Last updated: September 2026
Sizing a cable is not one calculation, it is an ordered process, and that is exactly where an AI assistant tends to be either genuinely useful or quietly dangerous. Used well, it drafts the step sequence, lays out the candidate sizes in a clean table, and keeps your units straight. Used badly, it invents an ampacity number that was never in the table and hands you a conductor that fails on site. This guide walks the full cable sizing workflow with AI in the loop, and marks the exact points where the standard and the engineer, not the model, have to decide.
For the underlying question of what AI can and cannot do here, and the voltage-drop math itself, see our companion piece on AI for cable sizing and voltage drop. For the input that starts the whole process, see AI for electrical load calculations. This article is the step-by-step procedure that ties those together, and it sits under our hub on the best AI tools for electrical engineering.
The short version: Cable sizing is an ordered process. Gather the inputs (load current, length, installation method, ambient temperature, grouping, allowed voltage drop, fault level, protective device), then pick the smallest conductor that satisfies three constraints at once: corrected ampacity, voltage-drop limit, and short-circuit withstand. An AI assistant is good for drafting the steps, tabulating candidates, keeping units consistent, and writing up the record, but it must not choose the final size or be trusted to recall a code-table number. The engineer confirms every governing value against the actual NEC or IEC table and signs off.

What actually decides a cable size
A cable size is not set by the load current alone. The final conductor is the smallest one that simultaneously passes three separate checks, and a size that clears one can still be forced larger by another. The three are corrected ampacity (current-carrying capacity after derating), voltage drop over the run, and short-circuit withstand during the time the protective device takes to clear a fault. Engineering references that work worked examples are explicit that you take the largest size any single check demands, not the average (ELEK NEC wire-sizing example; Enginist, cable sizing fundamentals). Get this framing right and the rest of the workflow is bookkeeping.
Step 1: Gather the inputs before you touch a model
The workflow only works if the inputs are complete, so collect them first: the design or load current, the run length, the installation or reference method, the ambient temperature, how many current-carrying conductors are grouped together, the allowed voltage drop, the prospective short-circuit level, and the protective device rating. This is a good place to let an AI assistant help you build a checklist and a blank input table, because it is organizing, not deciding. It is also where the load calculation feeds in: the design current you carry into sizing has to be the corrected, continuous-load figure, not a nameplate guess.
Step 2: Ampacity, then derate for the real conditions
Start from the base ampacity table for your code world, then correct it for conditions that are almost never the standard case. On the NEC side, the base allowable ampacities live in Table 310.16, quoted at 30 C (86 F) ambient with not more than three current-carrying conductors, in 60, 75, and 90 C insulation columns (ExpertCE summary of NEC 310.16). Two corrections then apply: a temperature-correction factor when the ambient is not 30 C, found in NEC 310.15(B), and an adjustment factor when more than three current-carrying conductors share a raceway or cable, in NEC 310.15(C)(1), where four to six conductors derate to 80 percent, seven to nine to 70 percent, and ten to twenty to 50 percent (ElectricalLicenseRenewal, reproducing the adjustment factors). Cite the section, not the subscript: the numbering shifted between the 2017 and 2020/2023 editions.
In the IEC world the same idea lives in IEC 60364-5-52, “Low-voltage electrical installations – Part 5-52: Selection and erection of electrical equipment – Wiring systems” (IEC webstore abstract). Its annexes give current-carrying capacities at a reference ambient with rating factors for ambient temperature, grouping, and installation method, and you size from the corrected capacity (Schneider Electrical Installation Guide). Keep the two systems separate: an NEC table value and an IEC table value are not interchangeable, and an AI that blends them is a real risk.
Step 3: Check voltage drop as its own constraint
Ampacity keeps the cable from overheating; it says nothing about whether the voltage at the far end is still usable. That is a separate check against your project limit. In the NEC, the familiar 3 percent branch and 5 percent total figures are a recommendation, not a rule: they appear in an Informational Note to 210.19(A) and the parallel note at 215.2(A), and Informational Notes are explicitly not enforceable unless a local authority adopts them (OrbitalJump analysis; EEPower). We keep the derivation out of this guide on purpose; the formula and the mistake most people make with it are covered in our voltage-drop article. Here it is simply one of the three gates a candidate size has to pass.

Step 4: Confirm short-circuit withstand
The third check is the one people forget: can the conductor survive the fault current for as long as the protective device takes to clear it? For faults up to a few seconds the heating is adiabatic, and on the IEC side the minimum size is governed by IEC 60364-4-43, clause 434.5.2, through the relation S = sqrt(I^2 t) / k, where S is the cross-section, I the fault current, t the clearing time, and k a material constant (Schneider Electrical Installation Guide). As an illustration, Schneider lists k values of 115 for PVC-insulated copper, 143 for XLPE copper, 76 for PVC aluminium, and 94 for XLPE aluminium; treat those as the reference’s figures, not ours, and read them from the current standard for your design. A conductor that passes ampacity and voltage drop can still be too small to survive a fault, which is why this check stays in the loop.
Where AI helps, and where it must not decide
AI is at its best on the parts of this workflow that are organizing rather than governing. It can draft the ordered steps, build the input checklist, tabulate several candidate sizes side by side, keep units consistent across the calculation, and turn a finished result into a clean calculation record. What it must not do is choose the final conductor or be trusted to recall a table number, because language models make arithmetic and solution-step errors even when the reasoning is set up correctly. Research on program-aided models found that offloading the actual computation to a Python interpreter beat chain-of-thought prompting by about 15 percentage points on the GSM8K benchmark (Gao et al., PAL, arXiv 2211.10435), and GSM8K itself showed that even large models struggle to robustly carry out multi-step arithmetic (Cobbe et al., arXiv 2110.14168). The practical rule follows directly: let the model structure and document, let a calculator or a validated tool compute, and let the engineer confirm every governing value. The same discipline applies to AI in engineering calculations generally.
Step 5: Document so the next person can follow it
The last step is the calculation record: the inputs you used, the table and edition you read, the corrections you applied, the three checks and which one governed, and the size you landed on. This is genuinely good work for an AI assistant, because it is turning your confirmed numbers into readable prose and a tidy table, not generating the numbers. A record that states its assumptions and its source tables is also what makes the design reviewable, which matters far more than shaving minutes off the write-up. If your inputs came from an older drawing, our guide to whether AI can read a wiring diagram covers how much of that you can trust the model to extract.
Frequently asked questions
Can AI size a cable for me?
It can draft the workflow, tabulate candidate sizes, and write up the result, but it should not choose the final conductor. The governing numbers come from the code ampacity table plus derating, the voltage-drop limit, and the short-circuit check, and every one of those has to be confirmed by the engineer against the actual standard. Treat the model as a drafting and bookkeeping assistant, not the authority on the number.
What are the three checks a cable size must pass?
Corrected ampacity (the table value after temperature and grouping derating), voltage drop within the project limit over the run length, and short-circuit withstand for the fault clearing time. The final size is the smallest conductor that passes all three, so a size that clears ampacity can still be forced larger by voltage drop or fault withstand.
Is the NEC 3 percent voltage-drop figure mandatory?
No. The 3 percent branch and 5 percent total values sit in an Informational Note to NEC 210.19(A) and 215.2(A), and Informational Notes are not enforceable requirements unless a local authority adopts them as an amendment. They are a sensible efficiency target, and many designs treat them as a limit, but the code itself frames them as a recommendation.
Should I use NEC or IEC values with an AI assistant?
Whichever governs your project, but never both at once. NEC Table 310.16 values and IEC 60364-5-52 values are built on different reference conditions and are not interchangeable. Tell the assistant which code world you are in, and confirm the number against that standard’s actual table rather than trusting the model to have the right one in memory.
Sources
- ExpertCE, NEC 310.16 ampacity tables guide
- ElectricalLicenseRenewal, NEC 310.15(C)(1) adjustment factors
- OrbitalJump, NEC voltage-drop recommendation vs requirement
- EEPower, computing voltage drop (NEC basics)
- IEC webstore, IEC 60364-5-52 abstract
- Schneider Electrical Installation Guide, general method for cable sizing
- Schneider Electrical Installation Guide, short-circuit withstand verification
- ELEK, NEC wire-sizing calculation example
- Gao et al., PAL: Program-aided Language Models, arXiv 2211.10435
- Cobbe et al., Training Verifiers to Solve Math Word Problems (GSM8K), arXiv 2110.14168
Written by the CognitiveFuture editorial team. Technical claims are drawn from the standards and engineering references linked above; NFPA 70 (NEC) and the IEC standards are paywalled, so code clauses are cited through authoritative secondary sources that reproduce them, and the exact section numbers can shift between editions. Confirm the current edition and table for your jurisdiction before sizing any conductor. This is general information, not a substitute for a qualified engineer’s judgment.


