Wall Assembly Comparison – Chat GPT
krackadile
| Posted in Green Building Techniques on
I have been looking at different wall assemblies for zone 4A. I used Chat GPT to compare two assemblies as seen below. I had assumed that a stick framed wall would be significantly more environmentally friendly based on what I’ve read on GBA but from what GPT says, that may not be the case. Why does GBA tend to sway towards stick framing and blown in cellulose and does what GPT provided appear to be correct (I know GPT can be off the mark quite a bit sometimes but sometimes it’s pretty close)? Thanks. Looking forward to some feedback.
Wall Assemblies
| Wall Assembly | Description |
|---|---|
| Wall A: 12″ CMU Wall | Reinforced partially grouted CMU, perlite insulation, exterior stucco, painted interior |
| Wall B: 2×6 Wood Frame | Dense-packed cellulose insulation, exterior OSB sheathing, house wrap, R-5 EPS rigid insulation, shiplap wood siding, interior painted drywall |
Key Assumptions & Data
| Aspect | Wall A (CMU) | Wall B (Wood Frame) |
|---|---|---|
| Embodied Carbon (kg CO₂e/ft²) | 1.08 | 1.70 (increased by EPS and OSB) |
| Operational Energy Savings (Annual CO₂e) | 270 kg CO₂e saved per 1000 ft² per year | Baseline (0) |
| Operational Carbon (50 years) (kg CO₂e per 1000 ft²) | 13,500 kg CO₂e saved | Baseline |
| Construction Cost ($/ft²) | $13.15 – $18.25 | $13.00 – $18.00 |
| Operational Cost Savings (Energy) | ~$540 saved over 50 years (assuming $0.12/kWh) | Baseline |
Explanation of numbers:
-
Embodied Carbon Wall B is higher due to EPS rigid insulation and OSB added.
-
Operational savings come from CMU’s thermal mass; wood frame baseline is zero savings.
-
Construction costs are roughly comparable because of added EPS and siding in Wall B.
-
Energy cost savings based on 6 kWh/ft² baseline, 10% reduction from CMU wall thermal mass.
Summary:
Aspect Wall A (12″ CMU) Wall B (2×6 Wood Frame + EPS) Embodied CO₂ Lower (1.08 t / 1000 ft²) Higher (1.7 t / 1000 ft²) Operational CO₂ Much lower (13.5 t saved over 50 years) Baseline (0) Upfront Cost Slightly higher or comparable ($13.15-$18.25/ft²) Comparable ($13-$18/ft²) Operational Cost Moderate savings (~$540/1000 ft² over 50 years) Baseline (0)
GBA Detail Library
A collection of one thousand construction details organized by climate and house part
Search and download construction details
Replies
krakodile,
These are not comparable walls - either in terms of R-value or appearance. A CMU wall with perlite, partially grouted cores, no interior finish or anywhere for services vs a wood framed one with cellulose, exterior insulation and drywall? And we are back into the whole Thermal Mass stuff.
Right. So, according to the IECC these both should meet the minimum requirements for a wall in zone 4A (at least I think they do). The interior finish is paint on the CMU, kinda like an institutional wall. The plan would be to either route the electrical in the cmu or on the surface in raceways. Chat GPT threw in the thermal mass talk, not me.
krackadile,
To get back to your original question: Given the wood framed wall will perform better, look nicer, and be easier to run services through than the CMU one, is it surprising that GBA posters would favour it?
Thanks.
Can you get Chat GPT to tell you what it thinks the R-value of a 12" CMU with perlite is?
Looking at the table here:
https://www.masonryandhardscapes.org/resource/tek-06-02c/
The r-value depends a lot upon the density of the concrete used. At 85 pcf the combined R-value is R11.47, at 135 pcf it's R5.68. At what I would consider a normal density, 125 pcf, it's R-6.47.
The framed wall has R5 continuous plus probably R19 batt. It's not even close.
Oh, looking at those charts I think you're right, the CMU wall wouldn't meet the IECC code requirements. I guess Chat GPT was off the mark on this one. I guess it won't be replacing designers just yet.
That's a good website. Thanks.
This is from Chat GPT:
The R-value of a 12-inch concrete masonry unit (CMU) with cores filled with perlite is approximately:
R-10.0 to R-11.0
Breakdown:
12" CMU alone (hollow): R-1.2 to R-1.5
Perlite fill: R-2.7 to R-3.0 per inch (loose fill)
When perlite fills the cores, the overall assembly R-value increases substantially due to reduced thermal bridging and added insulation.
The exact value depends on:
Density and quality of the CMU
How completely the perlite fills the cores
Presence of surface finishes (stucco, plaster, furring, etc.)
Rule of Thumb:
For typical 12" CMU with perlite-filled cores and standard conditions:
R-10 is a conservative, widely accepted estimate for total wall R-value.
Let me know if you need thermal modeling or comparisons with other insulation types.
Edit:
Google (note that google does not account for the entire wall assembly): A 12" CMU with cores filled with perlite has an R-value of approximately 23.6 when the cores are completely filled.
That is an excellent example of why I don't trust AI answers any farther than I can throw them. It is clearly disregarding the fact that CMUs have ends, webs and at least some concrete-filled cores, and instead treating a CMU as exterior and interior concrete with the middle area fully perlite. It's comparable to considering a framed wall to having no studs when doing the R-value computation.
Michael,
If you combine the webs, ends, and partially grouted cores, you are looking at least a third of the wall being R-1.5.
Malcolm, exactly--and that's assuming parallel heat flow, when in reality heat flow is three dimensional which would reduce the total R-value even more.
There is a thermal mass effect that would increase the apparent R-value, but only when temperatures are fluctuating, since thermal mass doesn't do a thing when temperatures are constant. And the effect is short-lived, since that amount of concrete can't store a lot of heat.
Chat GTP is way off on this. The operational cost of a 2x6+r5 wall is a small fraction of a filled CMU wall.
This is the problem with lot of these LLMs. Their answers seems right until you check.
Akos,
I suspect that's because the information they are being fed comes from CMU manufacturers or trade groups. Try searching for the R-value of a 12" CMU filled with perlite and see where the results come from - and their optimistic claims.
See post #6 above.
I suspect the answer might be coming from a blog where a "thermal mass" enthusiast claims that the extra "thermal mass" makes up for the far lower R-value.
This is a good point. I will run this through Trane Trace with a side by side comparison to see what it comes up with. Might take a minute.....
Edit: After double checking the numbers it seems the CMU wall is not actually code compliant so I will not be running the numbers through Trace (unless I get really bored).
I don't see costs being comparable. There is a lot more labor involved in building a CMY wall than there is in standing up a conventional studwall, and I don't see adding EPS being enough to offset the difference, or even siding. Even walls built with CMUs (many schools) tend to use an exterior facade or cladding some some type for aesthetic reasons, often some type of brick, which is an additional cost. I don't see how a CMU wall is in any way cost comparable there.
Also, running services in a CMU wall is a LOT more labor than running them in a studwall. Have your electricians quote both and you'll see what I mean.
Remember that the Chat GPT and the like are basically a huge memory bank -- a database, basically -- and a query tool that is more evolved. They have zero actual understanding of anything.
Bill
Off topic and just my opinion, I have no real interest what chat GPT has to say about anything, but if you care about being able to use chat gpt, stop posting its answers on the web.
The more AI created crap posted on the web, the more AI starts repeating itself
Excellent point.
Supposedly AI, at least Chat GPT, can recognize anything it's outputted. Presumably then they would have instructed it to not "learn" from it's own results? Otherwise it would like doing researching by reading your own book that you just wrote, probably not particularly useful.
Bill
There are now many various AI 'tools' and I do not know if they can tell each others drivel from actual information.
My opinion is that AI is hobbled by its basic goal, passing the Turing test.
It is taught to lie to you, so if it does not know the answer, it makes one up.
Note the near weekly new court filings with AI hallucinations, creating court case references that do not exist.
Look at the shear volume of bad information on the web, from sales exaggerations to oft repeated wives tales
i would request posters not quote Chat GPT or any other AI, but rather ask a question and compare it to the AI output, and if there is a conflict, bring that up.
#23 - There is no excuse not to seek out real expertise!
That's why I'm not interested in AI results either. At best (which is not the case) it will be learning from experts. More likely it will dumb down to the average or perhaps even a bit below. I'd rather seek out the experts myself.
In this case I like that the poster had the sense to double check those answers with real experts. And I'm fine with the idea of not posting AI derived dubious information as well.
Model collapse is already an issue in some areas.
Within the next couple of years, we will be awash with LLM content, the big challenge will be figuring out what information is real.
Like any tool, used the right way LLMs can be very useful. The other day I had it convert a data sheet into code definitions that would have take a lot of manual time. That is very real savings.
Best to learn how to use these.
If I were interviewing someone for a job where they were going to have to work without my direct supervision, there were two answers given by ChatGPT that would have immediately ended the interview.
The first is: "Operational savings come from CMU’s thermal mass; wood frame baseline is zero savings." Yeah, thanks for coming in.
The second is: "When perlite fills the cores, the overall assembly R-value increases substantially due to reduced thermal bridging and added insulation." This gets it exactly backwards. The problem with the CMU's is there's no overcoming the thermal bridging of the block material. It doesn't matter what you fill the interior with, it's never going to be a great insulator.
Excellent point that filling the blocks with insulating material does nothing to remove the thermal bridges, which are an inherent part of the design of the block.
I long ago decided to not rely on resume info when hiring people. Degrees weren't the big thing either. What we do is to setup a problem, then pretend to be a customer and ask the candidate to solve the problem. They can talk to us, and we respond like a customer would (i.e. "the internet is slow", etc.). What we're looking for is for the candidate to show some kind of logical progression that would eventually lead them to a solution to the problem. They have to think, and problem solve, in a simulated real-world setting.
BTW, I know "thermal mass" jumped right out at you too, DC, that was probably your first big red flag ;-)
Bill
I was going to stay out of this but what the heck…
As much as everyone loves shitting on AI, (and there is no shortage of valid criticisms) an important variable nobody mentioned is that these all seem to be using the free version of ChatGPT which has a tiny fraction of the power of its premium models. It's the equivalent of an alien talking to a bunch of five-year-olds and making the conclusion that humans are idiots.
The output with the paid 4o model below is much closer, even capturing some nuance with the thermal mass topic. Either breaking it into separate questions using the search function or using a deep research token (which I don't want to waste, sorry) would cite its sources and allow you to dig in and verify all the specifics while the latter would dedicate another order of magnitude more computing power to properly suss out more of the details.
My Original Prompt:
“Can you compare the two wall assemblies on the basis of 1. Embodied Carbon (kg CO₂e/ft²), 2. Operational Energy Savings (Annual CO₂e) 3. Operational Carbon (50 years) (kg CO₂e per 1000 ft²) 4. Construction Cost ($/ft²) 5. Operational Cost Savings (Energy)
Wall A: 12″ CMU Wall Reinforced partially grouted CMU, perlite insulation, exterior stucco, painted interior
Wall B: 2×6 Wood Frame Dense-packed cellulose insulation, exterior OSB sheathing, house wrap, R-5 EPS rigid insulation, shiplap wood siding, interior painted drywall”
The original output is a table that doesn't transpose into text easily, and a python script I’d assume most of you aren't interested in, so I asked it a follow up:
“Can you write out your methodology instead of the python script”
Response:
“Certainly. Here's the methodology I used to estimate and compare the two wall assemblies—Wall A (CMU with perlite) and Wall B (2×6 wood frame with cellulose and exterior foam)—across five categories:
1. Embodied Carbon (kg CO₂e/ft²)
Wall A: CMU + perlite
Concrete masonry has high embodied carbon (~0.15–0.20 kg CO₂e/lb).
Partially grouted 12″ CMU wall with stucco and paint finishes contributes significantly.
Perlite fill adds slightly, but much less than full grout.
Estimate: ~25 kg CO₂e/ft²
Wall B: Wood frame + cellulose + foam
Wood framing and cellulose insulation have relatively low embodied carbon; cellulose can even be carbon-negative.
EPS (extruded polystyrene) contributes more embodied carbon (~2–5 kg CO₂e/ft²).
Estimate: ~10 kg CO₂e/ft²
2. Operational Energy Savings (Annual CO₂e)
This refers to the annual CO₂ emissions avoided by using a more energy-efficient wall compared to a baseline (in this case, CMU).
Wall A: CMU
Used as the baseline wall.
Effective R-value ~5–6 with perlite fill.
High thermal mass offers some time-shifting of loads but not actual energy reduction unless climate and HVAC use align.
Savings: 0 CO₂e/year
Wall B: Wood + cellulose + R-5 EPS
Nominal R-value ~23; effective R-value ~18–20 after thermal bridging.
Better air sealing and reduced heating/cooling needs.
Based on EPA CO₂ conversion (~0.92 lb CO₂/kWh), energy modeling indicates ~100 kg CO₂e saved per 1000 ft² annually.
Savings: 100 kg CO₂e/year/1000 ft²
3. Operational Carbon (50 Years, kg CO₂e per 1000 ft²)
Wall A: CMU
If the building emits ~130 kg CO₂e/year/1000 ft² more than the high-performance wall:
130 kg/year × 50 years = ~6500 kg CO₂e
Wall B: Wood frame
100 kg/year savings × 50 years = ~1500 kg CO₂e
This reflects cumulative operational CO₂ emissions from heating/cooling differences.
4. Construction Cost ($/ft²)
Wall A: CMU
Includes cost of CMU units, reinforcement, partial grouting, stucco exterior, and paint interior.
Typical estimates: $20–30/ft² depending on region and labor rates.
Estimate: $25/ft²
Wall B: Wood frame
Includes 2×6 framing, cellulose, sheathing, EPS foam, wood siding, and drywall.
Common in residential; efficient to build; materials less expensive.
Estimate: $18/ft²
5. Operational Cost Savings (Energy, $/year per 1000 ft²)
Assuming $0.20/kWh electricity rate (California-like), and a 500–1000 kWh annual energy savings due to higher R-value:
Wall A: CMU
Baseline wall, no savings.
Savings: $0/year
Wall B: Wood frame
~500–1000 kWh/year × $0.20 = $100–$200/year
Estimate: $200/year per 1000 ft²
Let me know if you'd like to model actual site-specific values or account for climate zone, HVAC type, or net-zero goals.”
Is it a replacement for a qualified expert human? No.
Is it an amazingly powerful tool for skilled users to quickly do a tremendous variety of work at a tiny fraction of the cost and time of engaging an expert to do it for them, particularly if you use the more advanced models and or limit the variables in a given prompt? Yes, and they're getting better every day.
bcade,
I agree. The errors aren't inherent to AI, but rather to it's present state of development, and what is commonly available for casual users. That said, here we are discussing the misleading and useless answer given to someone trying to make informed decisions. It's important to know that the seemingly authoritative information it generates in response to factual inquiries can be very unreliable.
The same could be said for just about every building professional when talking about building science. Ai does not negate the need for critical thinking, it is a tool that you need to learn how to use and know its strengths and weaknesses. Figuring out how to structure prompts to limit hallucinations and get reference citations is a necessity. At least it specifically says ai can make mistakes on the prompt, I’ve never had an hvac contractor tell me that when they give me a ridiculously oversized system design.
Malcolm,
As a casual user myself, the $20/month is a pretty low barrier for the capabilities it provides. As Freyr alluded, there’s a disclaimer under every prompt, “ChatGPT can make mistakes. Check important info.” so that should go without saying.
Seemingly the OP and prior responses here are based on the free, lower performance version. Without that context, some of the critiques lacked the detail and clarity needed to present a full and accurate picture. One might even say some of the conclusions were “misleading and useless”, inadvertently hindering people from making informed decisions. I was attempting to provide some missing nuance.
While GusFHB is right, the GBA community is unbelievably generous and helpful with providing answers, many of the commenters in this thread no doubt have contributed thousands of unpaid hours to help others, but it can’t answer every question every person may have, nor could it scale if GBA became more popular. Look through the backlog of Q&A and even with only a handful of new posts a day there are plenty with zero replies.
As Freyr notes, there is no shortage of “experts” in the construction industry that will present all manner of pseudoscience, bias, and long since disproven theories as fact; the GBA community is very much the exception to the rule.
AI/ChatGPT isn't the boogie man, and it doesn’t need to be perfect to be useful. Just like any tool, you have to know how and when to use it.
bcade,
Yes and it's perhaps ironic that the misinformation about the merits of insulated CMU construction used by ChatGPT most likely came from industry sources, who unlike AI intentionally promote a false narrative as to the benefits of their products.
And there ya are, repeating AI spew for, uhh, no actual reason that I can think of
There are pretty smart people here who can direct people to the appropriate direction and thought process in designing a house/assembly/wall etc
If it were my site, I would delete any AI generated content, or quotes thereof, and if the poster became a repeat offender, I would ban them the same as AI questions or answers
Someday AI may free designers from worrying about design details
That is a fair bit in the future
I just wanted try and figure out what is the most environmentally friendly, cost effective, and energy efficient wall that can be built for basements and above ground. I live in an area that gets hit with tornadoes so I'm not a huge fan of stick frame after seeing some friends houses blown away but I was under the impression that concrete is not very environmentally friendly but GPT seemed to say otherwise so I was just as interested in the CO2 aspect as well as the efficiency aspect but DC pointed me to a link that explained it sufficiently and I saw the error in the equivalent r-values. I've read posts on GBA and it appears the consensus seems to me to be that double stud or dense packed cellulose stick frame above ground and ICF below ground are the way to go but I was trying to find something that was more environmentally friendly, cost effective, and/or efficient that may perform better in tornado prone areas.
ICF is expensive. Concrete is really the best option underground for foundations due to its longevity -- nothing else really comes close in that regard. If you have to rebuild your house due to a failing foundation, you have canceled out any advantage anything else may have had over concrete in terms of greeness.
Regarding stick framed structures in very high winds, I'd try building with hurricane ties. These ties are basically steel straps that tie the studs to the top plates and rafters, and the studs to the sill plates which are themselves bolted to the foundation. The goal is to resist uplift, and the results are very impressive. If you look down in Florida after a hurricane, you'll sometimes see a few houses standing with minimal damage while most others have been flattened. Hurricane ties are a big part of why those remaining homes are still there.
Bill
Thanks Bill. I will definitely plan to tie the roof down with hurricane ties.
I'd be interested in seeing a comparison of a wall made of AAC (autoclaved aerated concrete) to the wood framed wall, considering both upfront carbon and operational carbon, as well as consideration of the relative lifespans.
I'm an American living in France. AAC is very readily available here, as are masons. It's a pity the residential builders in the USA are less familiar with the material. It strikes me as a really useful alternative.