How to read an AI job description
How to read a job description for an AI role: separate must-haves from nice-to-haves, decode the level and the stack, and decide when to apply anyway.
On this page
To read a job description for an AI role properly, read the responsibilities first, sort the requirements into must-haves and nice-to-haves, and work out the level and the real role type from the wording. Learning how to read a job description this way takes ten minutes per posting and tells you three things: whether to apply, what to put at the top of your resume, and what the interviews are likely to cover.
What are the parts of a job description, and which matter most?
A job description usually has five parts, and the responsibilities section matters most because it describes the work you would do. The requirements section is often written last, copied from older postings, and padded with preferences.
| Section | What it tells you | How much weight to give it |
|---|---|---|
| Title | A rough label; inconsistent between companies | Low |
| About the team | The product or business area the role serves | Medium |
| Responsibilities | What you will actually do each week | Highest |
| Required qualifications | The filter the recruiter applies | High |
| Preferred qualifications | What would make a candidate stand out | Low to medium |
Read the responsibilities and ask: could I start doing most of this within a few months, and have I done something like it before? If the answer is yes, the requirements become a question of evidence, not eligibility.
How do you tell must-haves from nice-to-haves?
You tell must-haves from nice-to-haves by position, repetition, wording and connection to the responsibilities. A must-have typically shows at least two of these four signals.
- Position. Requirements listed first usually matter most. Writers start with what they care about.
- Repetition. A skill that appears in the title, the responsibilities and the requirements is central. A skill named once in a long list is not.
- Wording. "Required", "must have", "strong", "hands-on" and "proven" mark must-haves. "Preferred", "a plus", "familiarity with", "exposure to" and "nice to have" mark extras.
- Connection. A requirement that maps to a main responsibility is real. If the responsibilities say "build and evaluate retrieval pipelines", then retrieval and evaluation are must-haves wherever they appear in the list.
Take a representative posting that asks for "strong Python; hands-on experience building LLM applications; experience with RAG and evaluation; familiarity with a major cloud platform; exposure to Kubernetes is a plus; publications preferred". The must-haves are Python, LLM applications, and RAG with evaluation. Cloud is a soft requirement. Kubernetes and publications are extras, and missing them should not stop you applying.
How do you decode the level of a role?
Decode the level of a role from the verbs, the scope of the work and the experience range, in that order. Titles alone mislead, because "senior" means different things at different companies.
Entry-level and junior postings use verbs like "assist", "support", "contribute to" and "implement", and describe work on parts of a system under guidance. Mid-level postings use "build", "own", "design" and "deliver", and describe whole features or pipelines. Senior and lead postings use "lead", "define", "drive", "architect" and "mentor", and describe direction across teams and work with senior stakeholders.
The years-of-experience line is a guide to the pay band and seniority the company has in mind more than a strict gate. Being slightly under the range is usually worth an application if your work matches the verbs. Being far under it usually is not, because the role is budgeted and levelled for someone more senior.
How do you decode the stack and the real role type?
Decode the real role type from the stack and the responsibilities, because the same title covers different jobs. The tools named in a posting are the clearest evidence of what the day-to-day work is.
| If the posting stresses | The role is probably | Read next |
|---|---|---|
| Model training, feature engineering, model serving | ML Engineer | ML engineer role guide |
| LLM applications, RAG, prompts, agents, evaluation | GenAI / LLM Engineer | GenAI engineer role guide |
| Pipelines, deployment, CI/CD, monitoring, infrastructure | AI Platform / MLOps | MLOps role guide |
| SQL, data pipelines, warehouses, data quality | AI Data Engineer | AI data engineer role guide |
| Experiment design, statistics, research, publications | Applied Scientist | Applied scientist role guide |
A posting titled "AI Engineer" that lists APIs, microservices and one line about "integrating AI models" is mostly a backend role. One titled "Data Scientist" that lists retrieval pipelines and prompt evaluation is a GenAI engineering role. Compare what you find with the descriptions in AI engineering roles at GCCs or the role explorer, and see the ML engineer and GenAI / LLM engineer role guides for the two most common types.
The stack also predicts the interview. Skills named in the responsibilities are the ones the technical rounds will test, as explained in the GCC AI interview process.
Should you apply if you do not meet every requirement?
You should apply if you meet the core must-haves and can show evidence for each, even when you miss some of the remaining requirements. Job descriptions describe an ideal candidate, and hiring managers routinely interview people who match the central needs and can learn the rest.
Apply anyway when
- You meet the three or four must-haves tied to the responsibilities
- The gaps are tools you could learn in weeks
- You are slightly under the experience range but have done the work
- The missing items are marked preferred or a plus
- You have a project that proves the central skill
Skip, or wait, when
- The main responsibility is something you have never done
- The role is two levels above your current one
- A hard constraint does not fit: location, shift, work authorisation
- You would need to overstate your experience to pass the screen
- The role type is not the one you are aiming for
Applying anyway works best with a referral, because a person who can vouch for you helps a recruiter look past a missing keyword. How to get one is covered in GCC referrals.
How do you use a job description once you decide to apply?
Use the job description as the brief for your resume and your preparation. Once you decide to apply, the posting tells you exactly what to prove.
Highlight the responsibilities
Mark the three or four that describe the core of the job.
List the must-haves
Use position, repetition, wording and connection to pick them out.
Match evidence to each
For every must-have, name a project or result that proves it. Note the ones with no evidence.
Tailor your resume
Move the matching evidence to the top and use the posting's own terms where they are true for you.
Note the gaps
Prepare an honest line for each gap, and start closing the ones that repeat across postings.
Using the posting's exact terms matters because automated screening and busy recruiters both look for them; the mechanics are in writing an ATS-friendly resume for GCC roles. Keep every claim true. A keyword you cannot defend becomes a problem in the first technical round.
Reading several postings for the same role is also the fastest way to plan your learning. Collect ten postings, count which must-haves repeat, and that list is your syllabus. The free readiness check and roadmap on The GCC Engineer does a similar mapping between your current skills and your target role.
What warning signs should you look for in a job description?
Look for warning signs that the role is unclear or is not what the title suggests. None of them rules a job out, but each is worth a question to the recruiter.
- Responsibilities that span three jobs. A single role that covers research, data engineering, backend and stakeholder management may mean the team has not decided what it needs.
- A long tool list with no responsibilities. The posting may be a generic template; ask what the team is building.
- "AI" in the title and none in the duties. The role may be general software or support work with a fashionable label.
- No mention of the team or product. Ask which business area the role serves and who it reports to.
- A service or staffing arrangement described as a GCC role. Check who the employer is. In a Global Capability Centre, the employer is the enterprise itself, not a vendor.
Reading postings well is one skill in a larger process, set out step by step in how to get a GCC job. To choose which companies' postings to watch, see how to build a GCC target company list.
Key takeaways
- Responsibilities tell you the real job; the requirements list is often a wish list.
- Must-haves appear early, are repeated, and are tied to the main responsibilities.
- Verbs and scope reveal the level: 'assist' and 'contribute' differ from 'lead' and 'define'.
- The stack tells you the true role type, whatever the title says.
- Apply when you meet the core must-haves with evidence; skip when the core work is something you have never done.
Frequently asked questions
How many of the listed requirements should you meet before applying?
There is no fixed share. A better test is whether you meet the core must-haves: the three or four requirements tied directly to the main responsibilities. If you can show evidence for those, apply, even if several preferred items are missing. If the main responsibility is something you have never done in any form, a long list of matched extras will not carry you.
What does 'or equivalent experience' mean in a job description?
'Or equivalent experience' means the employer will accept demonstrated ability in place of the stated qualification. For a degree requirement, relevant work experience and projects can count. Make the equivalence easy to see: put the relevant experience and its results at the top of your resume instead of expecting the reader to infer it.
Why do job descriptions list so many tools?
Job descriptions are often assembled from older postings and several people's wish lists, so tool lists grow long. Few teams use every tool named. Look for the tools that also appear in the responsibilities or are named more than once; those are in daily use. A list separated by 'or' and 'such as' is asking for a category, not each item.
Should you apply to the same company for several roles at once?
Apply to the one or two roles that fit you best, not to every opening. Recruiters can usually see all your applications, and a scatter of unrelated roles suggests you have not decided what you want. If two roles are close, apply to both with a resume that fits each, and tell the recruiter which you prefer.
What if the job title and the job description do not match?
Trust the description over the title. Titles are inconsistent between companies, and one company's 'AI Engineer' is another's 'Data Scientist' or 'ML Engineer'. Read the responsibilities and the stack to decide what the job really is, then confirm with the recruiter in the first call by asking what a typical week involves.