Not reliably. ChatGPT or another AI can explain how a load calculation works and check your inputs, but sizing a replacement quote should come from Manual J software, which applies the full method to the actual house and produces a report you can file. Treat any AI figure as a sanity check, never as the quote's sizing basis.
The temptation is understandable. You describe the house, the assistant returns "about 3 tons" in four seconds, and it sounds authoritative. The trouble is that you cannot see which assumptions produced the number. Was the attic duct loss included? Which design temperatures? Were latent loads counted? A report from approved software lists every input, so you and the customer can see what it assumed. A chat answer hides them.
The free online calculators that rank for this search make the same trade: a quick figure from a handful of inputs, with duct location, glass orientation and air leakage reduced to defaults. Neither they nor a chat answer replaces checking the figure against a Manual J report before it goes on a quote. For the estimating side of replacement work, what AI estimating software measures and misses for HVAC sits next to this one.
What Manual J is, in plain terms
Manual J is the residential load calculation method published by ACCA, the trade body for heating and cooling contractors. Alongside it, Manual S covers equipment selection against the calculated load, and Manual D covers duct design. Together they take a job from "how much heating and cooling does this house need" to "which equipment, and how much air through which ducts". ACCA publishes a list of approved Manual J software on its website, including Wrightsoft Right-J8, Elite RHVAC and Cool Calc Manual J, and states that software not on the list is not ACCA-approved. Whether your permit office requires a Manual J report for a replacement varies, so ask them; do not assume.
The calculation looks at how heat gets in and out of the building and what is generated inside it:
- Envelope: walls, roof or ceiling, floors, with their construction and insulation.
- Windows and doors: area, orientation, glass type, shading from overhangs and neighbours.
- Air leakage: infiltration, from the tightness of the house.
- Ducts: where they run and how well they are sealed and insulated, because ducts in a hot attic add load.
- Internal gains: people, appliances and lighting.
- Latent load: moisture to be removed, which affects comfort as much as temperature.
- Design conditions: the outdoor and indoor temperatures the system is sized for.
Every item changes the answer, and several interact. That is why a one-line description to a chat assistant, however detailed, still leaves most of the inputs to guesswork.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA worked house: rule of thumb against calculated load
This is an illustration with invented figures. A two-storey house of 2,000 square feet has a failing 4-ton air conditioner. The old rule of thumb says one ton per 500 square feet, so the quote would be 4 tons, or 48,000 BTU per hour of cooling, the same as the unit being replaced.
A Manual J run for the house, with figures rounded and illustrative, might return these cooling loads:
| Source of load | BTU per hour |
|---|---|
| Walls | 5,400 |
| Ceiling and roof | 4,300 |
| Windows (conduction and sun) | 6,900 |
| Air leakage (sensible) | 2,100 |
| Ducts in the attic | 2,600 |
| People and appliances (sensible) | 2,400 |
| Sensible total | 23,700 |
| Latent load (moisture) | 3,000 |
| Total cooling load | 26,700 |
At 12,000 BTU per hour per ton, 26,700 is about 2.2 tons. The rule of thumb said 4 tons, which is roughly 1.8 times the calculated load. Manual S then tells you to select equipment whose capacity at the design conditions sits within an acceptable range of the load, so the choice might be a 2.5-ton unit, not a 4-ton one. The house in this example has newer windows and extra attic insulation added since the original system went in, which is why the old unit was oversized. That is common: the building improves, the rule of thumb does not.
The calculated load also works out at about 13.4 BTU per hour per square foot (26,700 divided by 2,000). That number is useful as a sanity check against similar jobs, but it only exists because the calculation was done first.
Why an oversized system is a real problem
Short cycling
A unit much bigger than the load cools the house quickly and shuts off. It then restarts shortly after, again and again. The compressor spends its life starting up, which is when wear is highest and efficiency is lowest.
A damp house
Dehumidification happens while the coil runs. A system that cuts out early removes less moisture, so the air feels clammy at a comfortable-looking temperature. Customers describe this as "it's cold but sticky" and call you back.
Uneven rooms and noise
An oversized unit pushes more air than the ducts were built for, which can mean noise, whistling registers and rooms that overshoot while others lag. Manual D exists to check the duct side.
The undersized case
Sizing too small is the opposite risk: the system runs flat out on the hottest days and the customer is hot. The calculation is there to avoid both, which a copy of the old unit's size cannot do.
Where AI helps around the calculation
Checking inputs against what you saw
After you gather survey data, ask an assistant to flag things that look inconsistent. It reads fast and flags gaps well. Give it the notes, not an instruction to compute.
Below are survey notes for a residential replacement quote. Do NOT calculate loads and do NOT supply values I have not given you.
Task: list (1) inputs a Manual J calculation normally needs that these notes do not record, (2) places where two notes conflict (for example window area against the plans), (3) anything suggesting ducts in unconditioned space, a closed-off room or a recent renovation that changes the envelope, (4) questions to ask the homeowner.
Notes:
[paste notes]An illustrative output: "Missing: orientation of the largest windows; glass type; attic insulation depth; duct location for the upstairs return. Conflict: notes say 14 windows, plan shows 17. Renovation: homeowner mentioned a sunroom addition; is it served by this system? Questions: what is the thermostat set to on the hottest days, and are any rooms never used?" Each flag sends you back to the house or the plans.
Sanity-checking the software's result
Give the assistant the finished report and your notes and ask it to question anything that looks out of line: a window load that dwarfs the wall load in a house with small windows, a duct load of zero for ducts you saw in the attic, a result per square foot far outside the range of your recent jobs. It cannot say the report is wrong; it tells you where to look again.
Explaining the report to the customer
Write a plain-English explanation (about 180 words) for a homeowner of this load calculation result: total cooling load 26,700 BTU per hour, about 2.2 tons; the old system was 4 tons. Explain why the calculation, not the old unit's size, is the basis for the new quote, and why a bigger unit is not automatically better. Do not promise energy savings or lower bills. Use only the figures I have given you.Illustrative output: "Before choosing equipment, we calculated how much cooling your house really needs, using the measurements and details of your home. The answer is about 26,700 BTU per hour, roughly two and a quarter tons. Your old system was four tons, which was sized when the house had older windows and less insulation. A system that is too large turns on and off in short bursts, which wears it out and leaves the air feeling damp. We have chosen equipment that matches the calculated load." What you fix: remove anything that promises comfort or savings, and confirm the equipment named matches the quote.
Writing the paperwork around the job
Cover letters, the list of assumptions, customer questions and follow-up are the places AI saves the most time without any risk to the sizing. AI tools for HVAC calls, quotes and service plans covers the wider toolset, and following up unapproved estimates handles the quote that goes quiet.
What goes wrong when you ask a chat model to size
It reaches for a rule of thumb
Ask "how big an AC for a 2,000 square foot house" and you will often get a per-square-foot rule. That reproduces the very error the calculation corrects.
It assumes values silently
Without an instruction to refuse, a model fills gaps with typical numbers: average insulation, standard windows, ducts inside. The answer reads as if it were calculated for your house.
It does not show a defensible record
A permit office, an inspector or a customer's lawyer cannot audit a chat transcript the way they can a software report with inputs, design conditions and method.
It can be inconsistent run to run
Ask twice, get two answers. A calculation that gives different outputs for the same inputs is not a basis for a quote.
The heating side: a furnace swap worked through
Manual J covers heating as well as cooling, and the furnace is the other place the old size gets copied. Say the same illustrative house has an old 100,000 BTU per hour input furnace at 80% efficiency, which delivers about 80,000 BTU per hour. The heating load calculation, with attic insulation added since the furnace went in and the windows replaced, comes out at 38,000 BTU per hour at the design outdoor temperature. The furnace delivers more than twice the load. A like-for-like swap carries that forward: the burner fires in short bursts, the house swings in temperature and the flue gases cool and condense in the wrong places on some designs. A quick sum shows how far down the replacement can go: a 95% efficient furnace with a 40,000 BTU per hour input delivers about 38,000, while a 60,000 input unit at the same efficiency delivers about 57,000, half as much again as the load. Manual S sets how far output may exceed the heating load, so check the current edition before you choose; a modulating or two-stage model is another route, because it runs longer at lower output. The approach is the same as for cooling: calculate, then select against the load, then check the ducts can carry the airflow.
A filled-in survey sheet to copy
Better inputs make both the calculation and the AI audit stronger. Here is a short filled-in example, illustrative, of the level of detail that works. You can adapt it into a form your technicians carry.
- House: two storeys, 2,000 sq ft conditioned, built 1998, slab on grade, ceiling height 8 ft.
- Walls: 2x4 framing, fibreglass batt, vinyl siding, no exterior foam seen; attic access confirms insulation depth about 10 inches blown cellulose.
- Windows: 17 total, double-pane low-e replacement windows fitted in the last 5 years, 5 on the west wall, 4 on the south with a 2 ft overhang, 3 on the east, 5 on the north.
- Doors: two exterior, one glass slider on the west.
- Leakage: no blower-door test; visible gaps around attic hatch and recessed lights.
- Ducts: supply and return in the attic, mastic seen at some joints, tape at others, insulation R-6.
- Occupants: four; a home office with two monitors and a small server.
- Complaints: upstairs warm in the afternoon, downstairs clammy.
Running the input-audit prompt on this sheet would typically ask for the glass specification (U-factor and solar gain), the duct leakage evidence and whether the home office has its own cooling arrangements. Those are the inputs where a guess moves the answer most.
How much the design conditions matter
Every load calculation is built on design temperatures, the outdoor conditions the system is sized to handle on all but a few hours of the year. Choosing a more extreme outdoor design temperature raises the load, and so does choosing a lower indoor setpoint. A calculation that assumes 75 F indoors and one that assumes 72 F will give different totals for the same house, and a chat assistant that does not state the setpoint it used has quietly picked one. Software lists the design conditions at the top of the report, which is another reason to keep the calculation there. When you review a report, ask the assistant to compare the design conditions against your own standard (for example "our standard is 75 F indoor cooling, 70 F heating") and to flag any deviation. That is a task it does well.
When a customer asks for a bigger unit
Customers often equate bigger with better, especially when last summer was uncomfortable. The calculation is your evidence. Give the assistant the facts and have it draft a short reply you edit.
Draft a friendly reply (under 140 words) to a homeowner who asks us to fit a larger air conditioner than our calculation recommends. Facts to use: our load calculation gives about 26,700 BTU per hour (about 2.2 tons); we recommend a 2.5-ton unit; a larger unit would switch on and off in short bursts, wear out sooner and leave the air damp. Do not mention savings or bills. Offer to go through the calculation report with them.An illustrative output: "I understand wanting extra capacity after a hot summer. We sized your system from a calculation of your actual house, which came out at about two and a quarter tons, so we recommend a 2.5-ton unit. A bigger one would cool the house in short bursts and shut off before it dries the air, which can leave it feeling clammy and shorten the unit's life. I'd be happy to go through the report with you, so you can see where each part of the figure comes from." What you would change: make sure the equipment and figures match the quote exactly.
A second house, a different lesson
Say a small, well-sealed new-build of 1,400 square feet needs its first system. The rule of thumb gives about 3 tons. The calculation, with tight construction, good windows and ducts inside the conditioned space, may show a load well under 2 tons, and in such cases the calculation might point towards a smaller unit with a variable-speed compressor. The lesson from the first house was that old rules of thumb oversize houses that have been improved. The lesson from this one is that they oversize houses built better than the rule assumed. Either way, the rule is a historical average, and your customer's house is not average.
The same point shows up when you quote a house where the owner plans to add insulation or replace windows next year. If those upgrades are certain, the load will fall, so ask whether to size for the house as it is or as it will be, write down the choice, and let the customer decide with the facts in front of them.
The 3.5-ton quote that came from a chat answer
An illustration of the failure this whole tutorial guards against. A small contractor asks an assistant for the cooling load of a house, giving floor area, number of storeys and a general description. The answer is 3.5 tons, and the quote goes out on that. After installation the customer complains of a clammy downstairs. A later Manual J run shows a load near 2.4 tons: the house has shaded windows and sealed ducts in conditioned space, none of which was in the chat. The unit is too big, short-cycles and under-dehumidifies, and the contractor absorbs a return visit and a re-size. The lesson is narrower than "never use AI": the sizing comes from the calculation, and the assistant works on the inputs and the wording.
Eight checks between the survey and the quote
- Survey data recorded for envelope, windows (area, orientation, glass, shading), leakage, duct location and condition, internal gains.
- AI input audit done on the notes; every flag resolved or noted.
- Load calculated in approved Manual J software, with design conditions stated in the report.
- Equipment selected against the calculated load (the Manual S step), at the design conditions, not from the label size.
- Duct capacity checked if the airflow changes (the Manual D side).
- Report saved to the job file.
- Customer wording checked against the report; no savings promises.
- Permit office asked what it requires for this job type.
What a calculation costs next to one callback
An experienced estimator can take roughly 30 to 60 minutes for a standard house once the survey is done; the first few jobs take longer. The AI input audit adds a few minutes and tends to save a second visit. A business plan such as ChatGPT Business ($20 per user per month billed annually, two seats minimum) is small beside a single return visit. The saving disappears if the load calculation is skipped, because the cheapest quote that comes back is the one that gets sized wrong.
If you hold off on the calculation for speed, ask what a wrongly sized replacement costs: a callback, a re-size, a warranty argument and a customer who tells others. That is why the load calculation is the one step in a replacement quote worth protecting. For the same discipline applied to heat pump work, see what MCS expects from a heat loss calculation, and for post-install paperwork, warranty paperwork with AI and its cost.
Load calculations and AI: common questions
Can ChatGPT replace Manual J software?
No. Manual J is a defined method with many inputs, and software applies it consistently and produces a report you can file. A chat model can explain the method and check your inputs, but it can skip a term or assume a value without telling you.
Is a rule of thumb per square foot good enough for a replacement?
It is a rough starting point, not a sizing method. Houses of the same floor area can differ widely in glazing, insulation, leakage and ducts. Oversized equipment can short-cycle and leave the air damp, so calculate the load instead of copying the old unit's size.
Do I need a Manual J for a like-for-like swap?
Requirements depend on your permit office and the job, so ask them. Even where it is not required, a calculation protects you from repeating an old sizing mistake and gives you something to show a customer who asks why you chose a size.
What can AI check in a Manual J report?
Give it the report and your survey notes and ask it to flag inconsistencies: a window area that does not match the plans, a duct location that differs from what you saw, or a result per square foot far from similar jobs. It points you back to the inputs; it does not decide.
Further reads
- How to Use AI in HVAC Design: Options, Duct Checks and Proposals — Where AI fits in system design, duct checks and proposals.
- Good-Better-Best HVAC Pricing: A Worked Replacement Quote Written With AI — Turn a correctly sized system into a three-option quote.
- HVAC Callbacks: Log the Cause on Every Job and Spot the Pattern With AI — Log callback causes and spot sizing problems early.
- Follow Up After Giving a Quote: An HVAC Replacement Sequence With AI — Follow up a replacement quote that has gone quiet.
- ChatGPT for HVAC: 10 Prompts for Plans, Replacements, Reviews — Ten ChatGPT prompts for replacement and plan work.
- AI Service Plan Reminders for Heating and Cooling Customers — Keep customers on service plans after the install.
- AI for HVAC Contractors: Getting Answers From Manuals and Spec Sheets — How an HVAC contractor can search manuals and spec sheets with AI: what to load, a prompt that forces page numbers, a ten-question test and safety limits.
- Air Conditioning Installation Business: Survey Notes to Quote With AI — Turn air conditioning survey notes into a structured record, unit options and install scope lines with AI; the engineer keeps sizing and refrigerant decisions.
- ChatGPT for Building Design: Briefs, Options and Code Checks — Brief intake, layout options and a code-check log for a small design practice, with the line ChatGPT must not cross.
- AI for Mechanical Contractors: Can It Read Specs and Drawings? — What AI reads reliably in a mechanical bid set, where it fails on drawings, and a 12-point hand check plus a 45-minute test to run on a past bid.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: ACCA's approved software page and its descriptions of Manual J (residential load calculation), Manual S and Manual D; ChatGPT Business pricing page. Check your own permit office for what it requires.