The short answer.
ADKAR works at the level of the individual. It gives you a way to say precisely where one person has stopped: they have not heard why, or they have heard and do not want to, or they want to and do not know how. Kotter works at the level of the organization. It gives you a sequence for how a large change gets started, carried and made permanent.
Most arguments about which is better are really arguments about which problem is in the room. If your program is stalling in aggregate, you have a Kotter problem. If your program is fine on paper and specific teams are not moving, you have an ADKAR problem.
ADKAR, in practice.
Five stages, in order: awareness, desire, knowledge, ability, reinforcement. The model is published by Prosci at prosci.com, and its usefulness is diagnostic. Because the stages are sequential, naming where someone is stuck tells you what to do next, and it stops teams from answering a desire problem with more training.
What it is less good at is everything above the individual. Sponsorship, sequencing, and the politics of a change that crosses several functions sit outside the frame. ADKAR will tell you that four hundred people are stuck at desire. It will not tell you that they are stuck there because two executives are visibly unconvinced.
Kotter, in practice.
Eight steps, set out by John Kotter in Leading Change and refined since. They run from establishing urgency and building a guiding coalition, through communicating the vision and enabling action, to consolidating gains and anchoring the change in how the organization works.
The strength is that it takes leadership behavior seriously as a mechanism rather than a nicety, and it is honest that early wins have to be visible or momentum dies. The limit is that it describes a shape rather than a measurement. Two organizations can both complete all eight steps and get different outcomes, and the model does not tell you why.
Which one to reach for.
Reach for ADKAR when the change is broadly agreed and the problem is that people are not moving through it. It gives managers language for a conversation they are otherwise having in adjectives.
Reach for Kotter when the change is not yet real to the organization. When there is no coalition, no visible sponsor, and no shared reason it matters now, an individual-transition model has nothing to work with.
Run both when the change is large. They are different altitudes, not rival theories.
What neither model was built to see.
Both were designed for change that a leadership team decides on and then rolls through an organization. AI at work has not arrived that way. It arrived in people's hands first, privately, unevenly, and often before anyone decided anything.
That produces failure modes neither model names. Tools adopted individually and never shared, so nothing anyone works out survives them leaving. Output that grows while the work does not get better. Time saved that returns to the same workload rather than to anything new. None of that is resistance, so an approach built for resistance does not find it.
This is the ground our own framework covers, and the reason it is organized as people, process and platform rather than as a sequence of steps. A fuller side-by-side of the models, including Lewin, is in our guide to AI change management frameworks.
Questions people ask.
What is the difference between ADKAR and Kotter?
Scope. ADKAR describes the change one person has to make, in five stages from awareness through to reinforcement. Kotter describes the change an organization has to make, in eight steps that begin with urgency and end with the new way being anchored in the culture. ADKAR explains why a particular person is not using the new tool. Kotter explains why the program lost momentum in month four.
Which is better for AI adoption?
Neither was designed for it. Both pre-date generative AI in the workplace and neither names the failure modes that are specific to it: tools adopted privately and never shared, output that grows while quality does not, and time saved that quietly returns to the same workload. Teams commonly run one of these as the transition model and add an AI-specific read on top.
Can you use ADKAR and Kotter together?
Yes, and many organizations do. Kotter gives the program its sequence and its leadership choreography. ADKAR gives the individual transition inside it a common language, so a manager can say which of the five stages a person is stuck at. They operate at different altitudes rather than competing.
Do we need to be certified in either one?
Certification exists for ADKAR through Prosci and is genuinely useful if you want a standard vocabulary across a large change function. Kotter's model is published and widely taught without a certification requirement. Neither certification tells you whether AI is paying back in your organization.
ADKAR is a registered trademark of Prosci Inc. Kotter’s 8-Step Process remains the work and intellectual property of its creator. Average Robot is not affiliated with, certified by, or endorsed by either organization. References here are for comparison and commentary.