Construction Project Managers Have Gone All-In on AI. Most Are Using the Wrong Tools.
Mastt's 2026 survey finds 66.7% of construction PMs use ChatGPT, but only 3.7% use Procore's built-in AI. Here's what that selection gap actually costs them.
Mastt's second annual State of AI in Construction Project Management survey has a number that should give every purpose-built construction tech vendor pause. Among 108 construction project management professionals surveyed between March and June 2026, 66.7% use ChatGPT regularly. Half use Microsoft Copilot. Only 3.7% use the AI built into Procore or Autodesk, the platforms most of them already run their projects on.
That's not an AI adoption story. It's a tool selection story, and the selection most construction teams are making is almost certainly costing them.
What Construction Teams Are Actually Using
The Mastt survey, gathered through industry newsletters, LinkedIn outreach, and direct email invitations to its subscriber base, breaks down tool usage in the past three months clearly:
| Tool | % of Respondents |
|---|---|
| ChatGPT | 66.7% |
| Microsoft Copilot | 50.9% |
| Claude | ~33% |
| Google Gemini | ~33% |
| AI inside Mastt | 8.3% |
| AI inside Procore / Autodesk | 3.7% |
| Specialist AI (Togal.AI, etc.) | 0.9% |
| No AI tools | 8.3% |
The Microsoft Copilot number makes sense. Enterprise Microsoft 365 rollouts have embedded it across industries, and construction companies aren't exempt. But the specialist tool numbers are almost embarrassingly low. Fewer than 1% of respondents use purpose-built estimating tools like Togal.AI. Less than 4% are touching the AI their own project management platforms have spent the last two years building.
The Blueprint Problem That ChatGPT Can't Fix
A benchmark examined by Digital Journal found that even the most advanced AI models remain poorly equipped to interpret construction drawings accurately. This matters because construction projects live and die by blueprint accuracy. AI that writes meeting notes faster doesn't reduce cost overruns. A misread takeoff does.
The tools designed specifically for construction drawings handle this differently. Togal.AI and Beam AI work with the annotation-heavy formats that blueprint sets actually use and output data that feeds directly into project budgets. ALICE Technologies and nPlan run schedule simulations against your critical path. Buildots uses hardhat-mounted cameras to track field progress automatically.
ChatGPT can't do any of that reliably. The benchmark data confirms it. And yet it's the tool two-thirds of the industry reaches for first.
Why Specialist Tools Aren't Winning
This isn't mysterious. Generic AI is free or already paid for. ChatGPT and Copilot require no new vendor relationship, no IT approval, no integration work. A project manager can start using them in five minutes, which matters in an industry where margins are thin and time is expensive.
Purpose-built tools carry a different cost structure. Mindcore's 2026 analysis of construction AI tools put it plainly: what decides whether a purchase survives past one project is whether the drawings are clean enough for a model to read and whether the field can feed it data without someone driving files back to the office.
That last condition rules out a lot of job sites.
The result is predictable. Construction teams get the generic productivity gains of AI: faster document searches, better-written RFIs, quicker meeting summaries. The higher-value applications stay untouched. The same pattern plays out across professional services. AI Is Flooding Insurance Underwriting in 2026 shows identical dynamics: broad adoption optimizing for speed, with precision applications still gaining ground. The Big Four Are Now Hiring More AI Specialists Than Auditors specifically to accelerate purpose-specific AI deployment across their construction and infrastructure clients.
The Market Pressure That's Coming
The Mastt survey found the AI in construction project management market is forecast to grow at a 21.77% compound annual growth rate, up from a prior estimate of 17.3%. The total market forecast has more than doubled since the 2023 target of $5.7 billion by 2028. Gartner, cited in the Mastt research, expects AI to run 80% of project management tasks by 2030.
Those numbers suggest specialist tools will eventually reach the teams that need them. The question is how much inefficiency accumulates in the meantime. In construction, where a single cost overrun can wipe out an entire project's margin, "getting there eventually" has a real price tag.
Similar adoption gaps show up in other operations-heavy sectors. The Top 9 AI Tools for Logistics and Supply Chain in 2026 covers comparable dynamics: industry-specific tools exist, but generic AI dominates actual usage. And how independent home inspectors are using AI in 2026 surfaces the same trade-off between free generic tools and purpose-built alternatives in the adjacent built environment space.
What Project Managers Should Do Now
Check what your existing platforms already include. The 3.7% figure for Procore and Autodesk AI use doesn't mean those features don't exist. They do. Teams already paying for these platforms should audit which AI features are active in their accounts before buying anything new.
Start with takeoff if you're evaluating specialist tools. The Mindcore analysis is direct: start with takeoff, because the drawings are input the contractor already has. No new data infrastructure required.
Treat data connectivity as a prerequisite for field AI. Scheduling simulation and field tracking tools are only as good as the data feeding them. If your superintendents run paper logs, that breaks any field AI tool. Fix the data flow first.
Don't let blueprint quality be a reason to avoid specialist tools. The benchmark finding that general AI struggles with construction drawings is exactly why purpose-built tools exist. They handle the annotation-heavy formats that general models can't.
Construction's AI adoption is real. The Mastt data makes that clear. But adoption concentrated in generic tools, while purpose-built applications sit below 1% usage, isn't a success story. It's the setup for the next round of project overruns that AI was supposed to prevent.
