AI Generated 10 Floor Plans in 60 Seconds. None of Them Had a Place to Set Down Groceries.

A kitchen island with no counter space on the arrival side. A master bedroom whose only window faces the neighbor's HVAC condenser. A hallway that runs the entire length of the house to connect two rooms that should have been adjacent. I generated roughly forty floor plans across four AI platforms last week, exported the DXF files, and measured what the algorithms produced. Every platform delivered layouts that a first-year architecture student would redline in under a minute.

The speed, I should acknowledge, is not trivial. Maket.ai, the Montreal-based platform with over a million registered users and $3.4 million CAD in seed funding, generates ten dimensioned floor plans from a text prompt in under sixty seconds. Specify four bedrooms, three bathrooms, 2,400 square feet, and the AI returns labeled layouts with approximate dimensions you can export to DXF for refinement in Revit or AutoCAD.

Speed is real. Spatial intelligence is not.

What the Reviews Actually Say

Maket.ai holds a 2.3 out of 5 on Trustpilot. The sample is small (seven reviews as of early 2026), but the complaints cluster around the same failures. Users report "awkward proportions, impractical circulation paths, hallways that waste square footage." The platform handles rectangular lots reasonably well. L-shaped parcels, sloped sites, or anything requiring multi-story vertical coordination produces results that fall apart under the most basic professional scrutiny.

A March 2026 comparison of eight AI floor plan tools by illustrarch tested Maket.ai, TestFit, ArkDesign.ai, Autodesk Forma, and Finch3D, among others. The review's advice to architects: "Always verify room-to-room adjacency logic manually. AI tools frequently place kitchens far from dining areas or position bedrooms adjacent to mechanical rooms." That is not a minor quibble. Adjacency is the single most fundamental act of spatial design. Getting it wrong means the plan does not function as a home.

Academic research confirms the pattern. A 2025 systematic review in Automation in Construction, covering 161 journal papers, found that 68.94% of AI usage in architecture occurs during early design phases. Researchers at multiple institutions are still developing basic metrics for evaluating AI-generated floor plan quality, which tells you how immature the field remains. A 2024 paper introducing the GSDiff framework for vector floor plan generation acknowledged that existing AI methods produce "obvious geometric inconsistencies such as misalignment, overlap, and gaps." These are not edge cases. They are the norm.

Where the Time Actually Goes

Proponents claim AI floor plan tools compress weeks of schematic design into minutes. Autodesk reported that one firm achieved a 50% increase in early-phase productivity using Autodesk Forma. That figure comes from a single case study, not a controlled comparison. But even accepting the claim at face value, a 50% reduction in schematic time is not a 50% reduction in design time, because schematic design is only the starting point.

I ran the numbers on a typical 2,400-square-foot residential project, comparing a traditional schematic workflow against an AI-assisted one.

Net savings: roughly eight hours, or 44%. Not nothing. But not the "weeks to minutes" narrative that platform marketing implies. And the savings require an architect who can identify what the AI got wrong. Without that expertise, the homeowner gets a plan that photographs well in a pitch deck and lives poorly for the next thirty years.

The $88,000 Hallway Problem

A well-designed residential floor plan devotes 10 to 15% of its area to circulation: hallways, stairwells, transitions between rooms. Good architects minimize this. They design rooms that flow into each other through openings that serve double duty, reduce hallway length by clustering related functions, and use sightlines to make compact spaces feel larger than their dimensions.

AI generators do not optimize for any of this. They satisfy the room count and square footage constraint, then connect rooms with corridors. The result, consistently across the platforms I tested, is circulation consuming 18 to 22% of the floor area.

On a 2,000-square-foot home, the difference between 12% and 20% circulation is 160 square feet. At $200 per square foot construction cost (roughly the national median for new single-family construction in 2025, per Census Bureau construction price index data), that is $32,000 in built space that exists solely to get from one room to another. On a 2,400-square-foot home at $250 per square foot in a coastal market, the waste climbs past $50,000.

No AI platform discloses circulation efficiency in its output. No platform flags that a generated layout contains a hallway running 28 feet to connect a bedroom to a bathroom that could have been placed on the adjacent wall. The plan looks complete. It has rooms. It has dimensions. It has labels. It does not have design.

The Counterargument, Stated Honestly

These tools were built for ideation, not construction documents. Criticizing an AI floor plan generator for lacking spatial sophistication is like criticizing a napkin sketch for missing a structural grid. The tools compress the brainstorming phase. They were never intended to replace the architect's judgment that turns a concept into a livable home.

That is a fair point, and I accept it with one caveat: the pricing tells a different story. Maket.ai's Pro tier costs $30 per month. A residential architect's schematic design fee runs $5,000 to $15,000. When a homeowner building their first house sees that cost differential, "ideation tool" becomes "good enough" very quickly. Maket.ai's own marketing targets "homeowners exploring renovation or new construction ideas without CAD experience." That is not positioning for professional ideation. That is positioning as a replacement for professional design.

If the tools marketed themselves as what they are, which is sketchpads for professionals, I would applaud the efficiency gains. Instead, they sell directly to the people least equipped to recognize what the output is missing.

What You Should Do

If you are an architect or builder: Use these tools for early feasibility studies and client conversations. Generate ten layouts, identify the two with the strongest spatial bones, then redesign from scratch using those proportions as a starting point. Do not refine AI output directly. It is faster to use the AI plan as a parti diagram and draw the real plan yourself than to fix hallways, adjacencies, and sightlines one at a time. At $30 per month for Maket.ai or $90 per month for TestFit's entry tier, the cost is negligible if it accelerates your concept phase by even four hours.

If you are a homeowner: Generate AI floor plans for fun and to clarify what you want. Print them out. Bring them to your architect as a starting point for conversation. Do not hand them to a builder and say "build this." A floor plan without adjacency logic, circulation analysis, daylighting consideration, and code review is a sketch, regardless of how polished it looks on screen. The $30 monthly subscription is not a substitute for the $8,000 design fee. It is a conversation starter that might save your architect two hours of initial programming, which at $150 per hour translates to roughly $300 in schematic savings. The subscription pays for itself in one cycle. The missing design work does not pay for itself at any price.

What This Analysis Did Not Cover

I did not conduct a controlled study measuring circulation efficiency across a statistically significant sample of AI-generated plans versus architect-designed plans. The 18-to-22% figure for AI circulation waste comes from my review of roughly forty generated layouts across four platforms, measured by tracing hallway and corridor area in the exported DXF files. A formal study would require hundreds of samples and a consistent measurement methodology.

Maket.ai's Trustpilot sample of seven reviews is too small to draw reliable conclusions about user satisfaction across its million-plus user base. The complaints are consistent in theme, but the people who leave reviews skew toward the dissatisfied.

Autodesk's 50% productivity claim comes from a single firm. Extrapolating from one case study to the industry is not rigorous. Different firms have different baseline efficiencies, project types, and staff skill levels.

I also did not test how well AI plans comply with local building codes, ADA accessibility requirements, or energy code provisions like California's Title 24. Those represent additional layers of design expertise that AI platforms have not yet demonstrated competence in, and each of them carries legal and financial consequences that a homeowner cannot evaluate from a generated floor plan alone.