Case 048
Partner since 2015
Connecting You to Opportunity (Randomized, Controlled, Peer-Reviewed)
- $140B
- Extraction Index
- 8
- Key initiatives
- 2015
- Rubicon crossed
In Summary
LinkedIn is Microsoft's professional network, whose documented extraction runs through data use rather than pricing. Between 2015 and 2019 it ran randomized experiments on more than 20 million members by varying the 'People You May Know' algorithm without informing them, and published the results in the journal Science; those members went on to file over 70 million job applications leading to roughly 600,000 jobs. In October 2024 the Irish Data Protection Commission fined it EUR 310 million for having no valid lawful basis for behavioural analysis and targeted advertising, and in August 2024 it enabled AI training on member data by default. EnshitifAi assigns LinkedIn an Extraction Index of $140B, an editorial figure anchored to these documented actions.
The Story
LinkedIn is the partner we cite when a client tells us their users would never accept it. Between 2015 and 2019 the client ran a randomized controlled trial across more than twenty million members, varying the mix of strong and weak ties each person was shown, and measured the result in employment. Seventy million applications. Six hundred thousand jobs. The members were not informed, the findings were published in Science with co-authors from MIT, Stanford and Harvard Business School, and the conclusion was that the intervention worked. We would ask you to sit with that last part, because it is the part clients underweight. The professional network did not sell your career. It assigned you to a condition.
Common Questions
Is LinkedIn enshittified?
By EnshitifAi's reckoning yes, though the mechanism is unusual. LinkedIn's degradation is not primarily about price or feature removal; it is about what the member graph is used for. Between 2015 and 2019 the company ran undisclosed randomized experiments on more than 20 million members that measurably changed who they met and where they worked. In August 2024 it turned on AI training against member data by default. In October 2024 an EU regulator found it had no valid lawful basis for the targeted advertising it had been running for years. The product still looks like a professional network. It is also a research instrument and a training corpus.
Did LinkedIn experiment on its users?
Yes, and the results were published. From 2015 to 2019 LinkedIn randomly varied the proportion of strong and weak ties surfaced by its 'People You May Know' algorithm across more than 20 million members, a design intended to test the sociological theory that weak ties matter more for finding work. Members were not told the tests were running. Those users created over 2 billion new connections, filed more than 70 million job applications, and landed roughly 600,000 jobs. The study appeared in Science, co-authored by researchers at LinkedIn, MIT, Stanford and Harvard Business School.
Does LinkedIn use my data to train AI?
By default, yes, unless you changed the setting. In August 2024 LinkedIn introduced a privacy setting permitting it and its affiliates, including Microsoft, to process member data for training AI models, enabled by default. It updated its privacy policy in September 2024 to state explicitly that personal data could be used for generative AI training, and clarified that opting out prevents only future use: data already collected remains embedded in models that have been trained. A proposed class action brought by Premium subscribers in January 2025 over InMail messages sought $1,000 per user under the Stored Communications Act and was voluntarily dismissed without prejudice on January 30, 2025.
Why was LinkedIn fined 310 million euros?
On October 24, 2024 the Irish Data Protection Commission fined LinkedIn Ireland EUR 310 million, finding it had no valid lawful basis for the behavioural analysis and targeted advertising it conducted on member data. The inquiry ran roughly six years, beginning with an August 2018 complaint that the non-profit La Quadrature du Net filed with the French data protection regulator. EnshitifAi's $140B Extraction Index is an editorial opinion anchored to that finding and the record around it, not a measured company financial.
Key Initiatives
- 01
Ran a randomized controlled trial on more than 20 million members between 2015 and 2019, varying the ratio of strong to weak ties surfaced by 'People You May Know' — the treatment variable was who you were shown, and the outcome variable was whether you got a job
- 02
Declined to inform the members that an experiment was underway, on the reasoning that the feed was always going to show them something and the only open question was what
- 03
Published the results in Science with co-authors from MIT, Stanford and Harvard Business School, converting an undisclosed intervention on 20 million careers into a peer-reviewed contribution to the literature on weak ties
- 04
Recorded the yield honestly: 2 billion new connections, over 70 million job applications, and roughly 600,000 positions filled — a number we encourage clients to read as evidence that the allocation of opportunity is a tunable parameter
- 05
Enabled AI training against member data by default in August 2024, extending processing rights to affiliates including Microsoft, and updated the privacy policy in September 2024 to say so plainly
- 06
Structured the opt-out prospectively, so that members who declined prevented future collection while their existing contribution remained embedded in models already trained — consent, in this construction, is a forward-looking instrument only
- 07
Absorbed a EUR 310 million Irish Data Protection Commission fine on October 24, 2024 for conducting behavioural analysis and targeted advertising with no valid lawful basis, concluding a six-year inquiry that began with a single August 2018 complaint from a French non-profit
- 08
Was acquired by Microsoft for $26.2 billion in June 2016, after which the professional graph of over a billion people became an input to a second company's product roadmap without any member being asked to re-consent to the new owner
“A control group is not a cohort we treated differently. It is the cohort we treated as the baseline. I have never once been able to explain, to anybody's satisfaction, why those two sentences feel like different sentences.”
Principal Architect, Strategy Layer
Opportunity Allocation
Want results like these?
Schedule your complimentary Value Extraction Assessment today.
The Extraction Index is EnshitifAi's subjective editorial opinion, grounded in the dated, publicly documented corporate behaviour described on this page. It is not a measurement of LinkedIn's revenue, profit, or conduct, and it is not reported by or endorsed by LinkedIn. EnshitifAi is a satirical parody publication.