LinkedIn job recommendations are supposed to make job hunting easier. The platform already knows a surprising amount about what a job seeker may want: job titles, preferred locations, remote or hybrid preferences, profile history, searches, alerts, and previous job-search activity.
So why do some users still open LinkedIn Jobs and think:
“Why am I seeing this?”
That question became the starting point for this research.
The issue is more interesting than simply saying that an algorithm sometimes recommends the wrong job. Recommendation systems will never be perfect.
The real question is what happens when a user has clearly communicated what they want, but the product appears to interpret that intent differently.
Quick Answer: Why Can LinkedIn Job Recommendations Feel Irrelevant?
LinkedIn says it personalizes job recommendations using both explicit preferences—such as job titles, locations, workplace type and employment type—and other signals including profile information and job-search activity.
This creates a difficult personalization problem.
What a person has done in the past may not always be the same as what they want to do next.
Recent public discussions also show users reporting mismatches involving remote-work preferences, location filters, highly specific searches and the newer AI-powered job-search experience.
The problem may therefore be less about whether LinkedIn has enough data and more about which signals should matter most at a particular moment.
LinkedIn Already Has a Lot of Job-Search Intent
LinkedIn allows members to tell the platform directly what they are looking for.
According to LinkedIn’s documentation, job preferences can include job titles, preferred locations, remote/hybrid/on-site preferences and employment types. LinkedIn says these preferences are used to personalize the jobs members see.
But explicit preferences are only part of the picture.
LinkedIn also says its job recommendations can use profile information and signals from job-search activity, including actions such as viewing or saving jobs.
This makes sense.
If someone repeatedly searches for data-product roles, saves several Data Product Manager jobs and updates their profile accordingly, those behaviours can help a recommendation system understand them.
The difficulty begins when these signals disagree.
Imagine someone who has spent ten years in software development but now wants to move into product management.
Their history says:
Software Engineer.
Their current search says:
Product Manager.
Which one should LinkedIn trust more?
There is no simple answer.
But from the job seeker’s perspective, the current goal is usually very clear:
“Show me where I want to go, not only where I have already been.”
AI Search Changes the Relationship Between the User and the Search Box
LinkedIn’s newer AI-powered job search makes this question even more important.
Instead of requiring exact keywords, LinkedIn says users can describe a desired role in natural language. The system then interprets the search intent and matches it against job descriptions. Users can include details such as location, experience level, skills, specialty and employment type.
That is potentially very useful.
A job seeker could search for:
“A remote product role where I can use my healthcare and AI experience.”
That is much more natural than trying to guess the perfect Boolean query.
But interpretation introduces a new ambiguity.
Suppose the user says:
“Remote Product Manager.”
Does remote mean:
Remote only
or
Remote preferred, but show me related hybrid roles too?
Humans naturally distinguish between a preference and a hard constraint.
An AI system may try to be helpful by broadening the result set.
The user may experience that same behaviour as the system ignoring an instruction.
LinkedIn’s own guidance hints at this challenge. It advises users to give AI job search enough context and notes that vague searches may return less relevant results.
The Strongest Public Friction Appears Around Hard Constraints
Recent public discussions provide a useful signal here.
In one August 2026 discussion, a user said a job alert that had been set to remote began showing on-site and hybrid roles in the AI-powered experience. The user described having to manually work through a much larger number of unwanted listings.
Another user reported carefully selecting company, posting-time and location filters but feeling that the AI search ignored the selected location.
A separate discussion complained that searches which had previously produced a small set of useful matches were returning hundreds of less-relevant results, turning a quick search into substantially more manual filtering.
These posts do not prove that most LinkedIn users have the same experience.
They do show something important about expectations.
When someone specifies:
remote,
a particular location,
a particular role,
or a narrow set of criteria,
they are often not asking the product to make a creative interpretation.
They are asking it to respect a boundary.
That suggests one useful distinction for recommendation design:
Some signals should help the system explore. Others may need to behave like constraints.
Relevance Is Also About the Cost of Filtering
A recommendation does not need to be completely absurd to create friction.
It only needs to be irrelevant enough that the user has to spend time rejecting it.
Imagine searching for a Senior Product Manager role and receiving ten results.
One is genuinely relevant.
Two are adjacent.
Three are in the wrong location.
Two are substantially different roles.
Another is hybrid when you need remote work.
And one is a listing you already evaluated.
The system has technically found several related opportunities.
The user, however, has still had to perform nine unnecessary evaluations to reach the useful one.
This is why recommendation quality is also an attention problem.
LinkedIn’s own product language recognizes the value of reducing this effort. Its AI-powered search is positioned as a way to help people reach relevant jobs without needing exact keywords, and LinkedIn continues to support filtering as part of that experience.
For the user, the ideal outcome is simple:
Spend more time evaluating jobs worth considering and less time filtering obvious mismatches.
Career Changers Create a Special Recommendation Problem
Career transitions deserve separate attention because they challenge one of the basic assumptions behind personalization.
Recommendation systems learn from history.
Career changes deliberately break with history.
Consider someone moving from finance into technology, engineering into product management, or operations into data analytics.
Their profile, work history and previous network may continue producing strong signals about the old career.
But their current search behaviour represents something new.
If historical signals remain dominant for too long, the system risks becoming very good at recommending a future the user no longer wants.
LinkedIn confirms that both profile information and user preferences can contribute to job recommendations.
That does not tell us how LinkedIn internally weights those signals.
It does give us a useful product question:
How quickly should current intent be allowed to override historical relevance?
For someone continuing along the same career path, historical signals may be extremely useful.
For a career changer, they can become noise.
Search Control Matters More When the User Knows Exactly What They Want
Not every job seeker wants AI to interpret their needs.
Some users already have highly developed search behaviour.
They know the job title.
They know the location.
They understand Boolean searches.
They know which filters matter.
For those users, additional intelligence can be valuable only if it preserves their ability to express precise constraints.
That tension appears clearly in recent public reactions to LinkedIn’s AI-search changes. Several users described missing the predictability and granularity of the earlier search experience, particularly around filters and remote searches.
This does not mean AI search is inherently worse.
It suggests there may be two different jobs to serve:
“Help me discover roles I may not know how to search for.”
and
“I know exactly what I want. Help me find only that.”
A good search experience may need to recognise which mode the user is in.
Feedback Creates Another Expectation: “Learn From Me”
Recommendations do not end when results appear.
Users interact with them.
They view jobs.
Save some.
Apply to others.
Ignore many.
Change preferences.
Run another search.
LinkedIn explicitly says that job-search activity can contribute to personalization.
That means users are not simply consuming recommendations.
They are continuously producing new signals.
And that creates an expectation:
“If I repeatedly show you what I do and do not want, the experience should become easier to steer.”
This is where relevance starts to overlap with control.
A recommendation system can be reasonably intelligent and still feel frustrating if the user cannot understand how to correct it.
The question is therefore not only:
“Can LinkedIn predict a relevant job?”
It is also:
“Can the job seeker reliably influence the prediction?”
The Research Points to Three Qualities That Matter
The evidence suggests that a better job-recommendation experience should perform well across three dimensions.
Relevant
The opportunities should make sense given the job seeker’s current intent, not simply their historical profile.
Efficient
The person should be able to reach useful opportunities without spending excessive effort rejecting obvious mismatches.
Controlled
Important preferences and feedback should feel meaningful enough that users understand how to steer what appears next.
These three ideas are closely connected.
If recommendations are relevant but difficult to control, users may struggle when their circumstances change.
If they are controllable but mostly irrelevant, users still waste time.
And if they are accurate but require too much manual filtering, the product is not reducing enough of the job seeker’s workload.
The Bigger Product Question
It would be easy to reduce this research to:
“LinkedIn needs a better recommendation algorithm.”
That conclusion would be premature.
The more interesting question is about the relationship between intelligence and intent.
A modern recommendation system has access to enormous amounts of context.
It can infer.
Predict.
Broaden.
Personalize.
But the user also knows something the algorithm cannot infer perfectly:
What I want now.
The strongest experience may therefore not be the one where AI makes the most decisions.
It may be the one that knows when an inferred recommendation is useful—and when an explicit user instruction should simply win.
That distinction matters far beyond LinkedIn.
It applies to almost every AI-powered product that tries to personalize choices on behalf of a user.
Frequently Asked Questions
How does LinkedIn decide which jobs to recommend?
LinkedIn says recommendations can use explicit job preferences, including job titles, locations, workplace type and employment type, along with profile information. LinkedIn also uses job-search-related activity such as viewing or saving jobs to help users find relevant opportunities.
Does LinkedIn use AI for job search?
Yes. LinkedIn’s AI-powered job search lets users describe the role they want in natural language. LinkedIn says the system interprets the meaning of that request and matches it against job descriptions rather than requiring exact keyword searches.
Why might LinkedIn show hybrid or on-site roles in a remote search?
There is no public information that allows us to identify a single technical cause. Recent users have reported this behaviour, while LinkedIn’s AI search works by interpreting search intent rather than simply relying on exact keywords. Possible causes would require internal LinkedIn data to verify.
Why are career changers difficult for recommendation systems?
A career changer’s historical profile may strongly represent their previous occupation while their current search expresses a different future direction. Because LinkedIn says both profile information and explicit preferences can contribute to recommendations, a career transition creates a natural tension between historical relevance and current intent.
What should a good job recommendation system optimize for?
For the user, three outcomes appear particularly useful: relevance, efficiency and control. Jobs should align with current intent, users should not need excessive filtering to find credible opportunities, and important preferences should give users a meaningful way to steer the experience.
Sources and Further Reading
LinkedIn Help — How we help job seekers and hirers connect. Explains the use of profile, preference and job-search activity signals in AI-powered job discovery.
LinkedIn Help — Discover new opportunities with AI-powered job search. Describes natural-language job search, supported search context and recommendations for improving relevance.
Recent public user discussions were also reviewed to identify qualitative patterns around remote searches, relevance and search control. These discussions are evidence of individual user experiences and are not treated as representative statistics for LinkedIn’s entire user population.