Last updated: August 2026
Choosing a material is one of the highest-impact decisions in a design. Get it right and the part is light, cheap, and lasts; get it wrong and you find out in the field. The promise of AI material selection in engineering is a faster route from “I need something light, corrosion-resistant, and good to about 120 degrees Celsius” to a shortlist of sensible candidates with the tradeoffs explained. That promise is real, but only if you keep AI in its lane: it is an assistant that shortlists and explains, never the source of a property value and never the engineer who signs off the choice.
This guide keeps the proven selection method at the centre and shows exactly where AI helps and where it will hurt you. It pairs with our wider roundup of AI tools for mechanical engineering and our honest look at the limitations of AI in engineering.
Short answer: Use AI to translate your requirements into function, constraints, and objectives, to shortlist candidate materials, and to explain tradeoffs and draft your selection rationale. Do not use it as a source of property numbers: every yield strength, modulus, or fatigue limit must come from a datasheet or verified database, because language models invent plausible, wrong values. The classic screen-and-rank method, backed by tools like Ansys Granta Selector and Matmatch, stays the source of truth, and a qualified engineer validates the final choice against verified data.
The method AI has to respect
Material selection has a well-established backbone, usually credited to Michael Ashby: translate the design into function, constraints, objectives, and free variables; screen out anything that fails a hard constraint; rank the survivors using a material index that captures your objective, such as specific stiffness for a light, stiff part; then apply supporting information like cost, availability, and manufacturability to the shortlist. AI can assist every step, but it cannot replace the logic. If you skip the method and just ask a chatbot “what material should I use,” you get a confident answer with none of the reasoning you need to defend it.
The reason this matters is scale. Engineers choose from a vast catalogue of materials, commonly cited as well over 100,000 once you count every alloy, polymer, and composite grade. No one holds that in their head, which is exactly why a systematic method plus good data beats intuition, and why AI is tempting as a way to move faster through it.
Where AI genuinely helps
Mapped onto that method, four uses stand out as real and defensible.
- Translating requirements. A language model turns a plain-English brief into a structured set of functions, constraints, and objectives, which is a fast way to start the screening step without missing an obvious requirement.
- Shortlisting and explaining tradeoffs. AI proposes candidate material families and explains the strength-versus-weight-versus-cost-versus-corrosion tradeoffs in prose, and it often surfaces a less obvious option, such as an aluminium alloy or a reinforced polymer where you defaulted to steel.
- Predicting properties of novel materials. This is the genuine machine learning end of the field. Materials-informatics platforms such as Citrine Informatics, Senvol for additive manufacturing, and Mat3ra predict properties from composition and processing, which is real modelling rather than datasheet lookup.
- Drafting the rationale. Once you have verified numbers, AI drafts the selection-justification narrative for your design review, so you spend your time checking rather than writing.
The payoff at the front of the pipeline is large because that is where cost and performance get locked in. The Materials Genome Initiative exists on exactly this premise: GE reports cutting a jet-engine alloy development cycle from fifteen years to nine by working data-first (NIST, Materials Genome Initiative). AI is one more lever on that same front end.
The tools, graded honestly
Most software sold as “AI material selection” is a curated database with excellent search and charting, not machine learning. That is still useful; it is just not what the label implies. The table separates the two. Pricing is shape only; check each official site.
| Tool | What it is | Genuine ML? |
|---|---|---|
| Ansys Granta Selector | Selection and Ashby-chart analysis over a large curated dataset | No, curated data |
| Matmatch | Free searchable materials database with charts and suppliers | No, curated data |
| MatWeb | Property-datasheet database for verifying numbers | No, datasheet lookup |
| Total Materia | Large commercial database with standards cross-references | No, database |
| Citrine Informatics | Materials-informatics platform for property prediction | Yes, real ML |
| Senvol ML | Machine learning for additive-manufacturing materials | Yes, real ML |
| Mat3ra | Cloud platform combining physics simulation and ML | Yes, physics plus ML |
| ChatGPT or Claude | Reasoning assistant for translation and rationale | Reasoning only, no data |
Ansys Granta Selector remains the industry standard for structured selection, with the vendor citing a dataset of more than 400,000 entries, and Matmatch is a strong free starting point. Both are curated data with powerful filtering, not AI. The true machine learning lives in the informatics platforms, and they are aimed at research and novel-material development rather than everyday part selection. General language models are reasoning assistants: excellent for shaping the problem, useless as a source of a fatigue limit.
Where AI fails, and how to stay safe
The failure modes here are specific and serious, because a bad material choice can end in a broken part rather than a bad paragraph.
- It hallucinates property values. A language model will state a yield strength, modulus, or fatigue limit with total confidence, and it will sometimes be wrong. Never trust an AI-quoted property number. Every load-bearing value comes from a datasheet, a verified database, or a mill certificate.
- It does not know your context. Service environment, code and standard requirements, real supply and lead times, and manufacturability are invisible to a generic model unless you supply them. Left out, it can quietly miss corrosion, creep, fatigue, and temperature-driven failure modes.
- It does not carry the liability. The final material choice is an accountable engineering decision. It must be validated by a qualified engineer against verified data and, where needed, physical testing. AI accelerates the shortlist and the write-up; it does not make the call.
The safe pattern is simple: let AI shape the problem and propose candidates, then confirm every number against a datasheet before it enters a calculation. For more on why AI arithmetic itself needs checking, see our guide to whether AI can do engineering calculations. Where to find and verify property data is its own topic, and one we cover separately; here the rule is only that the number must come from a trusted source, not from the model.
Frequently asked questions
Can AI choose materials for engineering design?
AI can shortlist candidates and explain tradeoffs, but it should not make the final choice. Use it to translate requirements into functions and constraints, to propose material families, and to draft your rationale. The screening and ranking logic, the verified property data, and the sign-off all stay with the engineer. Treat AI as a fast assistant on top of the established selection method.
Is ChatGPT reliable for material properties?
No. Language models can produce confident, plausible, and wrong values for yield strength, modulus, fatigue, and thermal properties. Never use an AI-quoted number in a calculation. Get every property from a datasheet, a verified database such as MatWeb or Ansys Granta, or a supplier mill certificate. ChatGPT is useful for reasoning about the choice, not for the numbers behind it.
What is the Ashby method?
The Ashby method is a systematic way to select materials: define the function, constraints, objectives, and free variables of the part, screen out materials that fail a hard constraint, then rank the survivors using a material index that captures your objective, such as specific stiffness. Material-property charts make the tradeoffs visible. It is the backbone that AI assists rather than replaces.
What software do engineers use to select materials?
Ansys Granta Selector is the industry standard for structured selection and charting, and Matmatch is a strong free database with filtering and Ashby charts. MatWeb and Total Materia are used to verify property numbers. For machine-learning property prediction of novel materials, engineers turn to informatics platforms like Citrine, Senvol, and Mat3ra, which are aimed at research rather than routine part selection.
Is materials informatics the same as AI material selection?
Not quite. Materials informatics uses machine learning to predict the properties of new or unlisted materials from composition and processing, which is genuine AI aimed at developing novel materials. AI-assisted selection is broader and more everyday: using a model to shortlist and reason about existing materials for a specific part. Most engineers doing routine design use the second, backed by curated databases.
Sources
- NIST: Materials Genome Initiative
- Ansys Granta Selector
- Matmatch materials database
- MatWeb material property data
- Citrine Informatics
- Senvol additive-manufacturing data and ML
Written by the CognitiveFuture editorial team. We build our guidance from established selection methodology and official product documentation, and we label vendor-reported figures as such. We do not independently benchmark any tool, and we do not treat AI-generated property values as reliable. A qualified engineer should validate any material selection against verified data before it is used.
Richard Johnson
Richard Johnson is an AI specialist at one of the world's largest technology companies, where he has spent the past three years helping organizations adopt AI. CognitiveFuture extends that work publicly: gathering the available evidence on each tool, from vendor documentation to independent reviews and user feedback, and cutting a crowded market down to the right choice for the job in front of you.