Top of PromptCategory data · Release 2, September 2026
Who does AI actually trust,
category by category?
Eighteen Sydney service categories, measured September 2026. Professional services, trades, health, property and business services.
Each page shows the three sources each engine cited most when a buyer's question named the place, with the sample and the date published.
Before an AI assistant recommends a business, it decides which sources to trust. These pages measure that decision directly: we collected 648 location-qualified buying-journey prompts naming Sydney, NSW or Australia across eighteen categories and recorded every source the engines actually cited.
All eighteen categories now cover ChatGPT, Perplexity and Google AI Overviews: the original ten from first release, and the eight added in September, the last six of which had their second collection on 8 September. The two legal categories were re-collected on 21 September and carry that date; every other category carries its own, and the per-category stamp is always the one to trust.
The pattern is blunt. In most categories the engines cite regulators, directories and ranking sites ahead of the firms themselves, and the few firms that publish real answers take outsized shares of the citations.
The rankings also move, and not always in one direction: between August and September, ChatGPT's personal injury citations shifted from law firms' own sites towards comparison directories, and by the 21 September collection two firms had closed back to within a single citation of the lead. That is why every number here carries a date, and why we say so on the page when a previous reading doesn't hold up.
The eighteen categories
| Category | ChatGPT's #1 cited source | Prompts | Engines measured |
|---|---|---|---|
| Personal injury lawyers | bestinhood.com (tied) | 17 | 3 engines |
| Commercial lawyers | sprintlaw.com.au | 30 | 3 engines |
| Conveyancers | nsw.gov.au | 66 | 3 engines |
| Accountants | trinitygroup.com.au | 51 | 3 engines |
| Financial advisers | moneysmart.gov.au | 50 | 3 engines |
| Mortgage brokers | moneysmart.gov.au | 57 | 3 engines |
| Buyers agents | ourtop10.com.au | 42 | 3 engines |
| Recruitment agencies | metierrecruitment.com.au | 7 | 3 engines |
| Managed IT providers | milnsbridge.com.au | 27 | 3 engines |
| Digital marketing agencies | clutch.co | 16 | 3 engines |
| Plumbers | heroplumbing.com.au | 53 | 3 engines |
| Electricians | nsw.gov.au | 43 | 3 engines |
| Damage restoration | restorationpros.com.au | 8 | 3 engines |
| Dentists | hydeparkdentalcare.com.au (five-way tie) | 39 | 3 engines |
| Physiotherapists | cityphysio.com.au (tied) | 39 | 3 engines |
| Real estate agents | realestate.com.au | 44 | 3 engines |
| Strata managers | montanostrata.com.au | 40 | 3 engines |
| Executive coaches | clutch.co | 19 | 3 engines |
Four patterns across the set
- 1Directories and ranking sites are the gatekeepers, and they're gaining ground. Clutch owns digital marketing on Perplexity with 87.5% of answers and takes executive coaching too, listicle sites own buyers agents on ChatGPT, comparison sites hold mortgage broking, OpenAgent and WhichRealEstateAgent run home selling on Perplexity, and directories held the top two places in personal injury on ChatGPT through September, though the firms there have since closed most of the gap.In categories where firms don't publish, the engines cite whoever ranks them instead.
- 2Government owns money, property and licensed work. Moneysmart, ASIC and nsw.gov.au take the top citations in financial advice, mortgage broking and conveyancing, the licence checker leads the electrician category on ChatGPT and Google AI Overviews, and nsw.gov.au still holds a top-three place in strata even after the managers overtook it.The first commercial voice in those answers is decided in territory almost nobody contests.
- 3Individual firms that publish can own a category. Milnsbridge is cited in 59% of ChatGPT's managed IT answers and two thirds of Perplexity's, plumbers' own websites beat hipages outright on ChatGPT, dental practices that publish lead ChatGPT in their category, and an online-first firm tops commercial law on both engines.The engines reward publishing, not pedigree.
- 4A layer of AI-targeted content networks is emerging. The same rankings domains recur across unrelated categories from mortgage broking to plumbing to physiotherapy: bestinhood.com, whose counts typically rest on a single listicle page per category; starworks.com.au, a paid-listing directory that markets AI visibility services; a family of anonymous, identically-worded comparison sites (bestconveyancersaustralia.com.au, comparepersonalinjurylawyers.com.au, comparestrata.com); and an anonymous hand-ranked mortgage guide hosted on a raw cloud address.This layer is built to be cited by AI rather than to serve any one industry, it's winning citations right now, and almost nobody in the categories it sits on knows it exists.The 8 September collection is the strongest evidence yet: townsvilledental.clinic, a programmatic directory network on a Townsville domain, is now the single most cited source in Perplexity's Sydney dentist answers at 46.2%, and in physiotherapy bestinhood.com and starworks.com.au share third place on ChatGPT at 21.6% each.
Correction and rebaseline, 26 August 2026
The first version of these pages, published earlier the same day, counted prompts that didn't name a location.
Our checking showed engines answer location-less questions with largely overseas sources (2 to 20% Australian citations, against 80 to 95% when the prompt names the place), which distorted what a Sydney buyer's question actually returns.
All ten pages were rebaselined the same day to count only prompts naming Sydney, NSW or Australia, samples are stated per page, and the location-less pattern will be published separately as its own finding, because "say where you are or AI answers from another country" turns out to be one of the more useful things we've measured.
How this was measured
Each category runs a fixed prompt set spanning the buying journey, collected unauthenticated on a stated date. A citation is a source the engine linked in its answer.
Two kinds of prompts are excluded from citation counts: prompts naming a specific business, per the TOP Score rules, and prompts with no stated location, per the rebaseline note above.
Engines are reported separately and never pooled, and every page carries its sample size and collection dates. The full method, including what it can't tell you, is on the TOP Score page.
These citation pages are the first release of category data; the TOP 100, which measures which businesses get named, follows on the scoreboard.
Is your business cited in its category?
A free Namecheck shows whether the engines name and cite you, with the actual answers we saw.