Career change resume for AI roles: what transfers

What should a career change resume for AI roles show?

It should show work you've already done that an AI posting asks for, written in that posting's words, with figures from your own records. A hiring manager should see the overlap within one bullet. It shouldn't use enthusiasm in place of experience, or attach AI vocabulary to work that never involved AI.

Plenty of jobs at AI companies sit outside research and engineering. On October 27, 2026, Anthropic's Data Operations Manager, Human Data posting asked for "3+ years in operations, consulting, product management, or program management roles." It also asked that you be "familiar with how LLMs work or have strong interest in understanding AI training methodologies." Neither line needs an AI job title.

The same posting says "We encourage you to apply even if you do not believe you meet every single qualification." That line invites near matches to apply, and it doesn't promise them an interview. The page on when to apply with missing required skills covers how close is close enough.

Which AI jobs list experience from outside AI?

Operations, customer success, enablement and support roles at AI companies often do. Four postings at Anthropic and OpenAI, checked on October 27, 2026, asked for years in customer success, operations, consulting or instruction. They treated AI knowledge as fluency to show or interest to bring. Research and engineering roles ask for more, and the gap there is wider.

• Anthropic's Customer Success Manager, Commercial posting asks for "4+ years of experience in Customer Success, Technical Account Management, or Solutions Engineering."
• OpenAI's Data Program Manager, Human Data posting says an interest or background in AI is "not required," and asks ideally for "at least 1-2 years of relevant experience."
• OpenAI's Enablement Lead, Government posting asks for "6+ years of experience in customer-facing, enablement, instructional, adoption, or transformation roles."
• Anthropic's data operations posting asks for operations, consulting, product or program management, as quoted above.

Each one also asks for something an old job may not show. The customer success posting wants "Technical fluency with ability to understand and articulate AI/ML concepts." Your old job may never have asked for that. Mark a line like that as a gap before you write a word.

Which of your existing facts transfer to AI roles?

The facts that transfer show the same work under a different name: writing instructions, checking quality, training people, managing vendors, keeping accounts and analyzing data. The table pairs made-up old-role facts with requirements quoted from real AI postings, checked October 27, 2026. Read across each row, then mark your own facts the same way.

Every old-role fact in the table is an example written for this page, about made-up people. The requirements are quoted from the postings as they read on October 27, 2026, and postings change or close. Reread the posting on the day you apply.

Look at the Mark column before the wording. The support lead's row is partial because the AI Support Engineer posting asks for more. It wants AI tools used "to script or engineer code (e.g., Python) for automating repetitive tasks." The last row is missing for most career changers, and it stays off the resume until a project shows it.

If you can't name your old work in plain words, O*NET can help. The US Department of Labor's O*NET page for training and development specialists lists the work as a task. It reads "Obtain, organize, or develop training procedure manuals, guides, or course materials." A shift lead who wrote the training binder did that task. Its wording sits closer to an enablement posting than "made the binder" does.

How do you rewrite an old-role fact in the posting's words?

Keep the fact and change the label. Start from the figure in your record, say what it measured, and describe the work with the posting's verb where your work matches it. The AI context belongs to the new job. Your bullet describes the old job, so it can't claim AI work that never happened there.

Here is one pair, as an example. The person and every figure are made up for this page. The target is a human data program role that asks you to "write instructions" and "define success criteria."

• Before: "AI-ready operations leader who drove a 50% improvement in data quality through AI-first thinking."
• After: "Wrote pick instructions and pass criteria for 30 seasonal staff and audited 5% of picks weekly; mis-picks fell from 2.1% to 1.4% in Q4 2025."

The number that changed is the 50%. Nobody measured "data quality," and no record held that figure.

The after line passes three questions. The figure is 2.1% to 1.4%, and it measured the share of orders picked wrong in Q3 against Q4 2025. It came from the warehouse system's monthly error report.

That drop is about a third in relative terms, a real result and a smaller one than the before line claimed. The page on how to turn your resume into facts walks through the three questions on a whole resume. If your old job never counted anything, start with what to do when your resume has no numbers.

Should AI courses, certificates and side projects go on the resume?

Yes, in their own section and at their real size. A course goes under training or education, and it shows you studied a topic. A project you built goes under projects, with what it does and who used it. Neither belongs under a past job, and neither turns a missing requirement into an evidenced one.

Some AI postings ask for exactly this kind of proof. Anthropic's Developer Education Lead posting was checked on October 27, 2026. It prefers "A portfolio you can show us: labs, courses, talks, or repos you built yourself." Its minimum list asks for hands-on experience building with LLM APIs and agents, "beyond prompting alone."

That last phrase matters if you're changing careers. Using a chat assistant at work is a fact, and a small one. For example, "Used ChatGPT to draft weekly client summaries for 12 accounts" is fine if your employer allowed it. Calling that "built AI workflows" is a bigger claim than the record supports.

A project counts once it leaves a record. For example, a teacher who built a small quiz generator on a model's API can link the repository and count the users. "Built a quiz generator on an LLM API, used by 2 classes of 26 students for 4 weeks" is a project line with a source.

What about AI terms you've only read about?

Leave them off until a fact shows them. AI postings carry terms such as RLHF, evals and agents, and copying them into a skills list may match a search and then fail the interview. Use a posting's term only inside a sentence about work you did. Reading about a method is study, and it goes under training.

Anthropic's data operations posting lists "Knowledge of RLHF, constitutional AI, or human-in-the-loop workflows" under what strong candidates "may also have." That is a preferred item, so a gap there costs less than a gap in the main list. The page on how to tailor a resume to a job description without lying shows how to mark each item before you write.

Employers in this field have said where their line sits. Anthropic's guidance on candidates' AI usage, last updated July 10, 2025, says "We want to see your real experience." A resume rewritten by a tool can pick up AI vocabulary you never used. The page on why AI resume tools invent numbers explains how a rewrite adds claims nobody checked.

Do you need a technical degree to move into AI?

For the six postings in the table, no. The three Anthropic postings accept "an equivalent combination of education, training, and/or experience," and OpenAI's support posting accepts "equivalent practical experience." The two other OpenAI postings name no degree. Data science and research roles are different, and government data shows a typical degree path there.

The US Bureau of Labor Statistics page on data scientists was last modified August 27, 2026. It says they "typically need at least a bachelor's degree in mathematics, statistics, computer science, or a related field." It projects employment to grow 35 percent from 2025 to 2035. That projection covers the occupation, and it says nothing about any one applicant's chances.

O*NET's task list for data scientists shows why the bar sits there. It includes "Test, validate, and reformulate models to ensure accurate prediction of outcomes of interest" and "Write new functions or applications in programming languages to conduct analyses." If your record shows those tasks, they transfer directly. If it doesn't, the operations and customer roles above sit closer to the facts you already have.

How do you tell the career change story the same way in every round?

Write two sentences once: why you're moving, and which of your facts the new role uses. Say them in every interview, with the same figures as your resume. A career change invites the question "why now," and an answer that shifts between rounds sounds like two different stories from one person.

For example: "I wrote the instructions and checks that 30 seasonal staff worked from, and mis-picks fell from 2.1% to 1.4%. Human data programs run on that same work, and I want to do it where the quality bar shapes how a model behaves." Every figure in it matches the after line above.

Interviewers will test the gap too. Name it once, say what you're doing about it, and use the same words each time. The page on consistent answers across interviews covers how a shifting answer gets noticed, and a fact-locked resume keeps your figures from drifting between versions.

After each round, write down how you explained the move and which facts came up, using the free 10-minute debrief template. Prept's job match marks each requirement in a posting as evidenced, partial or missing against facts you confirmed, which is the table above run on your own record.