Guides · Published
What Is the 30% Rule for AI? Meaning and Examples
The 30% rule for AI is a loose guideline: AI handles about 70% of repetitive work and people keep the rest. Here is what it means, where it fails, and related questions.
The 30% rule for AI is a loose guideline about how work should be split between AI and people. The usual version says AI should handle about 70% of repetitive, data-heavy or preparatory tasks, while humans keep the remaining 30% for oversight, judgment and creative decisions. Alta describes it the same way, and notes that the "30%" is not a fixed number. It changes with context.
The rule is a rule of thumb, not a standard, law or study. No regulator or research body set it. Different writers use the same name for different ideas, so check which version someone means before you act on it.
Where the 30% rule comes from
The phrase spread through business and education writing as AI tools got good enough to take over routine work. The core idea is simple: start by handing AI the tasks that are repetitive and need little judgment. AI Essentials puts it as automating the roughly 30% of tasks that are repetitive and low-judgment first. Other writers frame the split differently. Medium argues that in many complex roles about one third of tasks can be automated today, which is a claim about what is possible now rather than a target for how much to hand over.
Three versions you will see
| Version | What it says | Where you see it |
|---|---|---|
| AI does ~70%, humans do ~30% | AI handles repetitive, preparatory work; people keep oversight and judgment | Operations, sales and GTM teams |
| About one third automatable | A count of tasks in a role that AI can do today | Job and workflow analysis |
| Student AI use limit | Students may use AI for only a small share of an assignment | Schools and universities |
The third version is the one most people search for when they ask about essays. It is an institutional policy, and each school sets its own limit.
Why the split does not stay put
The weak point of the rule is that the line moves. LinkedIn author Neria Sebastien argues that once a team sees a speed gain from 30% AI, pressure builds to push it to 40%, then 50%. Moving past that point means handing over synthesis and problem solving, which is also how people build expertise. So a team that treats 30% as a ceiling can still lose skill over time.
The opposite mistake is treating 70% as a target. A fixed percentage ignores which tasks matter. Automating a repetitive data entry step is different from automating a client decision, even if both count as one task.
How to use the idea in practice
- List the tasks in one workflow, not the whole job.
- Mark which ones are repetitive, low-risk and easy to check.
- Automate those first and measure errors and time saved.
- Keep a human on tasks that involve judgment, customers or legal exposure.
- Review the split every few months as tools change.
Alta's guide makes a similar point: the rule is a starting place, not a ceiling.
Related: selling company records to AI developers
Some searchers who look up the 30% rule are also asking how companies can earn money from AI. Those are separate topics, but they overlap in one practical way: AI developers need examples of real work, and some companies have years of it sitting unused.
Smelted is a free matching service for that case. It matches US companies with AI data buyers who license anonymized internal records, such as support tickets, CRM history, project files and code. The company keeps ownership of its data. It licenses a copy, chooses what is included, and chooses which buyers it meets. Names, contact details and IDs are removed before anything is shared. The buyer pays Smelted's fee, and the company pays nothing.
Buyer listings describe deals from about $10K to $1M+, mostly for companies with 30 or more employees. Those figures are ranges, not offers. A real number only comes after a buyer has reviewed a sample of the data. Payment usually comes one to three months after the first call, though the timing depends on the deal.
The typical fit is a founder or C-level owner at a US company with 50 to 500 employees, founded in 2022 or earlier, with years of records in tools like Zendesk, Jira or GitHub, Salesforce or HubSpot, Bullhorn, AutoCAD or Revit, or Benchling. Core industries are staffing and recruiting, engineering and architecture, B2B software and science labs. Smaller companies can still ask. The buyers who take them tend to pay less, so the team will say so plainly. Law firms, healthcare and medical billing, accounting, call centers, IT managed-service providers, education, government and defense are not a fit. Physical books and published works are also out.
If you want a ballpark range, you can book a 15-minute call with Chris Irving at Smelted or use the 1-2 minute form. Before you share anything, check your client contracts and privacy obligations. Smelted does not give legal or compliance approval.
FAQ
What is the 30% rule of AI?
It is a guideline that AI should handle roughly 70% of repetitive, data-heavy work while people keep the other 30% for oversight, judgment and creative decisions. It is a rule of thumb, not a formal standard.
What is the 30% rule for AI in school?
Some schools set a limit on how much of an assignment a student may write with AI. The limit varies by school and course, so check the policy your instructor gives you.
Can you sell data to AI companies?
Some companies do. AI developers license business records to train systems, usually after removing names and identifying details. Whether your data can be shared depends on who owns it, your contracts and privacy rules, so review those first.
How much money do companies get for selling data?
Buyer listings describe deals from about $10K to $1M+, mostly for companies with 30 or more employees. A real figure only comes after a buyer reviews a sample of your data, so treat the range as a starting point.
How do you sell data for AI?
Start by listing which internal records you have, how many years they cover, and whether the data is yours or your clients'. Then talk to a matching service or a buyer, choose what to include, and review any license before signing.