The 40 ATS Keywords Data Scientist Resumes Need in 2026 (And 5 That Get You Filtered Out)

Every data scientist resume I review has the same problem. The candidate has shipped real models, moved real revenue, and still cannot get a phone screen. Then I run their resume against the job description and the keyword match is 38 percent. The parser is filtering them out before a human ever touches the file.

ATS parsers do not know that tree-based models and XGBoost are related. They do not infer that your experimentation framework work means you understand A/B testing. They match strings. If the posting says Python and your resume says python3, that might still work. If the posting says PyTorch and you wrote Pytorch, most parsers forgive it. But if the posting says Airflow and your resume says 'orchestration tool,' you lose the point. For more on this, see how to score 90+ on any ATS.

I tested 200 data scientist job postings across Greenhouse, Lever, Workday, and Ashby in Q1 2026 to find the keywords that actually show up in DS roles at Series B through Fortune 100 companies. Here is the list, ranked by frequency, split into must-have, nice-to-have, and buzzwords that actively hurt you.

Who This Guide Is For

Data scientists targeting 2026 DS, ML Engineer, Applied Scientist, or Analytics Engineer roles. If you are doing pure research science at a research lab this list is different. Everyone else, this is the one.

The Must-Have, Nice-to-Have, and Avoid Table

The 40 Keywords Your DS Resume Actually Needs

Grouped by category. Hit the relevant ones for your target JD. You do not need every single one on a single resume. You need the ones the specific posting asks for.

Core Programming (Must-Have in 90%+ Postings)

1. Python (spell it exactly, not Python3 or Python 3.11 only)
2. SQL (uppercase, standalone keyword)
3. Git (version control)
4. Jupyter (or Jupyter Notebook)

Data Manipulation Libraries

5. pandas
6. NumPy
7. Polars (rising, 2026 addition)

ML Frameworks

8. scikit-learn
9. PyTorch (dominant in 2026, use correct spelling)
10. TensorFlow (still requested in enterprise postings)
11. XGBoost (gradient boosting, still the workhorse)
12. LightGBM
13. Hugging Face (for NLP and LLM roles)

Data Infrastructure

14. Spark (or PySpark)
15. Airflow
16. dbt (data build tool)
17. Snowflake
18. BigQuery
19. Databricks
20. Redshift (AWS shops)

Cloud Platforms

21. AWS (add SageMaker, S3, Lambda specifics when relevant)
22. GCP (Google Cloud Platform)
23. Azure

Methods and Techniques

24. A/B testing
25. Experimentation
26. Feature engineering
27. Regression (logistic, linear)
28. Classification
29. Clustering
30. Causal inference
31. Time series
32. SHAP (model interpretability)
33. Bayesian

2026 LLM and GenAI Entries (New)

34. LLM (plus 'large language model' spelled out once)
35. RAG (retrieval-augmented generation)
36. Vector database (Pinecone, Weaviate, pgvector all count)
37. Embeddings
38. Evals (LLM evaluation, the 2026 buzzword that is actually real)
39. Fine-tuning
40. LangChain or LlamaIndex

That is the 40. Match them to the specific JD. If the JD mentions 7 of these, hit all 7. If the JD mentions 25, hit 25. You do not need to stuff every single one into your resume. You need to mirror the posting.

The 5 Buzzwords That Get DS Resumes Filtered Out

This is the part most keyword guides skip. Not every keyword helps. Some trigger negative signals on modern parsers and hiring managers that read LinkedIn on a second monitor. For more on this, see how ATS scoring works.

1. 'Big data'

This phrase went out of style around 2019. Using it in 2026 signals you learned DS from a 2015 Coursera course and never updated. Modern postings say 'large-scale data' or just name the tool (Spark, Hadoop, etc.). Replace with: Spark, Databricks, petabyte-scale, or the specific volume you handled.

2. 'AI ninja' or 'ML rockstar'

If you wrote this, you are telling the recruiter you do not know how to talk about your work. Nobody serious uses these phrases. They are a tell for a puffed-up resume with no real substance. Modern semantic parsers flag these phrases as low-quality-language signal.

3. 'Leveraged advanced analytics to drive business outcomes'

This says nothing. What did you build? What was the outcome in dollars or percent? Replace with the specific model, the specific metric it moved, and the specific number. 'Built XGBoost churn model that reduced monthly churn by 14%' beats ten bullets of 'leveraged advanced analytics.'

4. Every single framework you ever touched

A skill bar that lists 'Python, R, Julia, Scala, Java, C++, Go, Rust, SAS, SPSS, Stata, MATLAB' is a red flag. Recruiters assume you used one of these for a week in a class. Senior DS hiring managers will ask about the weakest skill on your list in an interview. Do not pad.

5. Any keyword you cannot defend in a 5-minute interview

The fastest way to fail a phone screen is to list a technology you touched once and get asked a specific question about it. If you cannot explain what a gradient does in XGBoost, do not put XGBoost on your resume. If you cannot write a window function in SQL, do not claim advanced SQL. This is where the 'keyword stuffing' meta falls apart. The parser gets you in. The human reads your claim. Then asks. Then you are done. For more on this, see pull ATS keywords from a JD.

How to Actually Place These Keywords

Placement matters almost as much as presence. Parsers weight keywords differently depending on section. Here is the hierarchy:.

1. Job titles and role summaries carry the most weight. If the JD says 'Senior Machine Learning Engineer' and your current title is 'Data Scientist II,' consider a functional title like 'Data Scientist (Machine Learning Engineer function)' when accurate.
2. Bullet points with metrics come next. Place technical keywords inside bullets with outcomes. 'Trained PyTorch transformer model that improved recommendation CTR by 23%.' The parser sees PyTorch and transformer and CTR and the number.
3. Skills section last. A skills section at the bottom gives you coverage for keywords you could not fit into bullets without it feeling forced. Do not treat it as your first-class keyword container.

A keyword in a bullet with a metric is worth five keywords in a skills list.

The Match-To-JD Approach

You cannot hit 40 keywords on one resume without it reading like spam. You do not have to. The trick is matching to the specific JD.

Take the JD you are targeting. Copy it. Strip out the boilerplate legal stuff. What is left is a list of keywords the ATS will score you against. Hit those. You are aiming for roughly 70-85% keyword match for a strong score. Over 85% and you start looking like you keyword-stuffed, which modern parsers increasingly detect.

This is the exact job AI Applyd's scorer does for you. Paste the JD, upload your resume, get the gap report back in 30 seconds. You see exactly which keywords the JD expects that your resume is missing. Fix the gaps, re-score, apply.

Role-Specific Adjustments

Not all DS roles are the same. The keyword mix shifts depending on subspecialty:. For more on this, see win the 6-second recruiter scan.

• Product DS / Analytics DS: Emphasize A/B testing, experimentation, SQL, dbt, causal inference, stakeholder communication. De-emphasize deep learning unless the product ships ML.
• ML Engineer / Applied Scientist: Emphasize PyTorch, TensorFlow, production deployment, ML infrastructure, model monitoring, feature stores, vector databases.
• LLM / GenAI Engineer: Emphasize RAG, evals, prompt engineering, fine-tuning, LangChain/LlamaIndex, vector databases, embeddings. This role barely existed as a dedicated title in 2023. It is a top-ten hiring category in 2026.
• Analytics Engineer: Emphasize dbt, SQL, Snowflake, data modeling, warehousing. Soft on ML framework keywords.

The Bottom Line

Your DS resume does not need to hit every one of these 40 keywords. It needs to hit the ones the specific JD asks for, with real metrics attached, in the right sections. Skip the filler phrases. Drop the 2015 buzzwords. Do not list technologies you cannot defend.

The parser is not your enemy. It is a blunt instrument. Treat it that way. Give it the exact strings it is scanning for. Then let the human see your actual work underneath.

Score your DS resume free or compare AI Applyd plans.

Frequently Asked Questions

What are the most important ATS keywords for a data scientist resume in 2026?

Python, SQL, pandas, NumPy, scikit-learn, PyTorch, and Git appear in over 80% of DS postings and are non-negotiable must-haves. Next tier includes Spark, Airflow, dbt, Snowflake, BigQuery, and Databricks which appear in 60-70% of postings. New 2026 additions include LLM, RAG, vector database, embeddings, and evals.

Should I add LLM and RAG keywords if I have not used them in production?

Only if you can defend them in an interview. If you built a side project with RAG and can explain chunking strategy, embedding models, and retrieval eval, yes. If you read one blog post, no. The parser gets you in, then a human asks specific questions. Listing keywords you cannot defend gets you rejected in the phone screen.

How many keywords should my DS resume have?

Aim for a 70-85% match against the specific job description you are targeting, not against a generic DS keyword list. Over 85% match can trigger keyword-stuffing detection on modern semantic parsers. Under 50% and you will likely be filtered before a human sees the resume.

Does using 'Pytorch' instead of 'PyTorch' hurt my ATS score?

Most modern parsers are case-insensitive, so 'Pytorch' and 'PyTorch' will match. Older parsers like Taleo may be case-sensitive. Use the exact capitalization the JD uses to be safe. PyTorch is the canonical capitalization.

Where should I place keywords on my DS resume?

Job titles and role summaries carry the most weight, followed by achievement bullets with metrics, with a skills section at the bottom for coverage. A keyword in a bullet with a metric is worth far more than the same keyword in a bare skills list. Place technical keywords where they appear in the context of outcomes.

How do I check my DS resume against an actual job description?

AI Applyd scores your resume against each specific DS job description in about 30 seconds and flags exactly which keywords are missing, which ones match, and what the overall compatibility score is. Free tier includes 10 scores per month so you can test the gap-to-JD approach before paying.