By Brad Ferris · 15 August 2026
The Australian Financial Review reported on 13 August that consultants are facing a backlash from clients over plans to use AI. The number underneath the story is the part worth keeping: about 70 per cent of consultancy clients said they would not trust a report if it was prepared using AI, according to a survey by Source Global Research.
Most people will read that as a story about the big four. It is a story about anyone who sells a work product.
Be precise about the question those buyers were answering. They were not asked whether AI produces worse analysis. They were asked whether they would trust a deliverable made with it. Those are different questions, and the gap between them is the whole commercial issue.
A buyer paying for a report is buying two things: the content, and someone's accountability for the content. AI does not weaken the first. It muddies the second, because the buyer cannot see where the machine stopped and the professional started. Absent any statement of method, the safest assumption a client can make is the least flattering one.
That logic does not stop at consulting. It applies to the engineering firm issuing a specification, the accountant issuing advice, the agency delivering a strategy, the builder pricing a variation, the broker writing a submission. If your business hands a client a document and charges for it, this survey is about you.
Two of them landed in the past fortnight, in very different settings.
The AFR reported in late July that PwC published leadership reports riddled with AI hallucinations, including a footnote that pointed to a media report which did not mention the survey it was cited for. In August, the AFR reported a workplace case thrown out after a claimant was caught using AI coaching in the witness box.
Different worlds, same shape. AI sat inside a process, nobody declared it, and nobody checked the output before it went out the door. The damage in both cases came from the undeclared part, not the technology.
For two years the quiet assumption in a lot of businesses has been that AI use is invisible, so disclosure is optional. That assumption is being engineered out of the tooling.
On 11 August, Anthropic began embedding C2PA provenance metadata in text and files generated across Claude, applying to models released after 2 August. The direct driver was European: the European Commission started enforcing AI Act transparency obligations on 2 August, covering disclosure that a user is interacting with AI, marking of synthetic content, and identification of deepfakes.
Australian businesses are not bound by the EU regime. They will still inherit its plumbing, because the vendors build one product and ship it everywhere. Provenance is becoming a default property of AI output rather than a feature somebody turns on. Expect it to arrive first as a procurement question in a tender, and later as an audit question.
The strategic point is simple. Whatever your disclosure position is, choose it deliberately, because the option of having no position and relying on nobody noticing has a short remaining life.
The practical response here is not a policy project. It is a method statement your clients can read, and it has three lines.
Where AI is used. Name the steps. Research and first drafts, transcription and summarisation, code generation, data extraction, image work. Be specific enough that a client can picture it.
What a human verifies before it leaves. This is the line that carries the weight. Every number checked to source, every citation opened, every recommendation signed off by a named person. If you cannot describe the check, you do not have one.
Who is accountable if it is wrong. A person, not a department. The same person who would have been accountable before any of this started.
Put it in the proposal where the client will read it, not buried in terms and conditions where it reads as a liability shield. An afternoon of work covers it.
The buyers in that survey were not refusing to work with suppliers who use AI. They were refusing to trust work whose method they could not see. Those are two very different market signals, and the second one is an opportunity for anyone willing to move first. Stating your method converts a hidden risk into a visible standard, and it lets you compete on the quality of the check rather than on a pretence that no machine was involved.
There is a second benefit, and in my experience it is the larger one. Most businesses cannot write those three lines today. Not because they are hiding something, but because nobody has mapped where AI has quietly entered the workflow, and staff have adopted tools faster than anyone has written down how the work is produced. Writing the method statement forces that map into existence. You end up with an honest description of how your business delivers, which is useful well beyond the disclosure question.
The client conversation is coming either way. The only real choice is whether you are the one who raises it.
Where does your business stand? The free AI Scorecard takes three minutes and shows you. If you want a straight steer from a person, book an AI Opportunity Call.