We collected the postings on August 28, 2026, and classified them in two runs. The first run covered 15 roles. The second run re-read the four data roles with a stricter method. The blog post draws on both.
Key numbers: the four data roles
The four data roles were read a second time on September 7, 2026. A posting counts when it asks the candidate for the skill. Company marketing and hiring boilerplate do not count. Each denominator is the whole role.
| Role | Postings analyzed | Asks for modern AI | 95% interval | Asks for machine learning | Asks for traditional analytics | Top AI skills asked (share of all postings) |
|---|---|---|---|---|---|---|
| Data scientist | 385 | 243 / 385 (63.1%) | 58.2 to 67.8 | 318 (82.6%) | 366 (95.1%) | GenAI and LLMs 36.9%, agents 21.6%, AI evaluation and governance 16.9% |
| Analytics engineer | 72 | 41 / 72 (56.9%) | 45.4 to 67.7 | 14 (19.4%) | 72 (100%) | AI-assisted analytics 36.1%, agents 20.8%, GenAI and LLMs 16.7% |
| Data engineer | 497 | 219 / 497 (44.1%) | 39.8 to 48.5 | 133 (26.8%) | 483 (97.2%) | GenAI and LLMs 18.3%, agents 15.3%, RAG and retrieval 9.9% |
| Data analyst | 340 | 81 / 340 (23.8%) | 19.6 to 28.6 | 26 (7.6%) | 319 (93.8%) | AI-assisted analytics 12.6%, GenAI and LLMs 7.1%, AI coding assistants 7.1% |
The three groups overlap. A posting that names an LLM and scikit-learn counts in both modern AI and machine learning.
Which kind of AI each role asks for
Each posting falls in one column only, so the rows sum to 100%.
| Role | n | Both modern AI and ML | ML only | Modern AI only | Neither |
|---|---|---|---|---|---|
| Data scientist | 385 | 218 (56.6%) | 100 (26.0%) | 25 (6.5%) | 42 (10.9%) |
| Analytics engineer | 72 | 9 (12.5%) | 5 (6.9%) | 32 (44.4%) | 26 (36.1%) |
| Data engineer | 497 | 99 (19.9%) | 34 (6.8%) | 120 (24.1%) | 244 (49.1%) |
| Data analyst | 340 | 18 (5.3%) | 8 (2.4%) | 63 (18.5%) | 251 (73.8%) |
The 15-role census
The first run covered 15 roles and 8,251 postings. It measured two things. Mentions means the posting names a specific AI or machine learning technology anywhere. Requires means a model read the posting and judged AI to be a requirement of the job. The gap between the two columns is the finding.
| Role | Postings | Mentions a named AI or ML technology | Requires AI | Note |
|---|---|---|---|---|
| ML engineer | 302 | 300 (99.3%) | 270 (89.4%) | AI-defined title |
| AI engineer | 286 | 273 (95.5%) | 212 (74.1%) | AI-defined title |
| Data scientist | 385 | 336 (87.3%) | 83 (21.6%) | |
| Software engineer | 1,051 | 554 (52.7%) | 114 (10.8%) | |
| Data engineer | 497 | 231 (46.5%) | 19 (3.8%) | |
| Analytics engineer | 72 | 32 (44.4%) | 2 (2.8%) | n = 72, directional |
| Product manager | 761 | 283 (37.2%) | 31 (4.1%) | |
| Marketing manager | 212 | 49 (23.1%) | 2 (0.9%) | |
| Data analyst | 340 | 75 (22.1%) | 3 (0.9%) | |
| Business analyst | 260 | 33 (12.7%) | 3 (1.2%) | |
| Financial analyst | 499 | 39 (7.8%) | 5 (1.0%) | |
| Accountant | 596 | 15 (2.5%) | 0 (0.0%) | |
| Nurse | 942 | 1 (0.1%) | 0 (0.0%) | |
| Electrician | 699 | 0 (0.0%) | 0 (0.0%) | |
| Teacher | 1,349 | 0 (0.0%) | 0 (0.0%) | |
| All 15 roles | 8,251 | 744 (9.0%) |
Two things to know about this table. The mention count has no term for plain "AI", so it undercounts postings that say only "experience with AI tools". Recomputed with generic AI language, data scientist moves from 87.3% to 92.7% and data analyst from 22.1% to 37.1%. And ML engineer and AI engineer are AI-defined titles, so their rates are partly definitional.
Which AI skills data roles ask for in 2026
For data analysts, the AI skills named most were AI-assisted analytics (12.6%), AI with no technology named (12.6%), generative AI (7.1%) and AI coding assistants (7.1%). SQL, BI and pipelines were asked for in 91.2%. These are shares of 340 US postings collected August 28, 2026, counting only what the posting asks of the candidate. The 22% in the summary at the top counts any mention of a named AI or ML technology.
Each cell is the share of all postings in the role that ask the candidate for the skill. A posting that names an LLM inside an agent counts in both rows, so a column does not sum to 100%. No single modern AI skill reaches half of the postings in any of the four roles. SQL, BI and pipelines are asked for in at least 82% of postings in all four.
| Skill | Data analyst (340) | Data scientist (385) | Data engineer (497) | Analytics engineer (72) |
|---|---|---|---|---|
| Modern AI | ||||
| AI-assisted analytics | 12.6% | 8.1% | 8.5% | 36.1% |
| AI, no specific technology named | 12.6% | 48.6% | 32.4% | 19.4% |
| Generative AI and LLMs | 7.1% | 36.9% | 18.3% | 16.7% |
| AI coding assistants | 7.1% | 4.9% | 9.5% | 13.9% |
| Agents | 4.4% | 21.6% | 15.3% | 20.8% |
| AI evaluation and governance | 2.4% | 16.9% | 6.2% | 8.3% |
| Prompting | 1.8% | 11.2% | 4.2% | 8.3% |
| RAG and retrieval | 0.6% | 14.3% | 9.9% | 12.5% |
| Machine learning | ||||
| Machine learning methods and frameworks | 7.4% | 82.1% | 24.9% | 18.1% |
| MLOps | 0.6% | 35.1% | 10.7% | 4.2% |
| Traditional analytics, for comparison | ||||
| Statistics, forecasting and experiments | 47.6% | 88.3% | 12.1% | 30.6% |
| SQL, BI and pipelines | 91.2% | 82.3% | 97.0% | 100% |
"AI, no specific technology named" is a posting that asks for AI and nothing narrower. The machine learning row counts named methods and frameworks. The analytics engineer column rests on 72 postings and is directional. Every figure comes from the published CSV, one row per posting.
By industry
Industry was assigned to each employer from its name, its short blurb on the board, and its domain. The job boards supplied an industry for only 31% of postings, so we classified the rest ourselves. The classifier agreed with the board's own label 63% of the time. Sector differences are real. Individual sector rates are not precise. Quote the ranking. Treat the number as soft.
Expected counts come from each sector's own role mix at the four base rates above. A ratio above 1 means the sector asks for AI more often than its mix of roles predicts.
| Sector | Postings | Ask for modern AI | Rate | 95% interval | Observed / expected | Employers |
|---|---|---|---|---|---|---|
| Media and entertainment | 38 | 22 | 57.9% | 42 to 72 | 1.23 | 28 |
| Tech and software | 175 | 102 | 58.3% | 51 to 65 | 1.21 | 117 |
| Banking and financial services | 107 | 61 | 57.0% | 48 to 66 | 1.20 | 65 |
| Legal | 13 | 7 | 53.8% | 29 to 77 | 1.13 | 8 |
| Retail and consumer | 90 | 46 | 51.1% | 41 to 61 | 1.09 | 44 |
| IT services and consulting | 233 | 108 | 46.4% | 40 to 53 | 1.05 | 119 |
| Manufacturing, energy, transport | 118 | 53 | 44.9% | 36 to 54 | 1.01 | 86 |
| Education | 34 | 10 | 29.4% | 17 to 46 | 0.90 | 29 |
| Other | 70 | 26 | 37.1% | 27 to 49 | 0.90 | 56 |
| Government and defense | 156 | 62 | 39.7% | 32 to 48 | 0.88 | 86 |
| Health care and pharma | 112 | 36 | 32.1% | 24 to 41 | 0.76 | 89 |
| Real estate | 30 | 12 | 40.0% | 25 to 58 | 0.73 | 18 |
| Insurance | 67 | 21 | 31.3% | 22 to 43 | 0.67 | 39 |
The same data cut by role. Cells under 15 postings are in brackets and should not be quoted.
| Sector | Data scientist | Analytics engineer | Data engineer | Data analyst | All four |
|---|---|---|---|---|---|
| IT services and consulting | 40/46 = 87% | [7/9] | 54/129 = 42% | 7/49 = 14% | 46% |
| Tech and software | 45/68 = 66% | 11/16 = 69% | 28/54 = 52% | 18/37 = 49% | 58% |
| Government and defense | 37/64 = 58% | [2/2] | 16/38 = 42% | 7/52 = 13% | 40% |
| Manufacturing, energy, transport | 21/33 = 64% | [3/7] | 17/44 = 39% | 12/34 = 35% | 45% |
| Health care and pharma | 17/32 = 53% | [2/4] | 12/34 = 35% | 5/42 = 12% | 32% |
| Banking and financial services | 25/32 = 78% | [3/13] | 26/42 = 62% | 7/20 = 35% | 57% |
| Retail and consumer | 20/33 = 61% | [1/3] | 18/34 = 53% | 7/20 = 35% | 51% |
| Insurance | 9/23 = 39% | [0/2] | 7/28 = 25% | [5/14] | 31% |
| Education | [4/5] | none | [3/5] | 3/24 = 12% | 29% |
| All sectors | 238/375 = 63% | 40/71 = 56% | 209/473 = 44% | 79/324 = 24% | 46% |
What to learn by role
The blog post reads these tables into advice for each role. We keep that in one place rather than repeating it here. Read the post: What 8,000 job postings say about AI in data roles.
Method
What was collected
We pulled 56,746 unique US postings on August 28, 2026, from Indeed and LinkedIn using JobSpy. Each posting carries its source URL. We searched 70 role families across 14 locations. Indeed supplied 56,001 postings and LinkedIn 745. The pull took the top results in each board's own order, so sponsored postings are over-represented. It is not a random sample.
How postings were kept
We kept a posting only when its title was the role. A search for "data analyst" also returns PowerApps developers and IT specialists. Those were dropped. After the strict title match and content dedup, 8,251 postings remained across 15 roles. Indeed supplied 7,644 of those and LinkedIn 607.
How postings were classified
Run one covered the 15 roles. A keyword taxonomy flagged named AI and ML technologies in every posting. Claude Sonnet 5 then read each of the 8,251 postings and judged whether AI was required, preferred, or absent. The run finished with zero errors. An earlier run with a 44% silent failure rate was discarded.
Run two re-read the 1,294 postings in the four data roles on September 7, 2026. A pattern scan first located every candidate passage in the full description. Claude Sonnet 5 then ruled on each passage: is this AI in the intended sense, is it asked of the candidate, and what action is expected. It had to answer every candidate, so a silent miss was impossible. Nothing from run one was reused. The machine learning column in the key numbers table counts MLOps as well as named methods and frameworks, so it runs slightly higher than the machine learning row in the skills table.
Validation
Claude Opus 5 re-read 183 rows chosen to over-sample the hardest calls. It changed 26 of them on at least one dimension. Most changes moved a passage between reject and company context, which does not touch the reported rates. Four over-claims and four under-claims roughly cancel. No human has checked a label. Treat the review as a consistency check. It does not measure accuracy.
Sample sizes per cut
| Cut | Postings | Detail |
|---|---|---|
| Collected | 56,746 | Indeed 56,001, LinkedIn 745. One day, 2026-08-28. |
| Strict title match, 15 roles | 8,251 | Indeed 7,644, LinkedIn 607. Every posting classified. |
| Four data roles | 1,294 | Data scientist 385, data analyst 340, data engineer 497, analytics engineer 72. Indeed 1,118, LinkedIn 176. |
| Independent model review | 183 | Stratified rows re-read by Claude Opus 5. 26 changed. |
| Industry assignment | 1,243 | 96.1% of the four-role postings. 784 of 799 employers placed. |
LinkedIn cells in the four-role run range from 29 to 61 postings and are directional only. Sector cells range from 13 to 233 postings. Five sectors have fewer than 40.
Known limits
The postings come from one day, so there is no time series. Postings describe employer intent. They do not describe hires or the work performed. The role mix is our search list, so there is no overall market rate. Confidence intervals cover sampling error only. They do not cover classifier error. Salary was not modelled in the four-role run.
The analytics engineer caveat
Analytics engineer has 72 postings. Its modern AI rate of 56.9% carries an interval of 45.4% to 67.7%, which overlaps data engineer. Its 36.1% for AI-assisted analytics rests on 26 postings. Every analytics engineer figure is directional.
How this page relates to the post
The post's numbers agree with the files above, rounded. The post says "almost 9 out of 10" data scientist postings. That is the 87.3% mention rate from the 15-role census. The candidate-facing modern AI rate for the same role is 63.1%, and modern AI or machine learning together is 89.1%. The post says "over the last few weeks" because the analysis ran from August 28 to September 9. The postings themselves are from one day. An earlier version of the study used 30 postings per role. Those numbers were superseded on September 4, 2026, and do not appear here.
Download the data
One row per posting for the four data roles, with the role, board, title, employer, source URL, and every derived flag. No personal data. Download the CSV (1,294 rows, 400 KB).
Cite this
AI Analyst Lab. "AI in data job postings, 2026." September 25, 2026. https://aianalystlab.ai/research/ai-in-data-job-postings