Engagement rate dropping for LinkedIn by 10% increased number of sponsored content

I’d first check whether shorter sessions reflect frustration or greater efficiency. Someone finding a relevant job quickly could have a shorter session and a better experience. The investigation needs to establish which behaviour is driving the decline.

1. Clarify the product and scope

LinkedIn helps professionals build networks, discover opportunities, share knowledge, and find jobs.

For this case, I’d focus on the core LinkedIn experience, including the feed, jobs, messaging, groups, and profiles. I’d exclude LinkedIn Learning and include both free and premium members.

The question initially refers to an engagement-rate decline, but the agreed metric is average session duration. I’d make that distinction explicit before proceeding.

2. Define the metric and verify the decline

I’d define:

Average session duration = Total measured session time ÷ Number of sessions

I’d clarify how sessions start and end, how inactivity is handled, and whether background time is included. If the existing metric counts background time, I’d preserve it for comparison while examining active foreground time separately.

I’d also check median duration and the distribution across sessions. A small group of very long sessions can materially affect the average.

The case describes a cumulative 10% decline over six months. For illustration, an average falling from 10 minutes to 9 minutes would represent that change.

Before investigating causes, I’d verify:

• Whether session definitions, tracking, or reporting changed.

• Whether missing events or early timeouts are shortening recorded sessions.

• Whether the comparison accounts for seasonal patterns.

• Whether a change in the mix of users or session types explains the average.

Users who never start a session should not contribute zero-duration sessions to this metric. Fewer visitors would affect traffic, but would not automatically reduce average session duration.

3. Segment the change and check external explanations

I’d compare duration by platform, country, membership type, new versus established members, and primary activity.

The case says the decline is global, spans platforms, and has no clear user-category pattern. I’d still examine differences in magnitude rather than assume every group is equally affected.

I’d then investigate external and behavioural changes:

• Are competing job or professional-networking services seeing similar declines?

• Have competitors improved job discovery, networking, or content?

• Have regulations, negative publicity, or operational disruptions affected usage?

• Have recruiter, employer, or affiliate relationships changed?

• Are employers directing more applications to their own websites?

• Have hiring conditions, work routines, or spending priorities changed?

The case provides no strong evidence for these explanations. Partnerships and hiring conditions are described as stable, so I’d prioritise changes within LinkedIn.

4. Locate the decline within the product

LinkedIn has several user journeys, so I’d examine time spent and task completion by product area.

I’d review releases, experiments, recommendation changes, interface updates, and monetisation decisions affecting the feed and jobs pages.

I’d also quantify how much each area contributes to the overall decline, accounting for users moving between them.

5. Test relevance, friction, and content-mix hypotheses

I’d start with three possibilities:

1. A new content format does not match what users want.

2. Discovery or application changes make important tasks harder.

3. Recommendations have become less relevant, potentially because sponsored content receives more prominence.

I wouldn’t assume creator-led content or short videos are inherently unhelpful. The question is whether the content serves the user’s professional interests.

The interviewer identifies recommendation changes and increased sponsored content on both affected pages. I’d test two related mechanisms separately:

• Relevance: Are the recommended posts and jobs less useful?

• Frequency: Are sponsored placements crowding out useful alternatives?

Evidence would include more hides or “not relevant” feedback, lower satisfaction, fewer useful saves, weaker job matches, and lower return usage after exposure.

For jobs, I’d examine application starts and completion. A shorter visit followed by a suitable application could indicate improvement; a shorter visit with repeated searches and no useful result could indicate frustration.

I’d compare exposed users with appropriate holdouts or rollout cohorts. The gradual six-month decline should also align with expanding exposure or changes in ranking and sponsored-content volume.

My leading diagnosis would be a deterioration in feed and job relevance associated with recommendation and sponsored-placement changes.

6. Address the immediate experience

I’d first reduce or roll back the ranking changes implicated by the evidence, then test improvements in a controlled rollout.

My short-term priorities would be:

• Make feedback such as “not relevant,” “hide,” or “see less” easy to find, improving existing controls where necessary.

• Apply clear feedback promptly to the relevant topic, advertiser, or job category.

• Introduce or tighten sponsored-content frequency caps.

• Require a relevance threshold for sponsored posts and jobs.

I’d avoid treating one dismissal as proof that someone dislikes an entire field. Feedback should be specific, reversible, and combined with other signals.

The source proposes Fibonacci spacing between ads. I’d interpret that as progressively wider gaps, such as 1, 2, 3, 5, and 8 organic items between placements. It could be an experiment, but the sequence itself has no demonstrated advantage here. I’d compare it with simpler spacing and frequency caps.

7. Improve personalisation over the longer term

I’d make professional interests easier to express and update. Users could optionally select a few areas, such as technology, startups, consumer businesses, or venture capital, then refine them as their interests change.

These preferences would inform both organic and sponsored recommendations. Sponsored content should earn its place through relevance as well as commercial value.

For jobs, I’d improve matching around role, skills, experience, location, working arrangements, and compensation expectations.

The source suggests using current and expected compensation. I’d prioritise an optional desired range and available job salary ranges. Current pay should not be required or become a ceiling on the opportunities shown.

I’d preserve room for discovery too. Personalisation should help users find useful adjacent opportunities, rather than repeatedly show the same narrow set of topics and roles.

8. Measure value alongside session duration

I’d continue tracking average session duration to evaluate the original decline, but I’d pair it with outcomes that explain whether the experience improved.

For the feed, I’d track:

• Reported professional relevance and satisfaction.

• Useful saves and substantive interactions.

• Hides and negative feedback.

• Return usage over subsequent weeks.

For jobs, I’d track:

• Relevant-job saves and application completion.

• User feedback on match quality.

• Downstream responses where reliably measurable.

• Time taken to find a suitable opportunity.

Interest-selection adoption and scroll depth would be diagnostic measures. More scrolling can mean interest, but it can also mean difficulty finding something useful.

Guardrails would include sponsored-content revenue, advertiser outcomes, repeated ad exposure, and exits following sponsored placements. I’d use experiments to determine whether those placements cause exits rather than infer causation from sequence alone.

My recommendation is to restore relevance on the feed and jobs pages, give users clearer control over recommendations, and test sponsored-content volume carefully. Longer sessions would be welcome when they reflect useful activity, but helping professionals achieve their goals is the stronger measure of success.