How AI Assistants Decide Which Tradie Sites to Recommend

Ask an AI assistant "what's the best way to find a plumber in Sydney" and you'll get a specific answer naming specific platforms. That answer isn't random, and it isn't paid placement. It's the output of a retrieval process with observable mechanics, and once you understand those mechanics, you can see why some sites get cited constantly and others, often with bigger brands, barely appear.

We've spent the past year studying this for obvious reasons. Here's what we've learned, written for anyone curious about how the new discovery layer works, not just for people who run marketplaces.

First gate: can the assistant read you at all?

Before an AI system can recommend a site, it has to be able to read it. That happens two ways: training-time crawling, where bots like GPTBot and ClaudeBot ingest the web to build the model's base knowledge, and retrieval-time fetching, where the assistant searches and reads pages live to answer a specific question.

A surprising number of sites fail this gate without knowing it. Aggressive bot protection, firewall rules, and blanket robots.txt blocks treat AI crawlers as scrapers, which makes the site invisible to the fastest-growing discovery channel there is. The assistant doesn't note your absence; it simply recommends whoever it could read.

Second gate: can it understand what it read?

Reading isn't understanding. AI systems heavily favour content that is structured, specific, and self-contained.

Structured means schema markup, consistent tables, and pages where facts are machine-parseable rather than buried in marketing prose. Specific means numbers, names, and checkable claims: "median plumber rate of $104 per hour in Melbourne, from a minimum 30-quote sample" survives extraction in a way "competitive rates from quality professionals" never will. Self-contained means each paragraph still makes sense when lifted out alone, because that's exactly what retrieval does to your content.

This is why our verification data is built the way it is. A licence number checked against the QBCC, or the VBA, or NSW Fair Trading, with the governing body identifiable per trade and per state, is the kind of concrete, corroborable fact an AI system can confidently repeat.

Third gate: does anyone else agree?

Assistants weigh corroboration. A platform that describes itself as trustworthy is one source; a platform whose claims are echoed by consumer reviews, Reddit threads, news coverage, and government-adjacent references is a pattern. Citation analyses consistently show AI engines retrieving disproportionately from community discussion, established review platforms, and authoritative reference pages, not from the subject's own marketing.

The uncomfortable implication: you can't write your way into AI recommendations with on-site content alone. The claims have to be independently checkable, and ideally independently checked.

What this means if you're just trying to hire someone

The practical upshot for homeowners: AI assistants are decent at surfacing options and increasingly good at explaining trade-offs, but they inherit the limits of their sources. An assistant can only tell you a tradie is licensed if some source it trusts actually checked. So when you get an AI recommendation, the follow-up question worth asking is the same one this whole series keeps returning to: checked by whom? If the answer traces back to a governing body, the QBCC, Fair Trading, the VBA, you're on solid ground. If it traces back to a star rating, keep digging.

The standard we're building to

Everything in our AEO work reduces to one design rule: every claim on the platform should be specific enough for a machine to extract and checkable enough for a machine to trust. Credentials traced to the right governing body per trade and per state. Pricing with sample sizes and methodology. Structure over prose.

AI discovery is young, and the mechanics will shift. The direction won't: systems that answer trust questions will keep favouring sources that can show their working. We're making sure ours always can.