Most AI use is not work. Google's AI & Economy ATLAS v1.0 study, which looked at 15 million de-identified interactions, found that more than 86% of conversational AI use happens entirely outside formal work. Google shared the figure on 15 September 2026, in a consumer post about using Gemini for household chores.
The post itself is lightweight: troubleshooting an appliance through the phone camera, planning meals from a photo of the fridge, tracking expenses from receipt photos, comparing big purchases. For a business reader, the number underneath is the part that matters, and it matters whether your team runs Claude or anything else.
What does it mean that 86% of AI use happens outside work?
It means most of your staff are already building AI habits at home, on personal accounts, with no guidance from you. They are learning to photograph a document and ask for a summary, or paste in a list and ask for a plan. Those habits arrive at work on Monday morning. The question for a business is not whether staff use AI, but which tools and rules those habits land in.
A habit formed on a free personal account does not come with a data policy attached. The same person who asked a chatbot to sort their household receipts on Sunday will, sooner or later, try the same trick with a supplier invoice or a client email. If the only tool within reach is their personal one, that is where the work goes.
Small time savings add up to large numbers
Google's study also estimates that about 30 minutes a week saved per household could be worth roughly US$100 billion a year to the US economy. That is a US figure built on household time, so it does not transfer directly. The logic does. Half an hour a week sounds trivial until you multiply it.
Apply the same half-hour to a workplace. At a loaded cost of $60 an hour, 30 minutes a week across 48 working weeks is about $1,440 a year per employee. For a Melbourne business with 50 staff, that is roughly $72,000 a year of time, from a habit most of them already have. The catch is that the saving is only yours to count if the work happens somewhere you can see and govern.
Where personal habits create business risk
Not every habit that crosses over is a problem. The table sorts the common ones by how much risk they carry when they move from home to the office.
| Habit learned at home | Work version | Risk on a personal account | Better home for it |
|---|---|---|---|
| Photographing receipts for expenses | Photographing supplier invoices | Financial data outside your systems | Sanctioned Claude workspace or finance tool |
| Troubleshooting with the phone camera | Photographing equipment or site faults | Low, unless images show people or client sites | Sanctioned tool with an image policy |
| Planning meals from a list | Planning rosters or project schedules | Staff personal information exposed | Sanctioned workspace with access controls |
| Comparing big purchases | Comparing vendors or quotes | Commercial terms leave the business | Sanctioned workspace, shared with the team |
| Drafting a personal email | Drafting a client email | Client personal information under the Privacy Act | Sanctioned workspace with review before sending |
The pattern in the right-hand column is deliberate. The fix for shadow AI is rarely a ban. Bans push the habit further out of sight. The fix is a sanctioned tool that is at least as convenient as the personal one, with clear rules about what can go into it.
What Claude users should take from a Gemini stat
If your business already runs Claude, the 86% figure is a reminder that licensing is only half the job. Staff will reach for whatever feels most natural, and a tool they use every weekend has a head start. Three things close that gap:
Show staff the work versions of their home habits inside Claude: invoice photos, schedule drafts, quote comparisons
Write a one-page AI use policy that says plainly what data can and cannot go into which tool
Look at real usage data after a month, so you know which habits actually moved across
Claude's usage reflection tools help with that last step, and our piece on turning everyday Claude time into measurable ROI covers how to read them. If you do not yet know which tools staff are using, start with a shadow AI audit, then follow the process in bringing unsanctioned tools into the open.
Reading the number carefully
A few limits are worth keeping in mind. The study measured interactions with Google's own conversational AI, so it describes that user base, not every Australian worker. A 15-million-interaction sample is large but says nothing about how many of those people also use AI at work on a separate tool. And a share of use is not a share of value: a single hour of well-governed AI work on a tender may be worth more than a month of meal plans. Treat the 86% as a signal about habit formation, not as a measurement of your workforce.
If you want help turning the habits your team already has into governed, measurable work in Claude, our services page explains how we approach it, or book a time with us to talk it through.
