Search for a top 5 AI researchers 2026 ranking by influential citations and you will find plenty of lists, few of which say where their numbers come from. This one does. We read the citation counts on 7 October 2026 from two scholarly databases with public APIs, OpenAlex and Semantic Scholar, and every figure below links to the record it was taken from.
Two things are worth knowing before the table:
- This is a ranking by citation counts. A citation count measures how often a researcher's work is referenced. It does not measure whose ideas matter most, and nobody should read it that way.
- The order depends on the database. OpenAlex puts Kaiming He first. Semantic Scholar's counts for the same five put Geoffrey Hinton first, by 1,913 citations. Neither is wrong; they count differently, and we show how below.
The top 5 AI researchers in 2026, ranked by citations
| Rank | Researcher | OpenAlex citations | OpenAlex h-index | Semantic Scholar citations | Semantic Scholar h-index | Most-cited work in OpenAlex |
|---|---|---|---|---|---|---|
| 1 | Kaiming He | 584,533 | 86 | 583,396 | 66 | Deep Residual Learning for Image Recognition |
| 2 | Geoffrey E. Hinton | 517,844 | 147 | 585,309 | 158 | ImageNet classification with deep convolutional neural networks |
| 3 | Yoshua Bengio | 486,871 | 191 | 546,977 | 209 | Deep learning |
| 4 | Ross Girshick | 422,542 | 89 | 492,803 | 82 | Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks |
| 5 | Shaoqing Ren | 356,136 | 18 | 364,547 | 12 | Deep Residual Learning for Image Recognition |
Rank follows the OpenAlex totals. All counts were read on 7 October 2026, and both databases keep adding citations as new papers are indexed, so treat the table as a dated snapshot rather than a standing fact.
Sources, per researcher:
- Kaiming He — OpenAlex record, Semantic Scholar record
- Geoffrey E. Hinton — OpenAlex record, Semantic Scholar record
- Yoshua Bengio — OpenAlex record, Semantic Scholar record
- Ross Girshick — OpenAlex record, Semantic Scholar record
- Shaoqing Ren — OpenAlex record, Semantic Scholar record
These links go to each database's public API, which is where we read the numbers, so they open as raw data rather than as a profile page. Semantic Scholar limits requests that come without an API key; if a link answers "Too Many Requests", wait a minute and try again.
How we built the ranking
A list of "top researchers" is only as good as the method behind it, so here is ours, step by step.
- Start from OpenAlex's own list. We asked OpenAlex for the authors it associates with its Artificial Intelligence subfield, sorted by total citations (the query).
- Remove the two records that are not AI researchers. "R Core Team" is in that list because of one entry, R: A Language and Environment for Statistical Computing, which carries 363,597 citations on its own. Steven L. Salzberg's record is in it too, but his most-cited paper is Fast gapped-read alignment with Bowtie 2 and his leading topics are in genomics. Each record qualifies because OpenAlex files one of its topics under the AI subfield ("Data Analysis with R" for R Core Team, "Algorithms and Data Compression" for Salzberg); neither is primarily AI research.
- Cross-check against computer vision. OpenAlex's Computer Vision and Pattern Recognition subfield returns the same five at the top, in the same order, with nothing to remove (the query). Two different starting lists agreeing is the main reason we are comfortable calling this a top five.
- Check for split records. A researcher whose papers are spread across several author records gets under-counted. In OpenAlex, searches for Kaiming He, Geoffrey Hinton, Yoshua Bengio, Ross Girshick and Shaoqing Ren found no second record under the same name holding more than 266 citations. Semantic Scholar splits more: its search for Yoshua Bengio returns 20 records, and the second-largest holds 35,609 citations. We did not add secondary records to any total, because we cannot confirm that each one is the same person.
- Read each researcher's main Semantic Scholar record and put its count beside the OpenAlex one, rather than averaging the two.
We also tried Google Scholar's author listing for the artificial intelligence label. It redirected to a Google sign-in page, so Google Scholar is not part of this comparison.
Why the order changes between databases
OpenAlex has Kaiming He ahead of Geoffrey Hinton by 66,689 citations. Semantic Scholar has Hinton ahead of He by 1,913. A gap that size can flip on how a database handles a few papers, and the paper-level counts show that the two databases really do handle them differently:
| Paper | OpenAlex citations | Semantic Scholar citations | Year in OpenAlex | Year in Semantic Scholar |
|---|---|---|---|---|
| Deep Residual Learning for Image Recognition | 229,475 | 241,105 | 2016 | 2015 |
| ImageNet classification with deep convolutional neural networks | 109,560 | 131,783 | 2017 | 2012 |
| Gradient-based learning applied to document recognition | 59,993 | 63,070 | 1998 | 1998 |
| You Only Look Once: Unified, Real-Time Object Detection | 51,651 | 48,678 | 2016 | 2015 |
Each row compares the two databases' records for the same DOI (ResNet in OpenAlex and in Semantic Scholar; ImageNet classification in OpenAlex and in Semantic Scholar; gradient-based learning in OpenAlex and in Semantic Scholar; YOLO in OpenAlex and in Semantic Scholar).
Two things stand out. Semantic Scholar is not simply bigger: it has more citations for three of the four papers and fewer for YOLO. And the two databases disagree about dates for the same DOI, by as much as five years for the ImageNet classification paper. That points to the real difference: a famous paper often exists as a preprint, a conference version and a journal reprint, and each database decides differently which versions to treat as one paper and whose citations to pool. Any ranking built on totals inherits those decisions.
What "influential citations" add
Semantic Scholar publishes a second count beside the total. In its own words, it "identifies citations where the cited publication has a significant impact on the citing publication", and these influential citations "are determined utilizing a machine-learning model analyzing a number of factors including the number of citations to a publication, and the surrounding context for each" (Semantic Scholar's explanation, which points to a paper titled Identifying Meaningful Citations for the method). The same page adds a limit worth keeping in mind: identification "relies on our access to the full-text of the citing paper", so some influential citations go uncounted where that text is not available.
Semantic Scholar reports this per paper, so we read it for a landmark paper by each researcher in the top five:
| Paper | Top-5 authors on it | Citations | Highly influential citations | Influential share |
|---|---|---|---|---|
| Deep Residual Learning for Image Recognition | Kaiming He, Shaoqing Ren | 241,105 | 32,978 | 13.7% |
| Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks | Shaoqing Ren, Kaiming He, Ross Girshick | 76,175 | 9,868 | 13.0% |
| Gradient-based learning applied to document recognition | Yoshua Bengio | 63,070 | 6,481 | 10.3% |
| ImageNet classification with deep convolutional neural networks | Geoffrey E. Hinton | 131,783 | 13,438 | 10.2% |
| You Only Look Once: Unified, Real-Time Object Detection | Ross Girshick | 48,678 | 3,396 | 7.0% |
All five rows are Semantic Scholar figures read on 7 October 2026. The Faster R-CNN row comes from Semantic Scholar's record for its arXiv version; the others are the DOI records linked above.
The share is the useful part. By Semantic Scholar's model, roughly one citation in seven of the ResNet paper is one where the citing paper builds substantially on it; for YOLO it is about one in fourteen. A high total with a lower influential share is consistent with a paper that is cited widely, often as a reference point, while a higher share suggests more papers that build directly on the work. It is a second lens on "influence", not a replacement for the first — and it comes from a model that needs full text, so it carries that model's judgement and that coverage limit.
The five researchers, briefly
Each paragraph uses only what the databases above report, plus one prize citation.
1. Kaiming He
The most-cited author in OpenAlex's AI subfield on the day we checked. His three most-cited works there are Deep Residual Learning for Image Recognition (229,475 citations), Faster R-CNN (56,068) and Feature Pyramid Networks for Object Detection (30,051) (his top works in OpenAlex). OpenAlex lists 158 works for him, against 1,376 for Yoshua Bengio — a reminder of how far a total can rest on a small number of papers.
2. Geoffrey E. Hinton
Ahead of Kaiming He in Semantic Scholar's counts, second to him in OpenAlex. His most-cited works in OpenAlex are ImageNet classification with deep convolutional neural networks (109,560 citations, with Alex Krizhevsky and Ilya Sutskever), Deep learning (85,418, with Yann LeCun and Yoshua Bengio) and Visualizing Data using t-SNE (35,353) (his top works in OpenAlex). He shared the 2024 Nobel Prize in Physics with John J. Hopfield "for foundational discoveries and inventions that enable machine learning with artificial neural networks" (NobelPrize.org).
3. Yoshua Bengio
The highest h-index of the five in both databases: 191 in OpenAlex and 209 in Semantic Scholar. His most-cited works in OpenAlex are Deep learning (85,418), Gradient-based learning applied to document recognition (59,993, with Yann LeCun, Léon Bottou and Patrick Haffner) and Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation (25,377) (his top works in OpenAlex). Where Kaiming He's total is concentrated, Bengio's is spread across a very large body of work.
4. Ross Girshick
His three most-cited works in OpenAlex are all on object detection: Faster R-CNN (56,068), You Only Look Once: Unified, Real-Time Object Detection (51,651) and Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation (32,287) (his top works in OpenAlex).
5. Shaoqing Ren
The clearest example of how totals work. OpenAlex lists 24 works and an h-index of 18 for him, far below the other four, yet his total puts him fifth. It comes from co-authorship of two landmark papers, Deep Residual Learning for Image Recognition and Faster R-CNN, followed by Delving Deep into Rectifiers (19,064) (his top works in OpenAlex). Three of our top five — He, Ren and Girshick — are co-authors of Faster R-CNN.
Just outside the top five
The next five in OpenAlex's computer vision list are Xiangyu Zhang (327,490), Jian Sun (280,933), Andrew Zisserman (274,677), Yann LeCun (264,458) and Ilya Sutskever (251,112), from the same computer vision query as above. We did not check these five records for split or merged identities the way we checked the top five, so treat ranks six to ten as approximate. Andrew Zisserman, for instance, is not among the first 15 names in the Artificial Intelligence subfield list at all, which shows how much the choice of starting list matters lower down.
The limits of ranking researchers by citations
- Totals reward landmark papers and co-authorship. Shaoqing Ren's 24 works outrank researchers with hundreds, because two of them are among the most-cited papers in the field.
- Totals depend on the database. The same five researchers come out in a different order in OpenAlex and Semantic Scholar, and one paper's count differs by 22,223 citations between the two.
- Older work has had longer to collect citations. A 1998 paper and a 2016 paper are not competing on equal terms.
- Citations count references, not quality. Influential citations are one attempt to separate the two, and they come from a model.
- Every number here is a snapshot. The OpenAlex records we read had been updated the day before we read them.
Key definitions
| Term | What it means |
|---|---|
| Citation count | The number of indexed works that reference a paper or, summed, an author's papers. |
| h-index | The largest number h such that h of an author's papers have each been cited at least h times. |
| Highly influential citation | Semantic Scholar's term for a citation where the cited work has a significant impact on the citing work, identified by a machine-learning model. |
| Author disambiguation | Deciding which papers belong to which person. Errors either split one researcher across several records or merge several people into one. |
| OpenAlex | An index of scholarly works and authors, queried here through its public API. |
| Semantic Scholar | A research tool for scientific literature that publishes per-paper citation and influential-citation counts, queried here through its public API. |
A different kind of citation
"Citation" now has a second meaning outside academia. When an AI assistant answers a question, it often names or links the sources it drew on, and being one of those sources is the AI-search version of being cited. That is the kind of citation TopCited works on: across a set of queries, which brands AI assistants mention, and how your share of voice compares with the competitors mentioned alongside you. If that is the citation you care about, start with what AI visibility is or how to track brand mentions in AI search.
Frequently asked questions
It depends on the database. On 7 October 2026 OpenAlex put Kaiming He first among authors in its AI subfield, with 584,533 citations. Semantic Scholar's count for Geoffrey E. Hinton, 585,309, was 1,913 higher than its count for Kaiming He. We compared these five in Semantic Scholar rather than ranking its whole author list.
By OpenAlex totals on 7 October 2026: Kaiming He, Geoffrey E. Hinton, Yoshua Bengio, Ross Girshick and Shaoqing Ren. Semantic Scholar's counts for the same five put Hinton ahead of He and keep the other three in the same order.
Semantic Scholar's highly influential citations are citations where the cited paper has a significant impact on the citing paper, as judged by a machine-learning model that looks at citation counts and the context around each citation. They are reported per paper.
Because citation totals reward landmark papers. He is a co-author of Deep Residual Learning for Image Recognition and Faster R-CNN, which together carry most of his 356,136 OpenAlex citations.
Each has its own coverage of citing documents and its own way of deciding which versions of a paper count as one. For the same DOI they can even record different publication years, as they do for the ImageNet classification paper: 2017 in OpenAlex and 2012 in Semantic Scholar.
No. Academic citations are references in scholarly papers. AI-search citations are the sources an assistant names or links in an answer. TopCited works on the second kind: which brands AI assistants mention across a set of queries, and your share of voice against competitors.
The short version
On 7 October 2026 the top 5 AI researchers by citations in OpenAlex were Kaiming He, Geoffrey E. Hinton, Yoshua Bengio, Ross Girshick and Shaoqing Ren, and Semantic Scholar's counts for those five put Hinton narrowly ahead of He. The ranking is real but fragile: it rests on a few landmark papers, it changes with the database, and influential citations tell a somewhat different story about which work others build on. Read any top-five list, including this one, as a dated snapshot of one way of counting.