Our roof started showing its age last autumn. Nothing catastrophic, a few slipped slates and some flashing that had seen better days, but enough that we knew we needed someone in before the winter rain arrived in earnest. We're based just outside Cork city, and we figured this would be a good chance to test something we'd been curious about for a while: how useful are AI tools when you actually need them for something hyperlocal and practical?
So we put the same prompt into four tools. Something along the lines of: "Can you recommend a reliable roofing contractor in Cork, Ireland? I need someone who can repair slipped slates and repoint flashing on a semi-detached house."
The results were, honestly, all over the place.
ChatGPT: Confident but vague
ChatGPT gave us a response that felt thorough at first glance. It talked about what to look for in a roofer, mentioned checking for Registered Gas Installers and other trade bodies (which aren't relevant to roofing, so that was odd), and then said it couldn't recommend specific tradespeople but suggested we try Google Maps and local Facebook groups. Useful general advice, maybe. But we already knew all of that. The confidence of the delivery didn't match the lack of actual local knowledge underneath it.
It did mention the Register of Contractors for Waterproofing and Flat Roofing, which is a real thing in Ireland, so that was a small plus. But the overall experience felt like asking someone who has read a lot about Cork without ever having been there.
Perplexity: Interesting, but verify everything
Perplexity was more interesting. It actually pulled in some named businesses and gave us what looked like real contact details, with citations linking to sources. That sounds great until you start checking. One of the businesses it named had a phone number that was disconnected. Another appeared to have closed or changed hands based on what we could find online. A third had reviews that were a few years old with nothing recent.
The citations gave Perplexity an air of reliability, but local business data goes stale fast. Tradespeople retire, change numbers, go sole trader, merge with someone else. Perplexity is pulling from web sources that may have been accurate eighteen months ago. For something like finding a roofer, that gap matters.
Gemini: Regional but shallow
Gemini managed to stay focused on Cork, which felt like progress. It gave a few names, mentioned Munster more broadly at one point, and generally kept its answer geographically relevant. But the names it gave were fairly generic, and when we searched for them it was hard to tell if they were real businesses or composite descriptions based on review site scraping. There was no obvious hallucination, but there was also nothing we could immediately act on. It felt like the kind of answer you'd get from someone who had skimmed a few pages of results without reading any of them closely.
Claude: Honest about its limits
Claude was the most upfront about what it couldn't do. It was clear that it doesn't have reliable access to current local business listings and that recommending specific tradespeople wasn't something it could do responsibly. It then did something the others mostly skipped: it gave us a structured way to search ourselves, including what to ask a roofer before hiring, what to check on a written quote, and which Irish organisations might have verified contractor lists.
We actually found that more useful in practice. It was slower to feel helpful, but the advice held up. If you're starting from scratch on how to write a proper spec for a builder or contractor, that kind of structured guidance is genuinely worth something.
What we actually learned
None of the four tools found us a roofer. We ended up doing what most people in Ireland do: asked a neighbour who had recently had work done, got a name, checked their previous jobs, and went from there. The roofer we hired wasn't listed anywhere online in any meaningful way.
That's the gap none of these tools can fully bridge yet. Hyperlocal trades in Ireland still run heavily on word of mouth, and no language model has access to the conversation your next-door neighbour had with someone on a GAA WhatsApp group last March.
Where AI tools do help is in the surrounding work: knowing what questions to ask, understanding what a fair quote looks like, spotting red flags in a contract. We've found them genuinely useful for that kind of preparation. Getting the name in the first place, though, that's still very much a human job.