Adaptive learning and personalised learning paths: the guide

What are we talking about?

Three terms partly overlap. Personalisation refers to any adjustment of a learning path to an individual, including adjustments based on stated preferences. Adaptive learning is the form based on continuous measurement of learning: with each answer, the system estimates what the learner has mastered and decides on the next activity. Individualisation (individualisation des parcours) is the term used in French vocational training, notably by the Qualiopi quality framework, for adapting a learning path after an entry assessment.

See the detailed definition of adaptive learning.

Why it works

Two findings from cognitive psychology underpin the approach. Being tested on information consolidates it more than rereading it (Roediger and Karpicke, 2006). Spreading learning over time improves retention, with an optimal gap that grows with how long the information needs to be retained (Cepeda et al., 2006 and 2008). An adaptive system applies both principles for each person and each concept, something no trainer can do by hand for a whole group.

What the evidence says

Bloom's "2 sigma problem" (1984), which attributed two standard deviations of progress to one-to-one tutoring, has not been found again at that magnitude. VanLehn's review (2011) measures 0.79 standard deviations for human tutoring and 0.76 for tutoring systems that intervene at each step; the meta-analysis by Kulik and Fletcher (2016) finds a median effect of 0.66 across 50 evaluations. These effects are large, but they depend on the quality of implementation and shrink in large-scale deployments.

Generative AI adds a new risk. In a trial published in PNAS in 2025, high school students with unrestricted access to GPT-4 did better on practice problems, then scored 17% lower on the exam without AI. Conversely, an AI tutor designed to guide without giving the answer more than doubled the median learning gains of Harvard students compared with an active-learning class. See what the research really says.

How AI personalises a learning path

The process has four stages: diagnosis of the starting level, building the sequence from the gap with the competency framework, adjusting difficulty and order with each answer, and spaced reviews. Systems that respond at each step of the learner's reasoning do clearly better than those that only correct the final answer. See how AI adapts a learning path.

Telling the mechanism from the façade

Many offers call simple catalogue filtering or predefined conditional paths "personalisation". Four questions help sort them out: is the path recalculated from the learner's answers? Is learning measured throughout the path and after a delay? Does the system rely on a structured framework? Do assessments require applying a skill, or only recognising an answer? See personalised learning paths and the categories of tools.

What learners expect

In 2025, 48% of French people aged 12 and over used generative AI according to France's Digital Barometer, and 72% of working people who use it at work said they needed training on it, according to Centre Inffo and CSA. Career progress remains the top motivation to learn according to LinkedIn. See what learners expect.

Designing training that lasts

Adaptation cannot make up for poor design. Crammed sessions, no reminders, content unrelated to the job: the causes of failure and the principles of effective design (spacing, retrieval practice, cognitive load, transfer) belong to instructional design. At company level, learning also depends on how work is organised: see learning organisation.

Training providers and CFAs in France: an obligation

For Qualiopi-certified providers, individualisation is not optional. France's National Quality Framework requires procedures for entry assessment of prior learning (indicator 8), adapting the service (indicator 10), assessing whether objectives are met (indicator 11) and measures against dropout (indicator 12). An updated framework, set by Decree No. 2026-728 of 1 August 2026, applies from 1 November 2026. See individualised learning paths in training providers and apprentice training centres.

Limits to keep in mind

Adaptive learning has mostly been validated in structured, assessable domains. It does not replace human assessment of behavioural skills, dedicated time, or the manager's involvement in applying skills on the job. See what AI cannot do in training and the role of an AI agent alongside the learner.