Why ChatGPT is Failing Your Construction Site

The technology wasn’t the problem. The data was. Their closeout reports were spread across three different formats, living on two SharePoint sites and a shared network drive nobody had bothered to migrate. Half the PDF files were scanned images with no searchable text at all. Naming conventions had changed twice in 18 months, so the model couldn’t distinguish a $2-million highway interchange from a $200,000 drainage repair. Both showed up as “Infrastructure — Misc.”

I’ve been consulting on AI and data strategy for AEC and manufacturing firms for about five years now. This story isn’t unusual. I’d say some version of it plays out at nearly every company that tries to jump straight to the exciting part without doing the boring work first.

I saw something similar at a regional MEP contractor, about 200 people. They wanted AI to speed up submittal reviews, comparing incoming submittals against project specifications and flagging discrepancies. Solid use case, but their specs came from a dozen different architects who all formatted things differently. Submittals lived in Procore, in emailed PDFs at that, and in one case, as photos of a whiteboard someone snapped on their phone. The AI choked on it, and honestly, so did the team trying to make it work.

Harder Than it Looks

What’s driving this? The pitch from AI vendors sounds incredible. Drop your documents in, ask questions in plain English, get answers instantly. After years of digging through Procore exports and buried email chains, that promise feels like it might actually fix something. I get why people buy in.

However, what the demo doesn’t show you is the gap between clean sample data and your actual files. In the demo, everything is formatted, digitized, consistent. In your firm, the data is whatever your team produced under deadline pressure over the last 10 years. That’s a very different animal.

I’ve started calling this the “foundation problem.” It’s the single biggest reason AI pilots stall in this industry. Everyone focuses on the technology layer. Almost nobody focuses on the data layer underneath it. Vendors aren’t going to bring this up, because their sales process depends on you believing the tool works out of the box. Some of them actually think it does. They’ve just never tried running it against a real AEC firm’s file structure.

Think about what a typical contractor is actually sitting on. Specs in PDF, a lot of them scanned from paper originals that are 15 or 20 years old. RFIs living in email threads, in Procore or some other cloud platform, sometimes still on paper logs in a filing cabinet. Submittals are tracked in spreadsheets where every project manager uses a different format. Lessons learned documents that, theoretically, exist but that nobody can actually find. Drawings scattered across Autodesk Construction Cloud, Bluebeam sessions, and local folders that were supposed to sync but didn’t.

Asking an AI to make sense of that is like handing someone a shoebox full of receipts from five different accounting systems and telling them to do your taxes. The tool isn’t broken. You just gave it garbage to work with.

And firms are spending real money on this right now. A 2024 survey from the Royal Institution of Chartered Surveyors showed that builders rate AI’s potential highly, but actual implementation on jobsites keeps lagging way behind expectations. There’s a pattern forming: enthusiastic pilot, quiet cancellation, nobody talks about it.

Fixing the Foundation Problem

The firms I’ve seen actually get value out of AI are doing something much less sexy. They’re cleaning up their data. Working through a checklist that nobody wants to talk about before they go anywhere near an AI tool.

They start by picking one project type and centralizing the docs. Not everything. Not the last five years of every project. Just one type. Highway bridges, K-12 schools, water treatment plants, whatever. Get all the relevant files into one place with a naming convention that actually makes sense.

Then they deal with the optical character regognition issue. If your specs and submittals are scanned PDFs — and in my experience that’s 30 to 50% of what firms over 20 years old are sitting on — then the AI can’t read them. Full stop. Running OCR on old documents is tedious, thankless work. Skip it and your AI is effectively blind to half your institutional knowledge.

After that, they find the one task that eats the most hours. Not the flashiest use case. Not whatever the CEO got excited about at a conference last month. The task where an engineer burns four-plus hours a week basically just searching for information. Spec lookups. Submittal cross-referencing. Safety reporting. That becomes the first AI target, because the ROI is obvious and you can actually measure it.

The step most firms skip? Finding someone inside the organization who genuinely feels the pain. AI pilots mandated from the top, without a project manager or lead engineer who’s actually bought in, stall out every single time. The champion doesn’t need to be technical. They just need to be frustrated enough with the current process to spend a couple hours a week testing the new one.

How to Prepare for AI the Right Way

I watched one firm do this well, a structural engineering group, about 80 people. They spent 10 weeks doing nothing but organizing their steel connection design library into a consistent, searchable format. No AI involved. Just a junior engineer with a spreadsheet and clear naming rules. When they finally connected that cleaned-up dataset to an AI tool, engineers who used to spend 45 minutes hunting for a comparable connection detail were getting answers in under two minutes. It worked because the data was actually ready for it.

Nobody’s going to write a press release about a 10-week file naming project. But that firm got a tool that works. The other firm got a $40,000 invoice and a folder full of meeting notes explaining why the pilot didn’t deliver.

There’s a security angle here too, and I don’t think the industry is paying enough attention to it. I’ve watched engineers at ITAR-regulated aerospace suppliers paste proprietary part specifications directly into ChatGPT because they needed a quick summary and the internal tools were too slow. They’re not being careless. They’re being resourceful. Yet, the compliance exposure is real. Data entered into public AI tools doesn’t stay private in any guaranteed way, and for firms handling ITAR-controlled technical data or operating under strict NDAs, one bad prompt could be a violation. This is happening at companies that have no idea, because nobody thought to ask what their engineers are pasting into browser windows at 10 o’clock on a Thursday night.

 Do the Hard Work First

The answer isn’t banning these tools or pretending the technology doesn’t work. It works. Sometimes remarkably well, but only when the conditions are right. The real move is to get your data foundation in order first, build a secure internal environment where AI can run against your actual project data without shipping it to a third-party server, and scale from there.

The firms that figure this out over the next couple of years are going to have a real edge. Not because they picked better software. Because they did the unglamorous work of making their institutional knowledge accessible and structured. The firms that keep chasing demos are going to keep writing off pilot costs and wondering why this technology that everyone else seems excited about isn’t working for them.

Construction has always been better at building things than at organizing information about what it builds. AI doesn’t fix that problem. It just makes it impossible to ignore.

Alex Ryan is CEO of Ryshe, a Data & AI consulting firm that serves as the Data and AI arm of Wiley|Wilson, a 125-year-old Architecture and Engineering company. He works with AEC, Manufacturing and aerospace firms on data strategy and secure AI implementation.