Digital learning in 2026: three approaches to explore

Microlearning: learning in a short format

Short se­quences fo­cused on one concept or task can make con­sulta­tion and re­vi­sion easier in some con­texts. There is no uni­ver­sal dur­a­tion: choose it ac­cord­ing to the con­tent, medium and learners’ needs, then check ac­cess­ib­il­ity and un­der­stand­ing.

Gamification: engagement within digital learning

Points, badges and chal­lenges are pos­sible mech­an­ics whose ef­fects vary with in­di­vidu­als and design. Rapid feed­back can inform learners about their an­swers; do not assume in­creased mo­tiv­a­tion or com­ple­tion without eval­u­at­ing the pro­gramme.

Artificial intelligence: assistance and personalisation need oversight

Adaptive learning

Some sys­tems can use re­sponses or pro­gress data to sug­gest dif­fer­ent con­tent. Check the tool’s actual cap­ab­il­it­ies, per­son­al­isa­tion rules and data pro­tec­tion ar­range­ments; a path­way pro­posed this way is not ne­ces­sar­ily op­tim­al for every person.

Automating time-consuming tasks

AI tools can help draft quizzes or handle cer­tain tasks, but avail­ab­il­ity varies and con­tent re­quires human check­ing. Mark­ing and is­su­ing cer­ti­fic­ates depend on the tools and pro­gramme rules. See the prompt toolkit for trainers.

Conversational assistants during training

Di­dacticBot demon­strates a spoken ex­change during a video: the course pauses, the AI an­swers using val­id­ated ma­ter­i­als, then play­back re­sumes on re­quest. This demon­stra­tion does not val­id­ate every po­ten­tial per­son­al­isa­tion or ana­lyt­ics cap­ab­il­ity. See the of­fer­ing for businesses.

Read trends as a portfolio of hypotheses

The 2026 label is used here to or­gan­ise cur­rent choices, not to date the emer­gence of three in­nov­a­tions or pre­dict their ad­op­tion. For each ap­proach, write down the prob­lem to ad­dress before watch­ing a demon­stra­tion: dif­fi­culty find­ing a pro­ced­ure, in­suf­fi­cient prac­tice or pre­par­a­tion work­load. Then for­mu­late a falsifi­able hy­po­thes­is. “People will find the right pas­sage without help” is test­able; “our train­ing will be more in­nov­at­ive” de­scribes neither a task nor an ob­serv­able be­ne­fit.

Create a sheet for each ap­proach cov­er­ing the audi­ence, start­ing situ­ation, pro­pos­al, re­sources needed, risk and sign of fail­ure. Avoid test­ing short se­quences, com­pet­i­tion and an as­sist­ant sim­ul­tan­eously: if the result changes, you will not know which ele­ment to ex­am­ine. The de­liv­er­able is a lim­ited port­fo­lio of ex­per­i­ments, ranked by need and veri­fic­a­tion cost. An ap­proach can remain under ob­ser­va­tion if the need has not been es­tab­lished; de­cid­ing not to launch is le­git­im­ate, not a tech­no­lo­gic­al gap that must be closed.

Choose a distinct test for each of the three approaches

For short se­quences, focus on self-con­tained use and con­tinu­ity: does an isol­ated re­source in­clude the pre­requis­ites and ex­plain when to apply the rule? Pre­pare an in­form­a­tion-find­ing task, then an ap­plic­a­tion task. For gami­fic­a­tion, ex­am­ine reward align­ment: are you re­ward­ing sound reas­on­ing, speed or simply at­tend­ance? Also test the ex­per­i­ence of people who do not want to be pub­licly com­pared. Points must not become a sub­sti­tute for suc­cess cri­ter­ia.

For AI, build a set of covered, am­bigu­ous and out-of-scope ques­tions. Check ac­cur­acy, the ref­er­ences used and the ex­pec­ted re­sponse when in­form­a­tion is miss­ing. Fic­tion­al ex­ample: a team com­pares a short in­form­a­tion sheet, a chal­lenge-based activ­ity and con­ver­sa­tion­al as­sist­ance across three dif­fer­ent needs, without rank­ing the tech­no­lo­gies. Each re­ceives its own pro­tocol. The de­liv­er­able is not a single in­nov­a­tion score but a con­clu­sion for each use: suit­able, re­quires modi­fic­a­tion or not demon­strated under the ob­served con­di­tions.

Separate evidence, governance and supplier promises

The UNESCO guidance of 7 September 2023 ad­dresses eth­ic­al and ped­ago­gic­al val­id­a­tion, data pro­tec­tion and a human-centred ap­proach to gen­er­at­ive AI in edu­ca­tion and re­search. It does not pre­dict market shares in 2026 or val­id­ate a par­tic­u­lar as­sist­ant. Use this source to build gov­ernance ques­tions without turn­ing it into evid­ence of per­form­ance or a com­mer­cial en­dorse­ment.

During a demon­stra­tion, divide in­form­a­tion into three columns: ob­served, stated and un­veri­fied. A suc­cess­ful answer to a pre­pared ex­ample be­longs in the first column, but tells you noth­ing about every pos­sible ques­tion. Ask about data pro­cessing con­di­tions, the limits of the doc­u­ment­ary sources, cor­rec­tion ar­range­ments and con­di­tions for leav­ing the ser­vice. For any claimed auto­mat­ic ad­apt­a­tion, re­quest de­tails of the data used and the de­cision rule. The de­liv­er­able is a list of re­quire­ments ac­com­pan­ied by ex­pec­ted evid­ence, rather than a cata­logue of as­sumed fea­tures.

Organise reversible decision gates

On 24 April 2024, the EEF described a four-phase process: ex­plore, pre­pare, de­liv­er and sus­tain. This guid­ance con­cerns edu­ca­tion­al set­tings; its use here in vo­ca­tion­al train­ing is a meth­od­o­lo­gic­al ad­apt­a­tion, not an es­tab­lished find­ing about busi­nesses. Its prac­tic­al value lies in dis­tin­guish­ing the se­lec­tion of an ap­proach from the con­di­tions needed for its day-to-day op­er­a­tion.

Adapt these phases into four gates: need con­firmed, trial ready, ac­cept­able use and jus­ti­fied con­tinu­ation. At each gate, name a de­cision-maker and the ex­pec­ted doc­u­ment. Before the trial, plan a way to return to the pre­vi­ous pro­gramme; before ex­pan­sion, check sup­port ca­pa­city and ne­ces­sary up­dates. Doc­u­ment in­cid­ents, access dif­fi­culties and human su­per­vi­sion time, not just en­thu­si­ast­ic feed­back. A review date pre­vents an ex­per­i­ment from be­com­ing per­man­ent through in­er­tia. The review must be able to re­com­mend stop­ping an ap­proach without call­ing the ori­gin­al learn­ing need into ques­tion. Archive de­cisions and re­ser­va­tions so that the next review starts from needs, not merely market an­nounce­ments.

Key takeaways

Mi­crolearn­ing, game mech­an­ics and AI are op­tions among others for design­ing train­ing. Eval­u­ate their fit with ob­ject­ives, learners, ac­cess­ib­il­ity and avail­able re­sources, then meas­ure res­ults rather than as­sum­ing an impact.