Which tradies does ChatGPT recommend? A Sydney trades index

When a homeowner asks who to hire, your business can be missing from the answer even while other local operators appear. We wanted a record that a trade business owner could inspect: the actual question, the names returned and the links shown beside them.

This is our Sydney trades AI search index, collected on 16 October 2026 and 17 October 2026. It describes what appeared in the saved answers. We have not assessed the quality of any business listed here.

Which tradies does ChatGPT recommend?

ChatGPT returned 122 distinct business names in 50 completed answers. A name counted when it appeared in the answer as a trade service provider. We excluded adverts, product manufacturers, utilities and businesses appearing only as source labels.

To examine the questions relevant to your work, talk with our AI SEO agency. Start with the saved answer that prompted your question.

The table below shows the most frequent names within each engine’s completed answers. A count is the number of answers containing that name, with a business counted once per answer. It is not a score for the business. Ties are shown together.

The full business appearances file includes the question, trade and order of first appearance for every recorded name. Read the saved answer data before interpreting an individual appearance.

Shortlists across engines

The same homeowner question did not always produce the same names. Among the 47 questions with completed answers from all four engines, 6 had a business name in common across every answer.

Each denominator in this table includes only questions with completed answers from both named engines. “Shared a name” means their answers contained the same normalised business name. An answer without a trade business stays in that denominator. A blocked check or an absent AI Overview does not.

This comparison does not establish why a business appeared. We did not change websites, review profiles or location settings to test a cause. Answers can vary.

Answers by trade

We included 10 trade categories, with 5 questions in each. The questions cover both broad recommendations and specific household jobs. For example, finding someone to replace a porch light is a different request from finding someone to upgrade a switchboard.

These counts show the coverage of our question set. They do not measure how many Sydney businesses exist in each trade or how often homeowners search for them. Several questions came from the same public discussion, so the prompts are related rather than independent consumer observations.

Where the links pointed

We recorded a business mention separately from a source link. A name could appear while the linked evidence came from a directory, review platform or community discussion. A source panel could also contain pages that were not linked directly in the visible answer.

The cells count completed answers with at least one captured source link in that category. An answer can link to several categories, so the columns should not be added into a percentage share. We included Google’s expanded source cards and counted Google business records separately from business websites. Perplexity’s research-only source-panel links are kept in the download and excluded from this table.

We counted an own-website match only when the answer presented an exposed destination as belonging to the named business. The table measures the name and website appearing in the same answer. It does not establish which sentence a grouped citation supports or verify legal ownership of the domain. Directory matches also require an exposed destination that the saved evidence associates with the named business. Both are lower bounds. Categories can overlap and unresolved cases are not evidence that a website was absent.

The citation review records the evidence for each match. A name appearance is counted once per business in an answer. It can include a warning or a statement that a provider may not suit the job, so these counts are not endorsement totals.

These are observed links, not a list of websites we advise every tradie to join. We have not tested whether obtaining a listing on any of them would change an answer. OpenAI also explains that its Sources view may contain cited sources and other relevant links, which is why our capture method keeps those observations apart.¹

Checks with no shortlist

An answer naming no trade business is different from a check that produced no usable answer. The table keeps both visible.

“No business” means the engine completed an answer but did not name a trade service provider. “No Overview” means the Google results page did not show an AI Overview. A block means the engine presented a barrier such as a login request or usage limit. A failed capture is a recording problem. None of those latter outcomes counts as a business being absent from an answer.

We flagged 1 completed answer as a clear market mismatch. Perplexity interpreted the Canterbury lock-change question as Canterbury in England. That frozen prompt did not specify Sydney. An empty provider list in such a response does not show that Sydney locksmiths were missing from a correctly located answer. Some other responses also used overseas sources for places with matching names.

How we ran the study

We froze a set of 50 unbranded questions before collection. The prompts were written from household jobs and recommendation requests found in 27 public Reddit and Whirlpool discussions. They are researcher-normalised questions rather than verbatim search logs. The question set and source register preserves the wording, trade, region and source for each prompt.³

We submitted each prompt once to each available engine: ChatGPT, Google AI Mode, Google AI Overviews and Perplexity. Each started a fresh conversation or search. We did not ask follow-up questions, regenerate answers or insert a business name into a prompt. The saved records identify any combination that was blocked before submission.

We used the existing signed-in Chrome account state. Google displayed Inner West Council, Sydney NSW, from the IP address. ChatGPT and Perplexity did not show an equivalent location label in the answer interface. We did not change location permissions or inspect account memories. ChatGPT showed Pro and Perplexity used its existing Search mode. These conditions limit how closely another person’s run can match ours.

We saved the full visible answer, capture time, names in order and visible source links. We waited for completion before counting a response. Grouped citations were recorded where their destinations were exposed; an unresolved group remains a capture limitation. Names retain the engine’s wording. For comparisons we normalised punctuation, capitalisation and the legal suffix “Pty Ltd”. Other name variants stay separate across answers even when they may refer to the same operator. Shared-name counts can therefore understate shared businesses.

The data bundle includes the records and recompute script. Repeating the arithmetic is possible from the download. Repeating the browser run may produce a different set of answers. The public download omits account identifiers and private conversation links while preserving the visible answer text and source URLs.

What this means for your trade business

Start with the work you want enquiries for. Use a small set of questions that name the job and the area you actually serve. Save the answers before deciding which page needs attention. If an answer asks for more detail, that is a useful question for your service page to answer clearly.

Check the evidence beside each named business. If it points to a service page, inspect how that page describes the job and service area. If it points to a directory or review page, check your own relevant profile for accurate details. These are practical checks suggested by the records, not tested causes of inclusion.

Make your own pages useful to a homeowner comparing options. Explain the jobs you take on, where you travel and how a customer can request a quote. Use genuine project details and publish information you can stand behind. Our Local SEO service covers the local search work around those pages and profiles.

If the shortlist keeps excluding your business, bring the saved questions and answers to our AI SEO agency. We can review the gaps with you and decide which work deserves attention first.

Use the data carefully

This is a convenience sample of questions across Sydney trades. It is not a survey of homeowners, a census of local businesses or a league table of workmanship. We did not verify the prices, reviews, licence claims or availability stated by the engines. Check a contractor independently before hiring; Service NSW provides an official licence lookup.²

The account state was not reset between prompts. Fresh conversations reduce direct carry-over from earlier messages but do not establish that account memory or other personalisation had no effect. The collection also spans time, so engine changes during the run cannot be ruled out. A missing name says only that we did not record it in the answers we captured.

Researchers and journalists can download the question set, answer records, business appearances and calculation script from this page. When citing a result, include the collection date, engine and denominator. Keep the question wording attached to any individual example.

Talk through your gaps

Bring a question your customers ask and the answer you saw. We can use that concrete example to discuss your next SEO step.

Search appearances are not guaranteed.

Talk about your SEO