Prompt Engineer Salary Guide 2026

Last updated: May 17, 2026 | By Tom Kenaley

Prompt engineers in the United States earn a base of $95,000 to $206,000 in 2026, with a national average near $129,500 across public salary aggregators. Frontier-lab packages at Anthropic and OpenAI push total compensation past $500,000 once equity and signing money clear. The headline range is honest. The hiring decision underneath it is the part most teams get wrong.

Tom Kenaley here, co-founder at KORE1. The “prompt engineer” search is the most title-confused role we have priced in two years. Same job board posting, three completely different paychecks, and the gap between what aggregators publish and what offers actually clear has only widened since the late-2025 AI hiring spike at Anthropic, OpenAI, Mistral, and the second-tier labs.

Standard recruiter caveat. KORE1 places AI and ML talent through our AI/ML engineer staffing practice, and we earn a fee when a client hires through us. So when this guide says the most expensive mistake clients make is hiring a $200K prompt engineer when they actually need a $300K AI engineer, take it with that grain of salt. We get paid on either one. The client only pays for one.

What Six Salary Sources Report a Prompt Engineer Earns

The aggregator spread on this title is wider than almost any technical role we benchmark, and the reason traces back to the fact that no two sources are looking at the same population of workers when they say “prompt engineer.” BLS has no SOC code for “prompt engineer” yet, so there is no government anchor. The closest BLS proxy is 15-1252 software developers at $132,270 median, which is a defensible floor for the engineer-grade end of the title but still misses the frontier-lab population entirely. Glassdoor pulls self-reported total pay from 29 anonymous filers, which is a tiny sample and leans toward people who took the time to brag about an offer in writing. Indeed scrapes 62 job postings over a rolling three-year window, which includes a lot of stale 2023 listings from the original prompt engineering hype cycle that never closed. PayScale leans junior. Levels.fyi captures the frontier-lab outliers that none of the others see.

Six sources. Six different sample populations. A $100,000 spread between low and high estimates.

The Glassdoor and ZipRecruiter midpoints line up close to each other near $129K. That number sells you a comfortable story. It is also the wrong number for almost every actual hiring decision we have seen this year, because it averages three different jobs that have stopped sharing a labor market.

Indeed’s $106,770 is dragged down by junior and contract listings, plus the wave of “prompt specialist” content roles at marketing agencies that got reclassified as engineering hires after a search consultant told the agency it would help retention, which is a story we have heard verbatim from three different clients in the past year. Those are not engineers. Those are content strategists with API access, and they read a lot more like the senior copywriter who learned to use Claude really well than the senior software engineer who learned ML alongside Python.

The Levels.fyi top end is real. It is also not what you are hiring against unless your last round closed in the same week as Anthropic’s.

Three Prompt Engineer Archetypes Drive the Spread

Same job title. Three different jobs. Until the hiring side picks which one they actually want, the offer they extend will land too low to close or too high to justify.

The frontier-lab prompt and evaluation engineer. This is the seat at Anthropic, OpenAI, Mistral, Cohere, Adept, Inflection’s remnants, and the well-funded model labs. Base lands $280,000 to $425,000. Total compensation regularly clears $500,000 once RSU grants and signing money settle. The work involves building evaluation harnesses, designing reinforcement learning from human feedback (RLHF) and constitutional AI workflows, red-teaming model behavior, and shipping the prompts that get baked into the model’s training data. It is a deeply technical role. Most hires we have seen at this tier have a graduate degree in machine learning or natural language processing, two to four years of industry ML experience, and a portfolio that mixes published research with hands-on production work. The candidate pool is small. The competing offers are not.

The applied AI engineer at a real-economy company. This is the seat at Snowflake, Databricks, Salesforce Agentforce teams, JPMorgan’s AI platform group, Lilly’s AI/ML org, Insight Partners’ portfolio companies, and the thousands of Series A through D startups now shipping LLM-backed product. Base sits $135,000 to $220,000 for mid-level and $190,000 to $275,000 for senior. The job is mostly retrieval-augmented generation (RAG) architecture, evaluation pipelines, agent orchestration with LangGraph or CrewAI or just hand-rolled state machines, vector database selection between Pinecone, Weaviate, and pgvector, and the unglamorous work of figuring out why a prompt that worked in the playground does not survive contact with production traffic. This is where the volume is. This is also where the title is moving away from “prompt engineer” and into “AI engineer” or “applied AI engineer” because the work has gotten too broad to fit under prompt-only framing.

The content and marketing AI specialist. Different job. Different paycheck. Base lands $75,000 to $120,000, occasionally to $140,000 at brand-name agencies in Manhattan. The work is mostly applied use of Claude, ChatGPT, Midjourney, and the rest of the consumer-facing model toolbox, plus building prompt libraries for the marketing org, training internal users, and producing AI-assisted content at volume. Real work. Real value. Not the same hire as either of the above, and a recurring source of compensation mispricing when a recruiter or hiring committee blurs the categories. We have placed exactly four of these this cycle and they read more like content operations leads than engineers.

Three roles. Three offer letters. Pick one before the JD goes out.

Salary by Experience Level

The level-banding below reflects the middle archetype, which is the one most KORE1 clients are actually hiring for. Frontier-lab numbers sit above the band. Marketing-AI roles sit below it. Both are real. Neither belongs in the same comparison.

The jump from mid to senior is steeper than what most aggregator views suggest, mostly because real senior candidates can ship production LLM systems end-to-end and the candidates who only claim that capability fall apart inside the first technical interview when they have to design an evaluation set for a multi-step agent under live constraints. The mid-level pool is much larger. Most candidates who claim senior land closer to upper-mid once you stress-test the resume against an actual eval design exercise.

Staff and principal is where the comp math starts looking like a generic senior ML engineer search, with a 15% to 20% premium for the prompt-engineering specialization on top, and the reason most companies underpay this seat is that the internal salary bands were calibrated against a 2023 view of the role that has not been refreshed since the foundation-model providers reset the entire skill ladder twice. Companies underpay this seat regularly. The market clearing rate is well above what most internal salary bands have been updated for.

What Prompt Engineers Earn at Anthropic, OpenAI, and Frontier Labs

The frontier-lab tier deserves its own section because the numbers are real but almost no one is hiring against them, and conflating “what an engineer makes at Anthropic” with “what we have to pay our first prompt engineering hire” has burned the budget on at least three searches we ran in the last six months alone.

According to Levels.fyi, Anthropic software engineers earn a median total compensation of $710,000, with base salaries running $300,000 to $425,000 and the balance in equity tranches that vest over four years with a one-year cliff in the standard package. OpenAI’s range is wider because the levels span L2 through L6 publicly, and the company has not standardized its compensation policy in the same disciplined way the older hyperscalers did. An L5 software engineer there reportedly takes home $1.15M total comp, with a $336K base and $774K in stock per year. These are software engineering levels rather than prompt-specific titles, but at both companies the most senior prompt and evaluation work is done by engineers at exactly those levels.

The smaller labs run 20% to 30% below those numbers on base, and the equity story varies wildly with stage, vintage, and the specific terms a candidate is offered on accelerated vesting or single-trigger acceleration during the next funding round. Mistral, Cohere, and the well-funded foundation-model startups still compete aggressively for prompt and eval talent. Series B and C model-application companies in San Francisco and New York will quote $250,000 to $320,000 base for senior prompt engineers and call it a market rate, which is honest market data for the population of companies they are actually competing against but which dramatically misreads the offer math at the foundation-model layer one tier above them. It is market for them. It is not market for the foundation-model layer.

One thing worth saying out loud. The candidates we have submitted into these searches do not come from a “prompt engineering bootcamp.” They come from machine learning, NLP research, applied data science, and software engineering backgrounds with a documented track record of shipping LLM-backed systems. The bootcamp graduates clear the marketing-AI tier. They do not clear the frontier-lab tier. The labor markets are not connected.

Salary by Metro

Geography still matters, even with remote work. The clustering of frontier labs, model-application startups, and well-funded enterprise AI teams pulls the local salary band up structurally in five metros and leaves most of the country closer to the national average.

Across our IT staffing placements in the past 12 months, the Bay Area premium has narrowed slightly against New York and Seattle but widened against everywhere else, mostly because the equity component is meaningful at the Series B+ companies in SF and almost nonexistent at the same-stage companies in second-tier metros where investor risk appetite for paying SF-style equity has cooled noticeably since the 2024 funding contraction. The reason matters when a candidate is comparing offers across geographies, because base alone undersells the SF package and overstates the offer from an Atlanta or Raleigh team.

Remote pricing has stabilized around the New York mid-level number minus 5%. The “remote means San Francisco rates” era ended in late 2024 for most non-frontier companies, and the “remote means Indianapolis rates” attempt ended faster, because the candidates who took those offers were the ones who could not get a competing offer anywhere else and turned out to underperform on the team measurably within their first six months. Most clients land in the middle.

The Hiring Question Most Teams Get Wrong

Here is the contrarian recommendation. Most companies that post a “prompt engineer” requisition would be better served by an AI engineer or applied ML engineer with prompt-design competence. The skills overlap. The salary band is more honest. The candidate pool is bigger by a factor of five, with comp ranges documented in our AI Engineer salary guide.

Three reasons.

First, prompt engineering as a standalone discipline has been folded into the broader AI engineering toolkit since mid-2024. The work that used to be “design a prompt” is now “design an eval set, build a retrieval index, write the orchestration code, ship the agent, monitor for regression.” Hiring for prompts in isolation produces a candidate who can do step one and stalls at step two.

Second, the talent supply curve for “prompt engineer” titles is bimodal. The top of the curve is competing with Anthropic. The bottom is competing with content agencies. There is almost no middle. Search for “AI engineer” and the middle reappears, full of people who can ship.

Third, our placement data backs it. Across 47 AI-adjacent searches at KORE1 in the last two quarters, the requisitions written as “AI Engineer” closed in an average of 19 days, while the ones written as “Prompt Engineer” closed in 38 days, often after the JD was rewritten partway through and the salary band was quietly raised by $25,000 on the senior end to attract the same pool of candidates the original posting had already failed to reach. Same money. Same role. Twice the time to fill.

If you are at a frontier lab, ignore everything in this section. You know what you are hiring for. If you are anywhere else, please consider rewriting the title before you post.

The Skills That Move the Number

Across our last 47 AI-adjacent placements, four skill clusters predict where a candidate lands inside the band. Listed in order of dollar impact.

Production LLM systems experience. Has the candidate actually shipped an LLM-backed product to real users? Built and maintained an eval framework using Braintrust, Langsmith, Promptfoo, or a hand-rolled equivalent? Monitored drift in production? This is worth $30,000 to $50,000 on base over a candidate who has only built prototypes. Almost nothing else moves the number as much.

Retrieval and vector database depth. RAG architecture is the table-stakes pattern. Real depth across Pinecone, Weaviate, pgvector, Chroma, plus an opinion on hybrid retrieval and reranking with Cohere or BGE models. Worth $15,000 to $30,000 over a candidate with cursory exposure.

Agent and orchestration framework familiarity. LangGraph, LlamaIndex Agent, CrewAI, OpenAI Agents SDK, plus a healthy skepticism about when not to reach for a framework at all. Most of the production work in 2026 is multi-step agents with tool use, and candidates who have built and debugged these earn the premium.

Fine-tuning and RLHF exposure. The premium here has moderated since the foundation-model providers shipped better off-the-shelf models. Still worth $10,000 to $20,000. At the frontier-lab tier it is non-negotiable.

One signal that does not move the number much. Certifications. The various “prompt engineering certified professional” credentials from third-party training shops do not show up as a positive in offer modeling. They do not hurt either. They are just neutral.

Common Questions Hiring Managers Actually Ask

Is “prompt engineer” still a real job title in 2026?

Yes, but it’s narrowing. Frontier AI labs still hire prompt and evaluation engineers as a distinct role. Almost everywhere else, the work has merged into “AI engineer” or “applied AI engineer” titles where prompt design is one skill among several. About 60% of our 2026 placements written as prompt engineer requisitions were retitled to AI engineer before closing, and they filled faster after the change.

How much do prompt engineers at Anthropic and OpenAI actually take home?

Total compensation at Anthropic and OpenAI for prompt and evaluation engineers runs $500,000 to $1.2M, with base salaries between $300,000 and $425,000 and the balance in equity and signing bonuses. The number depends heavily on level and equity vintage. Pre-IPO equity at OpenAI carries a markedly different expected value than RSUs at a public AI hyperscaler, and recruiter calls should price both scenarios when comparing offers.

Do prompt engineers need a degree?

For frontier-lab roles, usually a graduate degree in machine learning, NLP, or computer science is expected, plus published research or a documented production track record. For applied AI engineering roles at enterprises and startups, a bachelor’s degree in any STEM field combined with a real production LLM portfolio is sufficient. We have placed bootcamp graduates into the latter tier, but never into the former.

Will prompt engineering be replaced by AI itself?

Not the role. The narrow skill of writing one-off prompts is already being automated by tools like DSPy, Promptbreeder, and automated eval-driven prompt optimization frameworks that effectively run gradient descent over a prompt template against a labeled dataset, which is genuinely a better way to optimize a single prompt than any human can do by hand. The broader role of designing, evaluating, and iterating LLM systems is growing, because the systems themselves are growing. The skill is migrating from “writing prompts” toward “designing the system that automates prompt writing.” That work pays more, not less.

How long does a typical prompt engineer search take to close?

Our average time-to-hire across all KORE1 IT placements is 17 days, and applied AI engineer searches sit close to that mark when the JD is written cleanly. Prompt engineer requisitions take roughly twice as long, mostly because the title attracts an overly broad candidate pool that has to be filtered manually. The retention rate on placements we close runs at 92% over the first 12 months, including the AI/ML segment.

Should I open the search as contract, contract-to-hire, or direct hire?

Direct hire if the role is core to your AI roadmap and the team will be there in a year. Contract-to-hire works well for first AI engineer hires where the team is still figuring out what it actually needs. Contract-to-hire is also the right call if the budget exists but the headcount approval has not landed yet, which is a more common scenario this year than at any point we can remember.

What’s the realistic time to hire?

19 days is our placed-base average for AI engineer requisitions across the last two quarters, and 38 days for the same role written as prompt engineer. The variance comes from JD clarity more than candidate scarcity. If you want a faster search, the single highest-impact move is rewriting the title and tightening the must-haves down to three. When you’re ready, you can reach out to our team to scope the search.

A Plain Read on the Market

Prompt engineer is a real title with a real pay band, and the role is shifting under it faster than most internal compensation reviews are catching up to. The $129,000 average from public aggregators is honest for the middle archetype and misleading for the other two. Frontier labs pay multiples of that and hire a different population. Marketing-AI specialists are not engineers and should not be priced as such.

The cleanest hiring move for most teams in 2026 is to rewrite the requisition as an AI engineer search with prompt and evaluation skills called out as required. The candidate pool widens. The salary band becomes more defensible. The role keeps the same scope.

If you want to compare your internal banding against current market data, our salary benchmark assistant covers AI engineering roles across the same metros as this guide. If you want a faster path, talk to one of our recruiters and skip the calibration work. For a full hiring playbook covering JD structure, screening, and offer construction, see our how to hire prompt engineers guide.

About the author: Tom Kenaley is co-founder of KORE1, an IT and engineering staffing firm operating across 30+ U.S. metros since 2005. KORE1 specializes in placing AI/ML engineers, software engineers, and data scientists for venture-backed startups, mid-market technology companies, and Fortune 500 enterprises.