Claude vs ChatGPT, like Gemini vs ChatGPT, is usually argued over writing, coding and price, but for an Australian business the question that matters is simpler: when a customer asks each one who to hire, who does it name? On 24 September 2026 we asked ChatGPT, Claude, Gemini and Perplexity the same 24 Australian buying questions, and they mostly named different businesses. Claude and ChatGPT agreed least of all, sharing just 11% of the names they gave. Across 394 businesses named, only 11 were picked by all four, and 272 were picked by just one.
Then we asked 8 of those questions a second time. Perplexity gave back almost the same shortlist. The other three, Claude most of all, changed a third to a half of their names.
Key findings
- Only 11 of 394 businesses were named by all four assistants across 24 questions. 272 (69%) were named by only one.
- The most any two assistants agreed was 23% of names (Gemini and Claude). ChatGPT and Claude agreed on just 11%.
- Perplexity cited 15.1 sources per answer. ChatGPT cited 4.5, but ran 184 background searches across the 24 questions, and 49% of them were site: searches aimed at guides and registers such as Doyle’s Guide, HIA and QBCC.
- Asked the same question twice, Perplexity repeated 94% of its business names. Gemini repeated 65%, ChatGPT 55% and Claude 49%.
Why this comparison tests hiring questions, not essays
Most comparisons score these tools on tasks such as drafting an email or fixing code. That helps if you are choosing which one to pay for, but a business owner sits on the other side of the screen. When someone in Melbourne asks for a family lawyer, each assistant runs its own web searches, reads its own pages and writes its own shortlist. The business that gets named gets the enquiry.
So we tested that. Our 24 questions were the kind people type when they are ready to hire: 6 about lawyers, 6 about builders, 6 about marketing agencies and 6 about trades and professional services, each ending with “Give me a shortlist with reasons.” Every assistant had web search switched on and an Australian location. We recorded every business named and every website cited, then compared them with Google’s top 20 results and map pack for the same search. Our AI SEO services page explains how we use this kind of testing for clients.
ChatGPT vs Claude vs Gemini vs Perplexity at a glance
Here is the whole study in one table, built from the same 24 questions. If you came for Gemini vs ChatGPT specifically, compare the first two columns: the two named a similar number of businesses but found them in very different places.
| Measure | ChatGPT | Gemini | Perplexity | Claude |
|---|---|---|---|---|
| Businesses named per answer | 5.6 | 5.3 | 6.8 | 6.0 |
| Sources cited per answer | 4.5 | 6.1 | 15.1 | 5.8 |
| Citations that were business websites | 59% | 58% | 61% | 72% |
| Citations that were directories, reviews or ranking guides | 21% | 22% | 31% | 18% |
| Citations from government and industry bodies | 15% | 3% | 1% | 1% |
| Cited sites also in Google’s top 20 | 20% | 47% | 35% | 58% |
| Google map pack businesses it named | 22% | 28% | 25% | 19% |
| Its picks that no other assistant named | 57% | 36% | 51% | 46% |
Three things jump out. Perplexity reads far more than anyone else. ChatGPT is the most independent, with 57% of its picks named by nobody else, and it leans on government and industry bodies five times harder than the others. And Claude and Gemini stay closest to Google: more than half of Claude’s cited websites were also in Google’s top 20 for the same search, against one in five for ChatGPT.
They rarely name the same businesses
If the four assistants were reading the same evidence and reaching the same view, their shortlists would overlap heavily. They do not. We measured overlap for every pair as the share of names two assistants had in common out of all the names either of them gave (the Jaccard index), averaged across the 24 questions.
| Pair | Names in common |
|---|---|
| Gemini and Claude | 23% |
| Gemini and Perplexity | 19% |
| Perplexity and Claude | 18% |
| ChatGPT and Gemini | 14% |
| ChatGPT and Perplexity | 14% |
| ChatGPT and Claude | 11% |
Of the 394 businesses named, 42 were named by three or more assistants. A business that appears in one assistant’s answer is, more often than not, invisible in the other three.
Agreement also depends on the industry. Law had the most agreement, at 27% average overlap, and 7 of the 11 businesses named by all four were law firms, such as Coote Family Lawyers, Maurice Blackburn and MKI Legal. Trades had the least, at 12%, with 75% of trade businesses named by only one assistant. The reason is fairly clear from the citations: law has a shared referee, Doyle’s Guide, that several assistants read. A Melbourne electrician has no equivalent, so each assistant builds its list from whatever directory or review site it happens to land on.
How many sources each assistant reads
The quickest way to see how differently these tools work is to count what they cite.
Fewer citations does not mean less reading, though. Each assistant has a distinct habit, and the habit decides who it names.
ChatGPT goes looking for the referee
ChatGPT showed us the searches it ran before answering. Across the 24 questions it ran 184 of them, about 8 per answer, and 91 (49%) were site: searches, which only look inside one named website. The most searched sites were hia.com.au (9 searches), doylesguide.com (8), qbcc.qld.gov.au (5) and solarquotes.com.au (5). Some real examples from the study:
- site:doylesguide.com “Employment Law Firms” “Western Australia, 2026” MKI MDC (for the Perth employment lawyer question)
- site:liv.asn.au accredited specialist family law Melbourne directory (family lawyers, Melbourne)
- site:qbcc.qld.gov.au AMGCUBE 15616213 (checking a granny flat builder’s licence number in Brisbane)
- site:hia.com.au Perth renovation addition Humphrey Homes 2025 (renovation builders, Perth)
- site:solarquotes.com.au/installer-review Koala Solar reviews (solar installers, Perth)
The pattern is that ChatGPT finds a shortlist on a trusted guide or register, then checks each name. That is why only 20% of its citations were in Google’s top 20: it is not reading the results page, it is reading the referee. For a business, that means your licence number, your accredited specialist listing, your award and your guide ranking are what get you on ChatGPT’s list. We cover the mechanics in more depth in our guide on how ChatGPT decides which businesses to recommend.
Gemini vs ChatGPT: Gemini stays close to Google
Gemini ran about two plain searches per answer, never used site:, and 47% of its citations were in Google’s top 20. It named 28% of the businesses in Google’s map pack, the highest of the four. It cited Doyle’s Guide in 5 of 24 answers, and Reddit and Localsearch in 4 each. Of all four assistants it was the least independent: only 36% of its picks were named by nobody else. If you rank well on Google and have a strong Google Business Profile, Gemini is the assistant most likely to reflect that.
Perplexity reads the directories
Perplexity cited 362 sources across the 24 questions. Directories and review sites were 31% of them, more than any other assistant. The most cited were Localsearch and hipages (7 answers each), Oneflare and Airtasker (6 each), and Best Lawyers, LawConnect, ProductReview and Clutch (5 each). If you are listed properly on the directories for your industry, Perplexity is where that pays off. There is more on this in our post on getting cited by Perplexity.
Claude reads business websites
72% of Claude’s citations were the businesses’ own websites, and 58% of everything it cited was also in Google’s top 20, the highest overlap of the four. For Claude, a clear, specific website that ranks is most of the job (more on Claude below). Our study of what AI can quote from Australian homepages shows how few sites publish them.
Ask twice and you get a different shortlist
A single answer is one draw. We had already measured this for Google AI Mode, where two identical searches shared only 24% of their sources (see why AI answers change every time you ask). For this post we tested the four chat assistants. On 24 September 2026 we ran 8 of the 24 questions a second time with identical settings: two each from law, builders, agencies and trades. Then we checked how many of the businesses named in the first run came back in the second.
Perplexity was close to fixed. It cited 114 of the same 115 sources in both runs and repeated 49 of its 52 business names. On three of the eight questions its shortlist was identical. The others moved a lot:
- Gemini repeated 65% of its names, but on the Melbourne custom home builder question only 1 of its 5 names came back.
- ChatGPT repeated 55%. Only 4 of its 64 background searches in the second run were word for word the same as the first, even though the share that were site: searches stayed high (26 of 60, then 31 of 64). It keeps going back to the same kind of source, but writes a new set of searches each time.
- Claude repeated 49%, and only 13 of its 24 first-run top three picks appeared anywhere in the second answer. On the Melbourne SEO agency question it named 5 agencies, then 14, with 1 in common.
The top of the list was steadier than the tail. Across all four, the businesses named first were more likely to come back than the ones named fifth or sixth. In practice a business that is a strong fit keeps appearing, and a borderline one flickers in and out. That is why we measure visibility as a rate across repeated checks, as explained in our guide to tracking AI visibility properly.
Claude vs ChatGPT for business recommendations
Claude vs ChatGPT is the comparison Australians search for most, and for business recommendations it is also the pair that agreed least in our study. They shared only 11% of the businesses they named. That is not because one is right and the other is wrong. They find businesses in almost opposite ways, so the same business can be on one list and missing from the other.
| What we measured | Claude | ChatGPT |
|---|---|---|
| Businesses named per answer | 6.0 | 5.6 |
| Sources cited per answer | 5.8 | 4.5 |
| Citations that were business websites | 72% | 59% |
| Cited sites also in Google’s top 20 | 58% | 20% |
| Background searches per answer | About 2, all plain searches | About 8, half of them site: searches |
| Names repeated on a second identical run | 49% (23 of 47) | 55% (26 of 47) |
| First-run top three picks still named in run 2 | 13 of 24 | 18 of 24 |
| Questions it declined to answer with names | 2 (local trades and health) | 0 |
Claude follows Google, ChatGPT follows the referee
58% of the websites Claude cited were also in Google’s top 20 for the same search, the highest of the four assistants, and 72% of its citations were the businesses’ own sites. ChatGPT’s figure for Google’s top 20 was 20%. ChatGPT went to the referee instead: Doyle’s Guide for lawyers, HIA for builders, the QBCC register for Queensland trades. In short, Claude mostly rewards a business website that already ranks. ChatGPT mostly rewards third-party proof it can look up.
Claude changes its mind more
When we asked the same 8 questions twice, Claude repeated 49% of its names, the lowest of the four. ChatGPT repeated 55%. Claude also moved its top picks more: 13 of its 24 first-run top three picks were named again, against 18 of 24 for ChatGPT. So a single check in Claude tells you even less than a single check in ChatGPT.
Claude will sometimes say no
Claude declined two local questions outright, a Gold Coast dentist in the main study and a Melbourne electrician in our second run. Both times it said it could not verify local providers and pointed the user to review sites and licence registers instead. ChatGPT named businesses for every question we asked. For trades and health businesses, that means Claude may send a customer to hipages, ProductReview or a Google search rather than to you, so those listings matter even for Claude.
What this means for your business
If you want to show up in both, you need both kinds of evidence. For Claude: a service page that ranks in Google for your main search and says specific things in plain text. For ChatGPT: your licence, accreditation, awards and industry guide rankings, stated on third-party sites and repeated on your own. Search interest in the comparison has jumped too. Search volume data for Australia shows “claude vs chatgpt” at about 5,400 searches a month, up 312% on a year earlier, while “gemini vs chatgpt” sits at 1,300 and “perplexity vs chatgpt” at 720, both lower than a year ago.
Perplexity vs ChatGPT: the other big gap
Perplexity vs ChatGPT is the other pair with sharp differences, mostly in how much each reads. Perplexity cites 15.1 sources per answer, ChatGPT 4.5. Perplexity repeated 94% of its names on a second run, ChatGPT 55%. Perplexity’s citations lean on consumer directories (31%), while ChatGPT’s lean on official bodies and peer guides (15% from government and industry bodies, against 1% for Perplexity).
For a business, ChatGPT vs Perplexity comes down to where your proof lives. If your strongest evidence is a pile of reviews on hipages, ProductReview or Localsearch, Perplexity is more likely to find it. If your strongest evidence is a licence, an accreditation, an industry award or a Doyle’s Guide ranking, ChatGPT is more likely to go and look it up.
Which one should you care about as a business?
You do not have to pick one, because most of the work helps all four. (Paid placement is a separate question, covered in our look at advertising inside ChatGPT in Australia.) This table is the practical version of everything above.
| Assistant | What it cites most | Sources per answer | How to get in |
|---|---|---|---|
| ChatGPT | Business sites (59%), then official registers, industry bodies and peer guides (HIA, Doyle’s Guide, QBCC, LIV, SolarQuotes) | 4.5, after about 8 background searches | Be on the register, guide or awards list for your industry, and publish your licence number and awards on your own site so it can confirm them |
| Gemini | Business sites (58%) and Google-visible pages; Doyle’s Guide, Reddit, Localsearch | 6.1 | Rank in Google’s top 20 and in the map pack. Keep your Google Business Profile complete and your reviews current |
| Perplexity | Directories and review sites (31%): Localsearch, hipages, Oneflare, ProductReview, Clutch, LawConnect, Best Lawyers | 15.1 | Complete, consistent profiles with real reviews on the directories for your trade. Once in, you tend to stay in |
| Claude | Business websites (72%), mostly ones already in Google’s top 20 | 5.8 | A service page that ranks and states specific, checkable facts in plain text: areas, years, licences, prices |
Two practical notes. First, every one of these assistants has to be able to fetch your site. OpenAI says its OAI-SearchBot crawler is what surfaces websites in ChatGPT search, and Perplexity explains how PerplexityBot follows robots.txt. A security plugin that blocks them quietly removes you. Second, Google says its AI features rely on normal Search systems, and its guidance on AI features and your website asks for nothing special beyond good SEO. That matches our data for Gemini and Claude.
If your customers are lawyers’ clients, the referee matters most, and our post on how law firms get recommended by ChatGPT goes further. Builders face a different mix of HIA awards and review sites, covered in how AI recommends builders.
A note on Uprise Digital in these results
This is our own blog, and six of the 24 questions were about marketing agencies like us, so it is fair to say how we showed up. Uprise Digital was named by Perplexity and Claude for the question about agencies that help businesses get recommended by ChatGPT and Google AI, and by Claude for the question about marketing agencies for builders. No assistant named us for any of the other 22 questions, including the SEO agency questions for Melbourne and Brisbane and the law firm marketing question. In the second run of the AI agency question, Perplexity and Claude named us again, and ChatGPT and Gemini did not. We did not change anything on our site between runs. We report this as a disclosure, not a result, and we did not give ourselves any special treatment in the analysis.
How we did this research
On 24 September 2026 we ran 24 Australian buying questions (6 law, 6 builders, 6 marketing agency, 6 trades and professional services) through the API versions of ChatGPT, Gemini, Perplexity and Claude with web search switched on and an Australian location where the assistant supports one. These are not the consumer apps, which can add personalisation and memory. Each answer was run once. Business names were normalised across assistants, and each cited website was classed by type (business, directory or review site, government or industry body, and so on). Citations were matched at domain level against Google’s top 20 organic results and map pack for the matching keyword, searched from Australia on desktop.
For the stability test we re-ran 8 of the 24 questions (family lawyers Melbourne, employment lawyer Perth, custom home builders Melbourne, granny flat builders Brisbane, SEO agency Melbourne, AI SEO agency Australia, electrician Melbourne, solar installers Perth) on all four assistants with identical settings on the same day. We extracted the businesses from both runs with the same method so the spelling matched, and counted how many first-run names came back. The limits: 8 questions is a small sample, two runs is the minimum, and the stability we saw may change as these tools update their search layers. Search volumes are Google estimates for Australia.
Frequently asked questions
Is Gemini actually better than ChatGPT?
For writing or coding that depends on the task. For recommending Australian businesses, neither is better, they are different. In our study Gemini stayed closer to Google’s results (47% of its sources were in Google’s top 20, against 20% for ChatGPT), while ChatGPT searched industry guides and registers directly. They agreed on only 14% of the businesses they named.
Is Claude better than ChatGPT for finding a business?
Neither is better across the board. In our study Claude stayed closer to Google, with 58% of its sources in Google’s top 20 against 20% for ChatGPT, while ChatGPT checked licences, awards and industry guides directly. Claude was also less consistent, repeating 49% of its names on a second run, and it declined two local trade and health questions that ChatGPT answered.
What are the disadvantages of Perplexity?
It leans heavily on directories and review sites, which were 31% of its citations, and a directory ranking is only as good as that directory’s own methods. Businesses that are not on those directories can be left out even when they are well known. It is also very stable, which is good if you are on the list and hard to change if you are not.
Does ChatGPT use Google results?
Not directly in the way Gemini appears to. Only 20% of ChatGPT’s citations in our study were in Google’s top 20 for the same search. It runs its own searches, often aimed at one website at a time, such as Doyle’s Guide, HIA or the QBCC licence register, and builds its list from those.
If you would like to know how each of these assistants currently describes your business, and which sources are feeding your competitors’ names, that is where our generative engine optimisation work starts. We run your real customer questions through all four, more than once, and show you what to fix first.
