How to write resume bullets that show impact

Resume bullet points show impact when each one states what you did and what changed because of it. The pattern is four parts in one sentence: an action verb, what you built or changed, how you did it, and the result. A bullet with only the first two parts is a task; the last two turn it into evidence.

What makes a resume bullet show impact?

A resume bullet shows impact when a reader can tell what was different after your work. "Worked on the recommendation model" describes presence. "Retrained the recommendation model on session data and replaced the nightly manual refresh with a scheduled job" describes a change someone can ask about.

Hiring managers at Global Capability Centres read bullets as interview material. Each line is a claim they may spend ten minutes on. A bullet that shows impact is therefore specific enough to be questioned and true enough to survive the questions.

Three tests for any bullet:

• The "so what" test. Read the bullet and ask "so what?". If the answer is not in the sentence, add it.
• The swap test. Could any engineer on your team paste the same line into their resume? If yes, it describes the team, not you.
• The question test. What would an interviewer ask next? If you would not enjoy answering, rewrite or remove the line.

What is the action, what, how, result pattern?

The action, what, how, result pattern is a sentence structure that forces each bullet to include evidence. Each part answers one question the reader has.

Put together: "Built a document question-answering service for support staff, using retrieval over policy documents with vector search, so agents answer policy questions without escalating to a specialist."

The order can change. When the result is the strongest part, lead with it: "Stopped a recurring nightly pipeline failure by tracing it to schema drift and adding validation at ingestion." Both versions contain all four parts.

Choose the verb last and choose it for accuracy. "Designed" claims the decisions were yours. "Built" claims most of the code. "Led" claims other people followed your plan. "Contributed to" is honest when your part was small, and is better than a bigger verb you cannot defend.

What do you write when you have no metric?

When you have no metric, write a result that is qualitative but checkable. Most engineers, especially on client projects, never see business numbers, and a made-up percentage is worse than none.

Six kinds of result that need no dashboard:

• Something stopped. A failure, an alert, a class of bug that no longer occurs.
• Something was removed. A manual step, a spreadsheet, a weekly task someone used to do by hand.
• Something was adopted. A team, a service or a set of users now relies on what you built.
• Something became possible. A release, a migration or a feature that was blocked before your work.
• A comparison you made yourself. You timed the job before and after, or you scored two model versions on the same test set. State it plainly and be ready to explain the method.
• Scope. The number of services, pipelines or teams your work covers, if you know it for certain.

How do you write impact bullets for AI and ML work?

Write impact bullets for AI and ML work by stating how the system was evaluated, because evaluation is what separates a working system from a demonstration. A screener for an AI role reads "built a chatbot" on many resumes; a line about how you measured quality is far less common and far more convincing.

Illustrative bullets, by type of work:

• Retrieval and LLM applications: "Built a retrieval pipeline over internal engineering documents and wrote an evaluation set of real questions with expected sources, used to compare chunking and prompt changes before each release."
• Model training: "Trained a gradient-boosted model to flag likely duplicate invoices, validated on a held-out month of data against the existing rule-based check."
• Serving and platform: "Packaged a model as a containerised service with health checks and request logging, so the team could deploy new versions without downtime."
• Data work: "Added schema and null-rate checks at ingestion for the feature tables, catching upstream changes before they reached training."
• Monitoring: "Set up drift monitoring on input features and an alert to the on-call channel, after a silent upstream change degraded predictions."

Each example names a decision an interviewer can probe: why that evaluation set, why that baseline, why those checks. If the terms are new to you, the guide on how to evaluate an LLM application explains what a test set and a baseline look like in practice.

Personal projects follow the same pattern. A project bullet with an evaluation line carries as much weight as a job bullet, which is why AI portfolio projects that get interviews are worth the effort when your paid work has no AI in it.

How do you rewrite a weak bullet step by step?

Rewrite a weak bullet by adding the missing parts one at a time, then cutting anything that is background.

Tightening removes the background and keeps only the four parts.

If most of your lines start as "Responsible for" or "Involved in" because your experience is in client delivery, the guide to rewriting service-company experience on a resume deals with that format and with client confidentiality.

Which mistakes weaken resume bullets?

The mistakes that weaken resume bullets most are vague verbs, tool lists presented as achievements, and team results claimed as personal ones.

• Vague openers. "Worked on", "involved in", "exposure to", "responsible for". Each can be replaced by the actual verb.
• Tool lists. "Used Python, Spark, Kafka and Airflow" says nothing about what was built. Tools belong in the how, inside a sentence.
• Team claims. "Delivered a platform serving the whole organisation" when you built one component. State your component.
• Duties as achievements. Attending stand-ups, following coding standards and meeting deadlines are expectations.
• Unequal weight. More bullets on an old role than on the current one.

How do you check your bullets before applying?

Check your bullets by reading only the first three words of each one down the page. If the column reads as a list of distinct, specific actions, the bullets are working. If it reads "Responsible, Involved, Worked, Responsible", they are not.

Then confirm each bullet fits the role you are applying for. The AI engineer resume guide for GCC roles covers how the bullets sit within the summary, skills and projects sections, and the guide to an ATS-friendly resume covers using the job description's terms inside those bullets.

For feedback on a full draft, the free resume check on The GCC Engineer evaluates a resume for AI roles at GCCs. The free plan includes 5 resume evaluations, shared with job fit checks, so you can test a rewrite against the original.