Nature Machine Intelligence Submission Guide: Fit, Evidence, and Files
A practical guide to Nature Machine Intelligence fit, broad-readership evidence, initial submission files, and editorial-screen risks.
Readiness scan
Find out if this manuscript is ready to submit.
Run the Free Readiness Scan before you submit. Catch the issues editors reject on first read.
Quick answer: This Nature Machine Intelligence submission guide is for manuscripts offering a broadly important advance in artificial intelligence, machine learning, robotics, or their scientific and societal use, where the evidence supports more than a narrow leaderboard improvement. Think twice when the main contribution is a small gain on familiar benchmarks, a field-specific application without general insight, or a claim of generality tested in one controlled setting. Initial submission to this Nature Portfolio journal is format-flexible, but the manuscript, cover letter, data/code position, limitations, and supplementary evidence must tell one credible story.
Run a Nature Machine Intelligence submission readiness check before using the manuscript-tracking site.
From our manuscript review practice
Nature Machine Intelligence fit is rarely decided by the largest score in one table. The stronger signal is whether the work changes how a broad AI, robotics, or scientific audience understands a problem, method, or deployment constraint.
Nature Machine Intelligence submission facts
Requirement | Current official guidance |
|---|---|
Publisher | Nature Portfolio, Springer Nature |
Scope | Artificial intelligence, machine learning, robotics, and their broader scientific and societal implications |
Initial format | PDF, Word, or TeX/LaTeX accepted; submit compiled PDFs for TeX/LaTeX |
Initial package | Manuscript file, cover letter, and optional relevant Supplementary Information |
Article main text | Up to 3,500 words, excluding abstract, Methods, references, and figure legends; up to 6 figures and/or tables |
Review identity | Double-anonymized peer review is available when authors prepare anonymized files |
AI disclosure | LLMs do not qualify as authors; relevant use should be documented in Methods or another suitable section |
Submission route |
The official submission guidelines own the operational rules. This guide answers the harder pre-submission question: whether the paper's general insight, evaluation design, transparency, and limitations make it plausible for broad editorial consideration.
How this guide was reviewed
We reviewed the live Nature Machine Intelligence submission workflow, content-type definitions, preparation instructions, initial-formatting page, double-anonymized review instructions, editorial policies, and journal contact page on July 12, 2026. We separated explicit publisher requirements from Manusights editorial-risk interpretation.
In our pre-submission reviews of machine-learning and computational manuscripts, the recurring failure is contribution inflation: the abstract promises a general advance, while the evidence demonstrates a local optimization under one dataset, split, compute budget, or evaluation protocol. Nature Portfolio does not publish a formula for editorial selection. The diagnostic sections below show where claim breadth and evidence breadth commonly diverge.
Evidence boundary: this page does not use confidential Nature correspondence, private reviewer reports, or an unpublished acceptance model. The editorial-screen timeline is an author preparation tool, not a promise about how many days the journal will take.
Should your paper target Nature Machine Intelligence?
The journal covers technical AI and robotics research, but venue fit depends on why a broad reader should care. A specialist advance can fit when it changes a general method, reveals an important limitation, enables a new scientific capability, or alters the responsible use of machine intelligence. A large model or expensive experiment is not automatically broad.
Submit If
- the central finding changes a general understanding of learning, reasoning, perception, control, human-AI interaction, or scientific discovery
- evaluation spans the settings needed to support the paper's general claim
- comparisons include the strongest relevant baselines under fair data and compute conditions
- data, code, model access, and reproducibility limits are explained clearly
- the paper states practical and scientific limitations without hiding the conditions under which the method fails
Think Twice If
- the abstract presents a small benchmark improvement as a transferable insight without naming the mechanism
- the main results table omits the strongest baseline or evaluates it under a different resource budget
- the model's claimed generality rests on one dataset, institution, language, robot, or simulation environment
- a scientific application uses AI competently but does not change either the application field or machine-intelligence methods
- the abstract's societal, clinical, or safety claims extend beyond what the study measured
The first-screen question is simple: after removing the journal name and the largest metric, can a reader explain what is newly understood and why it matters outside this exact experiment?
The breadth-of-evidence test
Claim made by the paper | Evidence needed before submission | Common mismatch |
|---|---|---|
Better predictive performance | Fair baselines, uncertainty, repeated runs, leakage controls, and meaningful effect size | One favorable split or selective metric |
Generalization | Multiple datasets, domains, institutions, languages, tasks, or environments matched to the claim | A single held-out benchmark labeled general |
Robustness | Predefined perturbations, distribution shifts, failure analysis, and sensitivity checks | Average accuracy without worst-case behavior |
Scientific discovery | Domain-valid validation, mechanistic or experimental follow-up, and expert interpretation | A model-generated hypothesis treated as confirmed discovery |
Real-world utility | Workflow comparison, user or system outcome, deployment constraints, and error consequences | Retrospective technical performance only |
Responsible or safe AI | Measured risks, affected groups, misuse paths, mitigation limits, and governance context | Broad ethics language without an evaluation |
This is a Manusights decision artifact, not a Nature checklist. Its purpose is to prevent the title and abstract from outrunning the evidence package.
What belongs in the initial package?
Artifact | What to verify |
|---|---|
Manuscript | Clear title, abstract, main text, Methods, figures, Extended Data where relevant, and a claim-evidence sequence readable outside the niche |
Cover letter | Broad significance, closest competing work, content type, prior editor contact if any, and a precise reason the journal's readership benefits |
Supplementary Information | Only material relevant to understanding and replicating the conclusions, organized for review |
Data and code position | Repository or access route, licensing or privacy constraints, evaluation scripts, and a clear statement of what cannot be shared |
Reporting detail | Dataset provenance, sample construction, train-validation-test separation, compute, hyperparameters, statistical analysis, and human-participant oversight where relevant |
Author metadata | Complete names and affiliations, moved-address handling where needed, and anonymization if double-anonymized review is selected |
AI-use disclosure | No LLM authorship; document relevant LLM use in Methods or another suitable section |
Nature Machine Intelligence states that the cover letter should explain the importance of the work and why it fits the journal's diverse readership. That is a high-value constraint. A cover letter that only repeats the abstract misses the editorial decision.
What editors can inspect in the first several days
This model describes likely inspection order, not a published Nature timeline. It moves from package completeness through broad-interest fit, novelty, evidence width, transparency, and the practical ability to recruit suitable reviewers.
Day 0: Completeness and policy check
The journal can verify that the manuscript, cover letter, author information, optional Supplementary Information, and review-identity choice are internally consistent. Overlap with concurrent work, authorship, AI disclosure, and policy-sensitive data should be resolved before submission.
Day 1: Scope and broad-interest check
The title, abstract, first figure, and final conclusion reveal whether the paper speaks to a broad machine-intelligence question. A niche application can pass this check when it produces a transferable scientific or technical insight.
Day 2: Novelty and closest-work check
Editors can compare the claimed advance with the strongest recent methods and adjacent Nature Portfolio work. A novelty claim is vulnerable when the real change is scale, tuning, or dataset composition rather than a new principle or capability.
Day 3: Evidence-width check
The evaluation matrix shows whether the paper tested the dimensions implied by its claim: domains, shifts, languages, users, environments, compute budgets, or scientific validation. One benchmark cannot carry a general conclusion.
Day 4: Trust and reviewer-route check
Data provenance, code availability, limitations, statistics, and ethical context affect whether qualified reviewers can assess the work. If the method and application each require separate leaps of faith, the submission becomes difficult to route confidently.
In our pre-submission review work: Nature Machine Intelligence failure patterns
Across our Nature Machine Intelligence pre-submission reviews, we find a repeatable set of named failure patterns in machine-learning papers whose technical execution is stronger than their journal fit. The abstract often makes a general claim; the methods and data table reveal a narrower population or benchmark; the results omit the strongest resource-matched baseline; and the discussion treats failure cases as future work rather than evidence that bounds the contribution. These are component-level checks, not a claim about a private Nature screening formula. They let authors test whether the title, abstract, methods, figures, tables, code statement, and limitations section support the same breadth of conclusion.
Benchmark gain without a general insight
In our AI manuscript reviews, this pattern appears when a model improves one or two metrics but the manuscript cannot state what the result teaches about learning, representation, reasoning, control, or use. More experiments may confirm the gain without creating a broad contribution. The authors need to identify the principle, boundary condition, or capability that changes understanding.
Check if your Nature Machine Intelligence result adds a general insight ->
Generality claimed from a narrow evaluation
A paper can be technically sound yet overclaim across populations, languages, domains, or environments. The fix is either broader validation or narrower language. Editors can detect the mismatch by comparing the abstract's nouns with the datasets and settings in the methods table.
Check whether your Nature Machine Intelligence generalization claim is supported ->
Baseline and compute asymmetry
Comparisons become unconvincing when the proposed model receives more data, tuning, parameters, retrieval, annotation, or inference compute than the baselines. Report resource budgets and tune strong baselines fairly. If an advantage depends on scale, state that dependence rather than presenting it as an architectural result.
Check your Nature Machine Intelligence baselines and compute comparison ->
Transparency arrives too late
Code, data, prompts, model versions, preprocessing, exclusions, and evaluation scripts determine whether reviewers can reproduce the claim. A promise to release materials after acceptance leaves the strongest claims hard to audit. When privacy, licensing, security, or commercial restrictions apply, describe exactly what can be inspected and what cannot.
For Nature Machine Intelligence, these failures compound. A narrow benchmark encourages a broad abstract to lean on novelty language; missing resource accounting makes the gain difficult to interpret; and an incomplete code or data statement prevents reviewers from checking whether the result survives a fair comparison. In practice, we see the strongest revisions begin with a claim-evidence matrix: each abstract claim is mapped to a figure or table, a methods section, a baseline, and a stated limitation. Any claim without all four supports is narrowed or tested again.
This guide tells you what Nature Machine Intelligence editors look for in the public submission package; the review tells you whether your paper passes the broad-interest, evidence, and transparency checks. Paid reviews include a 60-day money-back guarantee, and we do not train models on submitted manuscripts.
Check whether the claim breadth matches the evaluation breadth before submission.
Readiness check
Run the scan against the requirements while they're in front of you.
See score, top issues, and journal-fit signals before you submit.
Nature Machine Intelligence versus nearby venues
Decision factor | Nature Machine Intelligence | Nature Communications | IEEE TPAMI |
|---|---|---|---|
Best fit | Broad AI, ML, or robotics advance with scientific or societal consequence | Broad multidisciplinary result where AI may be the method or subject | Deep pattern-analysis and machine-intelligence methods with strong technical validation |
First editorial question | Why does this change machine intelligence broadly? | Why does this matter across a broad scientific audience? | Is the method a substantial advance for pattern analysis or machine intelligence? |
Think twice when | Contribution is narrow or evidence does not support broad claims | Story is mainly for specialist AI readers | The work is primarily societal analysis or a narrow application |
Other prestige routes answer different reader jobs. Science Robotics is often the clearer target when embodied system evidence is central, Cell-family computational venues can fit biology-led contributions, and JAMA-family journals are more appropriate when clinical validity and care outcomes dominate. Nature Machine Intelligence remains the best fit only when machine intelligence itself is changed or newly understood.
If the contribution is primarily technical and field-specialist, compare the IEEE TPAMI submission guide. If the strongest contribution is a scientific application rather than an AI principle, the Nature Communications submission guide provides a different broad-interest test. For medical imaging, the IEEE Transactions on Medical Imaging submission guide may offer a more natural expert audience.
Cover letter strategy
Use the cover letter to explain the paper's consequence, not to stack adjectives. A useful structure is: the unresolved broad problem, the specific advance, the strongest evidence, the boundary of the claim, and why Nature Machine Intelligence readers across subfields should care.
Name the closest work fairly. Editors do not need a claim that nothing similar exists; they need to see what changes after this paper. If the difference is mainly data scale or engineering effort, say so and explain the capability that effort unlocks.
For interdisciplinary work, explain both directions of contribution. A medical, climate, chemistry, neuroscience, or social-science application should matter to that field, while also teaching the machine-intelligence audience something beyond successful tool use.
Initial formatting and later production
The journal accepts flexible initial formatting in PDF, Word, or TeX/LaTeX, with compiled PDFs for TeX/LaTeX submissions. Flexible format removes cosmetic work; it does not remove the need for readable figures, complete Methods, traceable data, and a coherent supplement.
Articles list a 3,500-word main-text limit excluding abstract, Methods, references, and figure legends. Do not use the Methods exclusion to hide the study's logic. The main text should let a broad reader understand the problem, advance, evidence, and limitation before specialist implementation detail begins.
If double-anonymized review is selected, remove author-identifying information from the manuscript and prepare the title page or cover-letter information according to the official instructions. Self-citations should remain scientifically useful without revealing identity through unnecessary phrasing.
Final pre-submission checklist
- The title states the real advance without claiming universal generality.
- The abstract explains why the result matters beyond one benchmark.
- The strongest relevant baselines receive fair data, tuning, and compute.
- Dataset provenance and split logic rule out obvious leakage.
- Repeated runs, uncertainty, and statistical choices are visible.
- Robustness tests match the conditions named in the claim.
- Limitations name where the method fails or remains untested.
- Data, code, prompts, models, and evaluation scripts have a precise access plan.
- The cover letter explains the diverse-readership value.
- The manuscript and supplement separate measured findings from proposed implications.
When the list exposes a claim-evidence gap, narrow the claim or expand the evaluation before upload. The Manusights review process separates journal fit, methods risk, claim support, and presentation, while example reports show how findings are tied to manuscript locations.
Last verified: July 12, 2026 against the live Nature Portfolio author pages.
Frequently asked questions
No special house format is required for initial submission as long as the study is clear enough for editorial assessment and peer review. The journal accepts PDF, Word, or TeX/LaTeX initially, with compiled PDFs requested for TeX/LaTeX submissions.
The journal does not promise one fixed timeline for every manuscript. Editorial assessment, reviewer recruitment, article type, and revision complexity vary, so authors should use the live journal information and their manuscript-tracking account.
Common manuscript-level risks are a small benchmark gain without a general insight, comparisons that omit the strongest methods, unclear train-test or data provenance, weak robustness across settings, and claims whose breadth exceeds the evidence.
Nature Machine Intelligence offers publishing-model choices whose costs and eligibility can change. Check the journal's current publishing-options page and institutional or funder coverage before making a financial decision.
Sources
Before you upload
Choose the next useful decision step first.
Move from this article into the next decision-support step. The scan works best once the journal and submission plan are clearer.
Use the scan once the manuscript and target journal are concrete enough to evaluate.
Anthropic Privacy Partner. Zero-retention manuscript processing.