Study
How many Australian celebrants can couples actually find online?
How many Australian marriage celebrants can couples find online?
Appear by creativebyjake. A study of the register of Commonwealth-registered marriage celebrants, read 8 October 2026; Google Maps and celebrants' websites read 13 and 14 October 2026.
About one in five. Of 400 celebrants drawn at random from the register, 19.0% have a Google Maps listing that a couple searching their name will find and that is provably theirs (95% interval 15.5% to 23.1%). Among the celebrants that ChatGPT and Claude name when a couple asks for one, it is 73.2%. The named celebrants also have far more Google reviews, and far more often a website.
Every table below is published as data.csv, under CC BY 4.0.
Five findings
- 19.0% of registered celebrants have a Google listing a couple can find by their name (76 of 400; 95% interval 15.5% to 23.1%). That is a floor: counting every brand listing our searches turned up that we could not tie to anyone as theirs, it is at most 23.5% (94 of 400).
- 73.2% of the celebrants the AI assistants named have one (112 of 153; 65.7% to 79.6%): three to four times the rate of the register as a whole.
- The named celebrants have about seven times the reviews, at the median. A named celebrant with a listing has a median of 81 Google reviews; a randomly chosen celebrant with a listing has 11.5. 67.0% of the named have 50 or more (57.8% to 75.0%), against 25.0% (16.6% to 35.8%).
- 76.5% of registered celebrants have neither a listing a couple can find nor a website address on their listing or at their email domain (306 of 400; 72.1% to 80.4%). Among the named, 20.9%.
- 75.2% of the named have a website we could reach, against 18.2% of the register (115 of 153; 73 of 400). Once a celebrant has a site, the sites differ less: 43.5% of the named sites describe the business in structured data (the machine-readable summary search engines read), against 34.2%, and the two intervals overlap.
In 100 answers across 25 Australian cities and regions, ChatGPT and Claude named at least 153 distinct registered celebrants: at least 1.4% of the 10,846 active celebrants on the register.
The two groups side by side
Each cell: the share, the count and its base, and the 95% interval.
| Measure | Random sample of the register | Named by ChatGPT or Claude |
|---|---|---|
| Has a Google Maps listing (of those read) | 19.0% (76 of 400; 15.5% to 23.1%) | 73.2% (112 of 153; 65.7% to 79.6%) |
| Listing found under their own surname (of those read) | 13.8% (55 of 400; 10.7% to 17.5%) | 56.2% (86 of 153; 48.3% to 63.8%) |
| Listing found under a brand name confirmed as theirs (of those read) | 5.2% (21 of 400; 3.5% to 7.9%) | 17.0% (26 of 153; 11.9% to 23.7%) |
| Ceiling: a listing, counting every unproven brand lead as theirs (of those read) | 23.5% (94 of 400; 19.6% to 27.9%) | 76.5% (117 of 153; 69.2% to 82.5%) |
| Listing has a website attached (of those listed) | 81.6% (62 of 76; 71.4% to 88.7%) | 96.4% (108 of 112; 91.2% to 98.6%) |
| Listing not claimed by the owner (of those listed) | 10.5% (8 of 76; 5.4% to 19.4%) | 0.9% (1 of 112; 0.2% to 4.9%) |
| No reviews (of listings with a readable count) | 22.4% (17 of 76; 14.5% to 32.9%) | 2.7% (3 of 112; 0.9% to 7.6%) |
| 10 or more reviews (of listings with a readable count) | 53.9% (41 of 76; 42.8% to 64.7%) | 87.5% (98 of 112; 80.1% to 92.4%) |
| 50 or more reviews (of listings with a readable count) | 25.0% (19 of 76; 16.6% to 35.8%) | 67.0% (75 of 112; 57.8% to 75.0%) |
| Has a website address (on their Google listing, or a celebrant's page at their email domain) | 20.0% (80 of 400; 16.4% to 24.2%) | 76.5% (117 of 153; 69.2% to 82.5%) |
| Has a website we could reach (its home page answered) | 18.2% (73 of 400; 14.8% to 22.3%) | 75.2% (115 of 153; 67.8% to 81.3%) |
| Neither a listing nor a website address (of those read) | 76.5% (306 of 400; 72.1% to 80.4%) | 20.9% (32 of 153; 15.2% to 28.0%) |
| Register email not on a free mailbox (their own or a workplace domain) | 16.0% (64 of 400; 12.7% to 19.9%) | 49.7% (76 of 153; 41.9% to 57.5%) |
| Measure | Random sample | Named by ChatGPT or Claude |
|---|---|---|
| Median Google reviews (listings with a readable count) | 11.5 (n 76) | 81.0 (n 112) |
| Median Google rating (listings with a rating) | 5.0 (n 59) | 5.0 (n 109) |
| Listings whose review count could not be read | 0 (n 76) | 0 (n 112) |
| Rows whose Maps search could not be read | 0 (n 400) | 0 (n 153) |
The ceiling row counts every not-listed celebrant whose search turned up a brand listing we could not tie to them, unless the business names register shows someone else holds that brand. It covers only the brand leads our searches turned up (up to two brand candidates a search), so it is a ceiling on those, not on every listing that might exist.
A website address is the one on their Google listing, or a page at their email domain that shows celebrant work or names them. A website we could reach is one whose home page answered when we read it.
Their websites
The home page of every website we could reach, checked for eleven basics.
| Website check (of sites checked) | Random sample | Named by ChatGPT or Claude |
|---|---|---|
| Page has a title | 78.1% (57 of 73; 67.3% to 86.0%) | 81.7% (94 of 115; 73.7% to 87.7%) |
| Page has a search description | 43.8% (32 of 73; 33.0% to 55.2%) | 57.4% (66 of 115; 48.3% to 66.0%) |
| Business described in structured data | 34.2% (25 of 73; 24.4% to 45.7%) | 43.5% (50 of 115; 34.8% to 52.6%) |
| Has a sitemap | 84.9% (62 of 73; 75.0% to 91.4%) | 94.8% (109 of 115; 89.1% to 97.6%) |
| Has robots.txt | 90.4% (66 of 73; 81.5% to 95.3%) | 98.3% (113 of 115; 93.9% to 99.5%) |
| Served over https | 89.0% (65 of 73; 79.8% to 94.3%) | 95.7% (110 of 115; 90.2% to 98.1%) |
| Phone number is tappable | 31.5% (23 of 73; 22.0% to 42.9%) | 46.1% (53 of 115; 37.3% to 55.2%) |
| Has a social preview | 79.5% (58 of 73; 68.8% to 87.1%) | 90.4% (104 of 115; 83.7% to 94.6%) |
| Page states its canonical web address | 89.0% (65 of 73; 79.8% to 94.3%) | 93.9% (108 of 115; 88.0% to 97.0%) |
| Set up for phones | 93.2% (68 of 73; 84.9% to 97.0%) | 92.2% (106 of 115; 85.8% to 95.8%) |
| Every image described | 21.9% (16 of 73; 14.0% to 32.7%) | 13.9% (16 of 115; 8.7% to 21.4%) |
By state
The random sample by state. The intervals overlap: no state differs clearly from the national figure. The ACT, Tasmania and the Northern Territory have fewer than 20 celebrants in the sample, so only counts are given.
| State | In the sample | Listed | Share listed (95% interval) | With a website | Share with a website (95% interval) |
|---|---|---|---|---|---|
| ACT | 8 | 3 | too few to say | 4 | too few to say |
| NSW | 124 | 23 | 18.5% (12.7% to 26.3%) | 24 | 19.4% (13.4% to 27.2%) |
| NT | 2 | 1 | too few to say | 0 | too few to say |
| QLD | 87 | 14 | 16.1% (9.8% to 25.2%) | 15 | 17.2% (10.7% to 26.5%) |
| SA | 24 | 3 | 12.5% (4.3% to 31.0%) | 3 | 12.5% (4.3% to 31.0%) |
| TAS | 8 | 0 | too few to say | 0 | too few to say |
| VIC | 107 | 26 | 24.3% (17.2% to 33.2%) | 27 | 25.2% (18.0% to 34.2%) |
| WA | 40 | 6 | 15.0% (7.1% to 29.1%) | 7 | 17.5% (8.7% to 32.0%) |
The whole register by state:
| State | Active celebrants | Published an email | Published a mobile | Email not on a free mailbox | Registered since 2020 |
|---|---|---|---|---|---|
| ACT | 222 | 75.2% | 67.1% | 14.9% | 60.4% |
| NSW | 3,330 | 77.8% | 74.4% | 15.9% | 46.8% |
| NT | 60 | 78.3% | 68.3% | 16.7% | 50.0% |
| QLD | 2,339 | 79.1% | 78.5% | 16.4% | 45.6% |
| SA | 645 | 82.6% | 78.0% | 16.7% | 46.0% |
| TAS | 210 | 83.8% | 76.7% | 21.0% | 40.5% |
| VIC | 2,887 | 78.6% | 75.1% | 16.2% | 50.5% |
| WA | 1,089 | 79.1% | 74.7% | 16.4% | 45.4% |
| unknown | 64 | 57.8% | 40.6% | 12.5% | 60.9% |
| all | 10,846 | 78.7% | 75.3% | 16.3% | 47.6% |
What the assistants named
| Assistant | Questions | Answered | Businesses named | Distinct names | Distinct names matched to the register | Registered celebrants matched |
|---|---|---|---|---|---|---|
| Claude | 50 | 50 | 262 | 220 | 119 | 96 |
| ChatGPT | 50 | 50 | 210 | 145 | 101 | 94 |
| ChatGPT or Claude | 100 | 100 | 472 | not counted | not counted | 153 |
A name counts as matched when, as the assistant gave it, it carries the surname and a first name of a celebrant registered in the state of the city asked about. Two spellings of one celebrant are two names but one celebrant, so the two right-hand columns differ. We did not check whether the unmatched names are brands, celebrants registered under another name, or invented. Eight celebrants are in both of our groups, about what chance would give (5.7 expected).
How we measured it
The register. The Attorney-General's Department publishes the register of Commonwealth-registered marriage celebrants at marriage.ag.gov.au. We read all of it on 8 October 2026: 10,849 entries, 10,846 of them registered and active, 10,782 of those with a state shown. We kept the name, state, year registered, ceremony type, whether they published an email and a mobile, and the domain of their email when it is not a free mailbox (a website signal). No addresses, email addresses or phone numbers were kept.
The random sample. 400 active celebrants with a state, drawn at random within each state in proportion to its size (NSW 124, VIC 107, QLD 87, WA 40, SA 24, ACT 8, TAS 8, NT 2), with a fixed seed so the draw can be repeated.
The AI-named group. On 8 October 2026 we asked ChatGPT (through OpenAI's Codex command-line tool) and Claude (through Anthropic's Claude Code, Sonnet), each with web search on, two questions a couple would ask, for 25 Australian cities and regions: "recommend a marriage celebrant for our wedding" and "who are the best wedding celebrants in {city}", each asking for the businesses' website and phone where the assistant could find them. That is 100 answers, all of which came back. Each business named was matched to the register as above: 153 distinct registered celebrants, measured exactly like the random sample.
Google Maps. For each celebrant we searched Google Maps for "first name surname marriage celebrant state" (Australian English, from one Australian connection that Google placed in Sydney) and opened up to four of the results. A listing counted as theirs only when it was provably theirs:
- Under their own name: the listing's name carries their surname and a first name (or their surname and the word celebrant), and its name or Google category shows celebrant work. Where the register has someone of the same name in another state, the listing's address must be in theirs.
- Under a brand name ("Married by ...", "Ceremonies by ..."): only with evidence. Its website is on the domain of the email they gave the register, or the website's address or pages name them in full, or the business name is registered to them on the Australian Business Register (ABN Lookup). A listing whose website is on the domain of their register email counts whatever its name, even one with no person's name in it.
Anything else was not counted. So "has a listing" here means a couple searching the celebrant's name finds a Google listing that is demonstrably theirs. It undercounts celebrants who trade under a brand that nothing public ties to their name.
Reviews. The count shown on the listing. A finished listing page that shows no rating and no reviews section has none.
Websites. The website on the listing or, failing that, a page at the domain of their register email that shows celebrant work or names them. Each reachable site's home page was checked for eleven basics (a title, a search description, structured data describing the business, a sitemap, robots.txt, https, a tappable phone number, a social preview, a canonical web address, a phone viewport and image descriptions).
Read dates. The register and the AI answers on 8 October 2026; Google Maps and websites on 13 and 14 October 2026. Every celebrant not listed was read on 14 October by the final version of the reader. The listings were read by it or by an earlier version on 13 October (35 random, 64 named) or 14 October (27 and 33), and every one of those passes the final version's rules. Websites reached through an email domain were read again on 14 October under the final rule.
How we checked the reader
A reader that matches people to Google listings can be wrong in both directions, so nothing here was published until it had been checked by eye.
- The first reader counted too much. Our first version accepted a listing that carried the celebrant's first name and celebrant work ("Married by" and a first name). Checked by eye, seven of ten such matches were someone else, and nine brand listings were credited to two different celebrants each. Fixing it removed almost half of the random sample's listings, and with them the websites those listings had handed to the wrong people.
- The fixed reader then missed some. A second check found celebrants whose listing carries no person's name at all but whose website is on the same domain as the email they gave the register. Those now count.
- What was checked by eye. Of 40 listings the second version accepted (20 drawn at random from each group), all 40 were right. Of 18 searches where it found nothing, checked by screenshot: two were the celebrants above, now counted; eleven show nothing in their name; five show a brand that might be theirs but that nothing public ties to them, the undercount the ceiling measures. The final version then added 29 listings that were not checked by eye: 24 whose website is on the celebrant's own register email domain and 5 that carry their full name.
- Websites. A review found pages counted as celebrants' websites that were not theirs: a mail provider, an internet provider, a law firm, a mowing business, parked domains. A page reached through an email domain now counts only if it shows celebrant work or names them, and 43 more mailbox domains are treated as mailboxes.
- Zero reviews. Google prints no "No reviews" on a listing page. All 20 listings counted as having no reviews were polled for 20 seconds each and checked by eye: finished pages with no rating, no star histogram and no reviews tab.
What this does not say
- It is a floor, not a census of listings. A listing counts only when it is provably the celebrant's. A celebrant trading under a brand that nothing public ties to their name is counted as not listed; the ceiling row shows how far the brand leads we found could move the figure.
- The comparison between the groups is tilted, so read it as three to four times, not a precise ratio (3.9 times on the floor figures, 3.3 on the ceilings). Three tilts: only celebrants whose name, as the assistant gave it, carried a first name and surname could enter the AI-named group, so brand-only traders, exactly where the random sample undercounts, cannot be in it; the questions asked for each business's website and phone, which favours celebrants with a website; and the answers came from the assistants' command-line tools with web search, not the consumer apps a couple would use.
- The AI-named group is not a random sample. It is the registered celebrants ChatGPT and Claude named in one pass of 100 questions. Answers change from one asking to the next. The comparison shows what the named celebrants have in common, not what made the assistants name them.
- Findable by name, from one place. We searched by the name on the register, as a couple who had heard of the celebrant would. A celebrant registered under one name and trading under another may be missed. All searches ran from one Australian connection, which Google placed in Sydney.
- Websites are the ones a couple could reach from the listing or from the celebrant's own email domain. We did not run a web search for every name, so a celebrant with a website that neither their listing nor their email points to is counted as having none.
- Small states are counts only. The ACT, Tasmania and the Northern Territory have fewer than 20 celebrants in the random sample, so their shares are not reported.
- Google changes. Listings, reviews and search results move daily; these are readings from 13 and 14 October 2026.
The data
Every table on this page is published as CSV beside it, in one file: data.csv, one row per figure. Counts and shares only, each share with its n and a 95% Wilson interval. No row describes a person, and no name, email address, phone number or street address is published. Names are kept only in the study's private working files.
Its columns are table, group, state, measure, count, n, share, ci95_low, ci95_high and value, each left empty where it does not apply; shares are fractions (0.19 is 19%). The table column says which of the five tables a row comes from:
online-presence: listings, reviews, websites and the eleven website checks, for both groups.figures: median reviews and ratings, and the rows that could not be read (none).by-state: the random sample by state.register-by-state: the whole register by state (published email, published mobile, email not on a free mailbox, registered since 2020).ai-naming: what each assistant named, and the two together.
Reuse it as you like under CC BY 4.0: credit "Appear by creativebyjake" and link this page.
Credit and licence
Source: register of marriage celebrants, Attorney-General's Department, © Commonwealth of Australia 2026, CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). We sampled, matched and aggregated it; the department has not reviewed or endorsed this study. Business names were checked on the Australian Business Register's ABN Lookup.
This study and its tables are published by Appear by creativebyjake under the same licence, CC BY 4.0: use them, with credit.
How to cite it
Appear by creativebyjake, "How many Australian marriage celebrants can couples find online?", October 2026, https://appeareverywhere.com/guides/celebrants-online-2026/. Data read 8, 13 and 14 October 2026.
Start with the free check
Find out whether AI and Google know your business.
Two minutes for your website basics, then the AI and Google Maps read by email.