Prompt engineer is one of the most frequently cited “jobs of the future” in hiring reports and think pieces. It’s also one of the hardest roles to actually recruit for — because the market still hasn’t agreed on what the role actually is.
Some companies want someone technical: Python, API fluency, experience building LLM pipelines. Others want someone with a feel for language, product thinking, and the ability to translate business requirements into model instructions. Still others need an operator — someone who takes existing AI tools and squeezes them into a specific business process.
All three are right. The problem is they’re looking for completely different people under the same job title.
This article answers one question: before you start searching for a prompt engineer, you need to know which one you’re looking for. Because the candidate profile, sourcing channels, job ad requirements, and interview questions — all of it hinges on the answer to that single question.
What a Prompt Engineer Actually Is in 2026
The role has evolved faster than its definition. Two years ago, “prompt engineering” mostly meant manually crafting queries in ChatGPT. Today it’s a term covering at least three very different professional profiles.
Profile A — Technical
Builds prompt systems, integrates language models into production applications, designs RAG (Retrieval-Augmented Generation) architecture, writes Python. Thinks of a prompt as an interface between a model and a system — not as a command typed into a chatbot. Works shoulder-to-shoulder with ML Engineers and backend developers. In practice, this role sits closer to engineering than to communication.
Profile B — Product
Designs how users interact with AI inside a product. Defines system prompts that shape the model’s personality and behavior, builds output quality evaluations, collaborates with designers and product managers. Understands both model capabilities and end-user needs. It’s closer to a Product Manager than a programmer — except the “product” they’re designing is how a language model behaves.
Profile C — Operational / Content
Takes existing AI tools and deploys them inside specific business processes: content generation, customer support, CV pre-screening, reporting. Optimises prompts for concrete tasks, measures effectiveness, trains teams on tool usage. Closer to an analyst or process operator than an engineer. What matters here is deep knowledge of the business process and the ability to think at scale.
Before you write a job ad — answer this: which of these three profiles actually solves your specific problem? The answer changes everything you do next.
Requirements: What Actually Matters
Prompt engineer job ads are full of requirements that sound reasonable but verify nothing in practice. Before you write your list of expectations, it’s worth knowing which ones actually separate a strong candidate from an average one.
What genuinely matters
Documented examples of prompts and their outcomes. Every practitioner leaves traces — in repositories, public projects, case write-ups. A candidate with nothing to show worked with AI too superficially to have a portfolio. The absence of examples is a warning sign, not a neutral observation.
Practical knowledge of model differences. “I’ve used GPT-4” isn’t enough. A strong candidate can say: “For this task I chose Claude because it handles long documents better and hallucinates less when citing sources. For fast content variant generation I use a faster, cheaper model.” Conscious model selection is a sign of maturity.
The ability to design evaluations. This is the point that separates practitioners from users. A good prompt engineer doesn’t just write prompts — they measure whether they work. They know how to define a success metric for generated text, how to build a test set, how to compare two prompt versions quantitatively rather than just by gut feel.
Iterative thinking. Prompt engineering isn’t writing — it’s testing. A candidate who describes their work as one-time prompt creation followed by deployment doesn’t understand the role. A practitioner iterates dozens of times, documents what worked and why, and reacts when a model update changes the behavior they’d built around.
For the technical profile, additionally: Python sufficient for API integration, familiarity with LangChain or similar frameworks, understanding of RAG architecture and the basics of fine-tuning.
What sounds important but verifies nothing
“Familiarity with ChatGPT” — in 2026, this is like requiring familiarity with Google Search. Too generic to mean anything.
“Creativity” — unmeasurable in a hiring process. Every candidate will say yes. Ask instead for a concrete example of a non-obvious problem they solved with AI.
Prompt engineering certificates — the market is too young for certificates to carry real weight. No recognised institution has established a standard yet. A certificate signals interest in the topic, not competence.
Years of experience — the prompt engineer role in its current form is 2–3 years old. Every candidate has a similar tenure. What counts is depth, not length.
Where to Look for Prompt Engineer
Prompt engineers — especially the technical and product profiles — rarely browse job boards. They’re passive candidates, but highly active in niche online communities. If you’re sourcing exclusively through LinkedIn, you’ll reach people actively looking for work. The best ones usually aren’t.
Hugging Face — forums, model comments, public Spaces. People who build and publish projects there are active practitioners. Look at who’s commenting, asking technical questions, publishing their own solutions.
GitHub — repositories connected to LangChain, LlamaIndex, PromptFlow, Semantic Kernel. Look not just at authors but at active contributors and PR reviewers. A contributor understands the code well enough to improve it.
Discord — Anthropic, OpenAI Developer Forum, LangChain, PromptLayer servers, and local AI community channels. These are places where practitioners ask questions, solve each other’s problems, and share what they’ve discovered. An active discussion participant signals high engagement.
X / Twitter — a public AI thread is often the best portfolio a candidate can show you. Look for people regularly writing about prompt engineering, LLMs, and their own experiments. Check whether their followers are practitioners or random accounts.
LinkedIn — but not through job postings. Through posts and comments. Someone who regularly comments on LLM articles with specific, substantive perspective is a valuable candidate — even if their profile doesn’t look like a “classic prompt engineer.”
Local AI meetups — in major cities, there are growing communities of AI practitioners who gather regularly. The networking value of these events for a recruiter is disproportionately high relative to the cost of attending.
How to Write a Good Job Ad
Most prompt engineer job ads share the same mistakes. Naming them is faster than describing what to do instead — because if you avoid the traps below, your ad will already be better than 80% of what’s out there.
Mistake 1 — Requirements too vague to mean anything
“Familiarity with AI tools” is not a requirement. “Experience designing system prompts for GPT-4 or Claude, knowledge of chain-of-thought, few-shot prompting, and RAG techniques, ability to write output quality evaluations” — that’s a requirement. Specificity filters out candidates who added “AI” to their CV from those who actually know what chain-of-thought means.
Mistake 2 — Copy-pasting ads from 2023
The role has changed radically. A job ad written for “prompt engineer” two years ago will attract candidates with an outdated skillset — and put off the people you’re actually looking for, because they won’t recognise themselves in a description written before multimodal models and AI agents changed what the job entails.
Mistake 3 — Missing product context
A candidate can’t evaluate whether this is a role for them if they don’t know which model they’ll be working with, in what product, at what scale, and in what team. “Working with the latest AI models” is not context. “Designing system prompts for a customer support assistant handling 50,000 queries per month, integrating with GPT-4o via API, collaborating with a 3-person product team” — that’s context.
Mistake 4 — Not telling candidates how they’ll be measured
A strong candidate wants to know what success looks like. What are the KPIs for this role? Output quality? Time to deployment? Token cost reduction? State it clearly — and you’ll attract people who think in outcomes, not just in tasks.
What every prompt engineer job ad needs:
- A clear signal of the role profile (technical / product / operational)
- Concrete context: model, product or process, scale
- Example tasks from the first 90 days
- Whether the role is standalone or embedded in an AI team
- Expectations around portfolio or prior work examples
How to Evaluate Candidates
This is the most important part of the entire process — and the one where most companies get it wrong, because they apply standard evaluation methods to a non-standard role.
Level 1 — The interview
Five questions that actually verify competence:
“Show me a prompt you’re proud of — and walk me through how you iterated on it.” This question does two things at once: asks for a concrete example and tests the thinking process. A candidate with nothing to show hasn’t been active enough. A candidate who describes only the final version without the iteration history doesn’t understand that prompt engineering is a process, not a one-time creative act.
“How do you measure whether a prompt is working? What’s your definition of success?” This is the best question for separating practitioners from amateurs. A practitioner has a concrete answer: “I build a test set of 50 examples and check accuracy automatically,” “I measure human ratings on a sample of outputs,” “I track the escalation rate to a human agent.” No concrete method means the candidate is operating on instinct, not data.
“When do you change the model instead of fixing the prompt?” This tests depth of understanding of the ecosystem. A prompt engineer who tries to solve every problem by rewriting the prompt doesn’t understand that sometimes the problem is the model, the data, or the system architecture. A good answer includes specific examples of situations where switching models was the right call.
“Describe a time when a prompt stopped working after a model update. What did you do?” Every practitioner has this story — models get updated, behavior shifts, prompts that worked suddenly produce worse results. This question tests resilience to unpredictability and the ability to diagnose the problem. A candidate without this story either hasn’t worked intensively enough with AI, or hasn’t been tracking output quality consistently.
“How do you explain to stakeholders why the model gave a bad output?” The prompt engineer role isn’t just about working with models — it’s also about communicating with people who don’t understand models and have unrealistic expectations of them. A good answer shows the candidate can translate model limitations into plain language and manage expectations without hiding behind technical jargon.
Level 2 — The live task
Set aside 30–45 minutes. Give the candidate a specific, non-trivial business problem and access to a model. Say: “You have 30 minutes. Use any tools you like.”
Don’t evaluate the solution. Evaluate the process:
- Does the candidate start by understanding the problem, or by writing a prompt?
- How do they frame their first query — do they give the model context?
- How do they react when the output is wrong — do they tweak the prompt, change approach, ask the model to explain itself?
- Do they document what they tried and why?
- At the end: ask them to explain the final solution. Do they understand why it works?
Level 3 — Portfolio review
Before the interview, ask for examples of previous prompts with context: what was the problem, what approach did they take, what was the outcome. Format doesn’t matter — a GitHub link, a document, a case description all work.
No portfolio at all is a warning sign. Anyone seriously working in prompt engineering leaves traces — in public repositories, notes, project write-ups. “I did this work but have nothing to show” is possible, but rare.
Salaries and Market Reality
A few hard facts before you start negotiating offers.
Salary ranges for prompt engineers (B2B contracts, 2026):
- Technical profile (LLM integrations, Python, architecture): €3,500–€6,000/month
- Product profile (AI interaction design, evaluations, product collaboration): €2,800–€4,800/month
- Operational / content profile (process optimisation, content workflow): €1,900–€3,500/month
Three caveats that matter more than the numbers themselves:
First, compensation in this role is moving faster than in any other specialisation. Data from six months ago may already be stale. If you’re building an offer based on benchmarks older than a quarter, you risk either overpaying or not reaching the right people at all.
Second, candidates with a documented portfolio and concrete measurable outcomes negotiate well above market ranges. A prompt engineer who can show they reduced operational costs by X% or tripled the speed of a specific process is priced individually — not against a benchmark.
Third, vacancy cost in this role is different from classic positions. A prompt engineer typically unblocks an entire team or AI project. Every week without the right person isn’t just a recruitment cost — it’s the cost of a delayed rollout.
Summary
Hiring a prompt engineer is 90% work done before you write the job ad: defining the profile you actually need, choosing the right sourcing channels, and preparing a verification protocol that separates practitioners from candidates who’ve learned the right vocabulary.
If you know which of the three profiles you’re looking for — you already have an advantage over most companies hiring for this role. If you don’t know — start by answering: what specific problem should this person solve in their first 90 days?
For the verification side of the interview, it’s also worth reading 10 Interview Questions for an AI-Native Developer — the underlying logic of evaluation overlaps significantly between the two roles.
If you’re not sure which profile to start with, or want to sense-check whether your requirements reflect what actually exists in the market — let’s talk. We start every recruitment project with a feasibility study, precisely so we don’t waste time searching for candidates who don’t exist.
