How to Get AI Citations for Tradies

Someone asks ChatGPT why their hot water has gone cold while the pilot light is still burning. The answer comes back as a few paragraphs of explanation with small links underneath, or numbered through the text. Those links are citations: your page was used as a source.

The kind of question matters, and this is where most tradie sites aim at the wrong target. Citations are won on informational questions: how something works, why it fails, what it costs, whether it is a job for a licensed tradie. Ask an AI who to call in Joondalup on a Sunday and you get names rather than sources, because that is a recommendation, which is a different job with different levers. Citations come from the questions people ask before they have decided to hire anyone.

The dullest part of the job, and the one that stops most tradie sites dead. You cannot be cited by a system that never read the page. Check three things first.

Your robots.txt

There are three kinds of AI bot. Training crawlers like GPTBot and Google-Extended collect content to train models. Search crawlers like OAI-SearchBot build the index an AI searches when it answers. Fetchers like ChatGPT-User grab a page because somebody in a chat just asked about it. Blocking the training bots while allowing the search ones is a legitimate position. Blocking the lot is how you disappear.

Your CDN or firewall

This one catches people out, because it is not in the file they checked. Cloudflare and similar services have added one-click controls to block AI crawlers, and some turned them on by default. Your robots.txt can be perfectly welcoming while the bot is being turned away before it ever reaches your server. If a web designer set this up years ago, go and look at the dashboard yourself.

Whether your content exists without JavaScript

Some builders assemble the page in the browser, so the text a customer reads is not in the file a bot downloads. Turn JavaScript off and load your own service page. If the copy vanishes, that is roughly what the AI ecosystem sees.

Query fan-out is the mechanism behind everything above, and it is the part worth understanding properly because it changes what you write. When someone asks an AI a real-world question, it does not run that one search. It fans the question out into several smaller ones, searches those separately, and assembles an answer from whatever ranks for them.

So a homeowner asking whether to repair or replace a twelve-year-old gas hot water system fans out into half a dozen searches: how long one lasts, what replacement costs, whether a repair is worth it at that age, how long the job takes, whether a plumber is required. The AI reads what ranks for those, not for the question the person typed. Which means you are not competing for the question. You are competing for its parts.

1. 1

   Write down the real question a customer asks you

   Not the keyword. The thing they say on the phone. 'Is it worth fixing or should I just replace it?' is the question; nobody types it into Google that way.

2. 2

   List the smaller questions it contains

   Cost, lifespan, timeframe, whether it is a DIY job, whether it needs a licensed tradie, what goes wrong if you leave it. Google's People Also Ask boxes are a free source of these.

3. 3

   Answer them on one strong page, not six thin ones

   Expanding a page you already have beats publishing a new page per question. One page that covers the whole decision has more chance of being the source for several sub-questions at once.

4. 4

   Check what currently ranks for each sub-question

   If a competitor owns three of the six, you know exactly what you are writing against and where the gap is.

Generic advice is everywhere, and an AI has no reason to cite the four hundredth page telling homeowners to get three quotes. What it cannot generate is checkable detail from someone who does the work — the one advantage a working tradie has over every marketing site in the country.

• Your actual call-out fee, and what it includes, rather than 'competitive rates'
• How long a job genuinely takes in a 1970s house versus a new build
• What a repair typically costs in your area, as a range, with what moves it up or down
• The specific conditions that catch people out locally: limestone, reactive clay, salt air, old copper, a suburb with pressure problems
• What you find on the majority of jobs of a given type, which is real field data nobody else has
• Which brands or parts you refuse to install and why, said plainly

It is also the safest content to write, because it is true and you can defend it. If a number changes, update the page.

An AI assembling an answer is looking for a passage it can take cleanly. Give it one. Make the heading the question, in the words a person would use, then answer it directly in the first forty to sixty words underneath before you go into detail.

Two things worth saying plainly, because both are sold as AI strategy right now.

Mass-producing suburb pages gives an AI nothing to cite, because there is no answer in them, and Google's spam enforcement has sharpened on exactly that pattern.

Nor is there a hidden setting. No schema tag, meta tag or file makes an AI cite you. Structured data helps a machine understand a page that already says something useful; it cannot rescue one that does not.