred thousand U.S. postings containing AI-related terms from July 2024 through June 2025. Its contextual classification placed 52% in a broad category combining development and interaction with models; roughly a quarter had no clear thematic fit. The authors caution that broad mentions can signal an employer’s AI posture without explaining the job’s substance. Because the categories merge different activities, these figures describe posting language, not your odds of needing engineering skills or an employer’s actual adoption.

A UK government vacancy analysis offers a more useful distinction than a single “AI job” label: it searched for broad implementer roles, technical specialists, and experts. The study analyzed hundreds of thousands of online postings, but its categories depend on search definitions and can overlap; the data do not verify what workers actually do. Use that taxonomy as context, then classify the specific opening from its duties rather than assuming every AI mention signals technical ownership.

Sources: How Employers Are Talking About AI in Job Postings; AI Skills for Life and Work: Job vacancy analysis

Which clues separate building AI from using it?

Use four clues together: verbs, deliverables, technical stack, and accountability. They are practical reading signals, not proof. A job can mix both kinds of work, and a posting may describe them poorly.

First, look at the verbs. “Design,” “train,” “evaluate,” “deploy,” “integrate,” and “maintain” suggest someone owns a system or connects one to a product. “Use,” “draft,” “summarize,” “analyze,” and “review” more often describe applying an existing tool to ordinary work. Then check the deliverable: a model, production service, or data pipeline points toward building or integration; a market brief, customer response, forecast, or campaign points toward using AI as one means to produce a business result.

Next, look for a stack and prerequisites. Python, machine-learning frameworks, model evaluation, cloud deployment, or data-pipeline work strengthen the technical interpretation. A named workplace tool or a request for workflow fluency may indicate applied use, though even that can involve substantial judgment. Integration deserves its own category: connecting a vendor model through an API can require software and testing skills without requiring you to train a foundational model.

Finally, ask what you would be accountable for. If the role owns system reliability, evaluation, or production performance, technical responsibility is central. If it owns the accuracy and usefulness of a report or customer decision, an existing tool may assist while the worker checks its output. The UK analysis separates broad implementer searches, which can include use of generative AI or language-model tools, from specialist and expert searches tied to implementation and deeper technical work. Those are market-level search categories, not a universal rule for reading every advertisement.

Sources: AI Skills for Life and Work: Job vacancy analysis

How much should the wording change your application?

Match your response to the work described, not to the loudest keyword. For applied use, show a relevant workflow and how you verify its output. For a build or integration role, show the particular technical capability needed by the deliverable. If the mention is vague or preferred, do not treat it by itself as a reason to abandon your experience or pay for a broad course.

For example, an experienced business analyst applying to a role that asks for AI-assisted research could describe the task, tool category, checks against source material, and report delivered. That is evidence of applied fluency; it does not imply model-development experience. A candidate for a role responsible for deploying an AI feature should instead be ready to show relevant software, data, integration, or evaluation work if those are explicit requirements. The title matters less than the promised output and ownership.

Separate “required” from “preferred,” then compare each criterion with evidence you can honestly show. A small gap in a tool workflow may justify a short, task-specific practice project; a central technical requirement may call for deeper study or a different opening. Andrew Green’s 2024 OECD working paper gives broad context: its first estimates concern skill demand in occupations highly exposed to AI, and its abstract says most exposed workers will not need specialized AI skills such as machine learning or natural-language processing. It also reports changing task and skill demand, including an emerging decline in some skills in panel-establishment evidence. This is an aggregate labor-market analysis, not a study of individual job ads or a guarantee about one vacancy. It argues against an automatic detour into ML engineering, not against learning a skill the stated deliverable requires.

Sources: Artificial intelligence and the changing demand for skills in the labour market

When should you ask before deciding what to learn?

Ask the recruiter or hiring manager when the ad says AI is required but gives no use case, tool, deliverable, or ownership level. Try: “Could you share one recurring task where this role uses AI, and whether the hire develops or configures the system or works with an existing tool? What would the person be expected to verify or own?” A concrete answer helps you compare the job with your experience. A vague answer leaves uncertainty; it does not prove hidden engineering duties exist.

Postings are recruitment documents, not verified descriptions of day-to-day work. Indeed found a sizable group of sampled AI-related ads with little contextual detail. The UK analysis uses Lightcast’s proprietary online-posting database, which deduplicates postings and extracts skills; its report flags possible selection bias where employers recruit through closed platforms or other channels, as well as classification risk from search definitions. Neither source establishes how an employer uses AI internally, what a particular worker will do, or displacement.

The verdict is simple: infer from tasks, then verify ambiguity. Underline the AI-related verbs; identify the output and who owns its quality; check the technical stack and whether the criterion is required; ask for an example if those clues conflict. Apply with evidence of relevant work when the core role fits. Choose one bounded learning step only after naming the actual gap. If you are still mapping which tasks in your current work could change, the free task-level checker can help organize that reflection, but it cannot tell you what an employer intended by a job-ad phrase.

Sources: How Employers Are Talking About AI in Job Postings; AI Skills for Life and Work: Job vacancy analysis