In 2013, Ian Duncan-Smith said “looking for work should be a full-time job”. This was to be policed through the ‘claimant commitment’ a document that would detail, among other things, the number and type of jobs that someone would be expected to apply for. People would then present evidence that they were spending up to 35 hours a week trying to meet those targets when they signed-on.

Proving you are spending time looking for work has been a component of the British welfare system for a long time. Before the Claimant Commitment, people had to fill out a paper form detailing the jobs they have applied for, and you can go right back to 1921 and the introduction of the “seeking work” test.

Also in 2013, an anonymous group of benefits campaigners released an app called Universal Automation. Universal Automation automatically submitted random job applications on behalf of users of the government’s Universal Jobmatch website - a service that benefit claimants are told to use to look for jobs (and which allows Jobcentre staff to see what jobs they are applying for).

Universal Automation was an exercise in digital disobedience rather than a useful service, but it was an interesting weak signal too. It asked the question: what if finding jobs became something that took seconds rather than a day?

Last week, Google announced ‘Google for Jobs’ which promises to use AI to determine which jobs people should apply for. This is possible, in part, because of the slow but steady adoption by job boards of a standard for publishing structured data about job vacancies* - the schema.org jobPosting.

Traditionally, job search services have used free text search plus a basic understanding of location to help people find jobs. With jobPosting, in addition to the textual description of a job, structured machine-readable data about things like location, job category, skills, hours and benefits can be included.

Better data, in combination with increasingly ubiquitous AI, mean it’s possible to do smarter things - like recommend jobs that fit both your skills and when you need to pick the kids up from school, or tell you which jobs you could apply for if you developed a particular skill.

Google for Jobs currently looks quite basic, but it (or something like it) will soon start to make looking for work into a ‘push’ rather than ‘pull’ activity, and release an enormous amount of wasted human potential in the process.

So what should happen to welfare policy when digital assistants replace the job of looking for a job?

One response could be to carry on regardless - forcing benefit claimants to continue to present evidence of time spent. But if digital assistants get good enough, then this will become tantamount to getting people to dig holes and fill them in again.

A more mature response would see changes both to what is expected from benefit claimants and to the data infrastructure that government provides to support job seeking.

Benefit claimants should be freed up to spend more time on learning and other activities. They should be able to submit digital proofs that apps were looking for jobs on their behalf or that they completed a course online.

The government should support services like Google for Jobs by investing in data infrastructure that increases the quality and quantity of open data about the labour market and skills.

It should prioritise creating open, real-time datasets about career progression, job titles, available skills and demand from people’s interactions with job centres and HMRC **.

Data about child care availability and training courses should be made available, and research into what skills are needed for particular jobs needs to be invested in.

Finally, government needs to create a regulatory framework for services like Google for Jobs to ensure that they support people from all backgrounds and that any potential for bias baked into the service can be spotted early and addressed.

As good design and AI start to become a feature of the working-age welfare state, one of the most effective interventions government can make will be to improve the quality of data those systems use. The days of having to prove “actively seeking work” are nearly over.

• I’m reading between the lines here, but it fits with how Google has approached other sectors, like events ** aggregate, not individual