Australian professional services firms are advising clients about AI. Almost none will say what they do with it themselves.
Published surveys put the share of Australian law firms with an AI policy at anywhere from under a third to 52%. Every one of those figures is self-reported. We checked something different and verifiable: what 616 Australian law, accounting and financial advice firms actually publish about their own use of AI. Of the 570 firms whose websites we could reach, 0% publish an AI policy and 1.2% tell clients how AI touches their matter.
51
firms publish commentary advising others about AI
0
of those publish their own AI policy
2
of those tell clients how they use AI
Free to reuse with attribution (CC BY 4.0). Every positive finding in the dataset carries the URL it came from.
What 570 firms publish
Percentages are of firms whose website we could reach. Ranges are 95% confidence intervals.
Mentions AI anywhere
12.6% (72/570, CI 10.2–15.6%)
Any reference to AI on any page we checked.
Publishes commentary about AI
8.9% (51/570, CI 6.9–11.6%)
Articles or insights advising others about AI, at an article-style URL.
Publishes its own AI policy
0% (0/570, CI 0–0.7%)
A page whose title or heading announces the firm's own AI policy or position, written in the first person.
Discloses AI use to clients
1.2% (7/570, CI 0.6–2.5%)
A sentence linking AI to client data, matters, confidentiality or privilege.
States where client data lives
7.7% (44/570, CI 5.8–10.2%)
A sentence naming a storage location or residency, excluding legal boilerplate.
By profession
| Measure | Law firmsn = 260 | Accounting firmsn = 181 | Financial advice firmsn = 129 |
|---|---|---|---|
| Mentions AI anywhere | 13.5%(35/260) | 8.3%(15/181) | 17.1%(22/129) |
| Publishes commentary about AI | 9.6%(25/260) | 4.4%(8/181) | 14%(18/129) |
| Publishes its own AI policy | 0%(0/260) | 0%(0/181) | 0%(0/129) |
| Discloses AI use to clients | 1.5%(4/260) | 1.7%(3/181) | 0%(0/129) |
| States where client data lives | 7.7%(20/260) | 7.7%(14/181) | 7.8%(10/129) |
How we measured it
Firms were drawn from OpenStreetMap: every Australian entity tagged as a law, accounting, tax or financial advice office carrying a website, deduplicated by domain. We used OpenStreetMap rather than commercial place data specifically so the dataset could be republished, and so anyone can rebuild the frame themselves.
Each firm's public website was then read: homepage, policy-looking links, candidate paths such as /ai-policy, and URLs found in the site's own sitemap. We obeyed robots.txt, rate-limited to one request every 1.5 seconds per site, identified our crawler honestly, and never submitted a form or contacted a firm. No business spent a minute of staff time on this study.
A firm counts as publishing an AI policy only when a page title or heading announces its own position in the first person. Articles advising other businesses about AI regulation are counted separately as commentary, because a law firm writing about AI governance has told you nothing about its own use of it.
46 of 616 firms (7.5%) had a website we could not load at all: dead domains, moved sites and misconfigured hosts. Those are excluded from every percentage rather than counted as silent.
A further 4 rows were removed by hand because OpenStreetMap had mis-tagged them: Jacobs Engineering Group, ACCA, American Express, Morningstar. They are not Australian professional services firms (an engineering group, an overseas accounting body, and two global corporates). The exclusion list, with a reason for each, ships with the dataset rather than being applied silently.
Limitations, in full: OpenStreetMap is a convenience frame, not a register, so it skews toward mapped, urban, established firms and these figures should be read as “of mapped Australian firms with a website”. A policy published only as a linked PDF, or on a page with no policy-like URL and no sitemap entry, can be missed, so these are lower bounds on publication. The study measures publication, not governance: a firm may have an excellent internal policy and publish none of it. And publishing nothing is not evidence of not using AI.
Frame data © OpenStreetMap contributors (ODbL). This dataset is published under CC BY 4.0. If your firm appears and a finding is wrong, including if you publish a policy we failed to find, email info@zatersio.com and we will re-check it and publish a correction.
Why we ran this
We build AI systems for Australian professional services firms, including on-premises deployments for firms whose client material cannot leave the building. The question we get asked most is not “can this work” but “what do we tell clients”. This study is our attempt to find out what the profession is currently telling them, which turns out to be very little. That is a commercial interest, and you should read the findings knowing it. It is also why the raw data and the code are published: you do not have to take our word for any of it.
