LinkedIn started using European member data to train content-generating AI models on November 3, 2025. That date has passed. Advice written before it is still in circulation, in the future tense, built around an opt-out deadline that has already expired.
LinkedIn's own help pages describe something narrower than a deadline. The limits on the setting matter far more than the date did.
What the Data for Generative AI Improvement setting actually controls
Data for Generative AI Improvement is a member setting that controls whether LinkedIn and its affiliates use your data to train AI models that generate content. It is on by default. It works going forward only. LinkedIn's help page for the setting, checked on September 29, 2026, says so plainly: "Opting out does not affect training that has already taken place."
That same page links the control directly, at linkedin.com/mypreferences/d/settings/data-for-ai-improvement.
What LinkedIn confirmed happened on November 3, 2025
A second document, LinkedIn's terms and data use update, states: "On November 3, 2025, we started to use some data from members in these regions to train content-generating AI models that enhance your experience and better connect our members to opportunities." The regions named are the EU, the EEA, Switzerland, Canada and Hong Kong. The UK is covered in its own section, same date.
European members are not exempt. LinkedIn says it started using data from these regions on that date, which means they sat outside that scope before. Any guidance telling a European executive that stricter local rules keep them out is describing the world before November 3, 2025.
The legal basis is legitimate interest, not consent: "We rely on legitimate interest to process your data for this purpose." That is why it is an opt-out instead of an opt-in and why a formal objection route exists alongside it.
What the opt-out does not cover
Three of the carve-outs inside this setting are documented on LinkedIn's own pages.
Training that already happened. Not retroactive, by LinkedIn's own account. The Irish Data Protection Commission put the technical reason under it in an AI Insights Report published on September 24, 2026: "With the current state of the art, once an AI model is trained it can be very difficult to remove any training data that might be part of the model." The same report notes that "there may be no effective way for users to exercise their Article 21 rights if not provided the opportunity to opt out or object prior to the processing beginning."
Every model that is not a content generator. The setting "does not apply to our training of AI models used for other purposes, such as models used to personalize your LinkedIn experience or models we use for security, safety, or anti-abuse purposes." Models that rank the feed and models that run trust and safety sit outside the toggle. Objections to those go through a separate LinkedIn Data Processing Objection Form. Scoring systems were never member-controlled here, and the Social Selling Index shows one slice of what runs underneath.
Feedback you submit. LinkedIn states that feedback and improvement data may be used to improve its generative AI models and is not covered by this setting. A thumbs-down on an AI suggestion travels on a different track.
The setting does not downgrade the account: "Your setting choice does not change the access you have to GenAI products and features offered on LinkedIn."
Every profile field LinkedIn lists as in scope
A third document, LinkedIn's generative AI FAQ, names the profile fields used to train content-generating models: name, photo, current position, prior work experience, education, location, skills, certifications, licenses, volunteering experiences, publications, patents, endorsements and recommendations. Fourteen fields.
The same page adds content typed into AI features, responses to screening questions, resumes saved for future use, group activity and group messages, plus member posts, articles, poll responses, contributions and comments in all formats.
LinkedIn's exclusion list names private messages including InMail and Inbox, login credentials, payment and credit card data, plus salary or job application data attributable to a specific member.
Private messages are out. Group messages are in. For anyone treating a private LinkedIn group as a back room, that is the line that matters.
LinkedIn's systems already read profile language, though not through a keyword affinity score, which is not a real LinkedIn metric. What changed on November 3, 2025 is the purpose, not the reading.
What the Irish regulator actually said
LinkedIn's European lead supervisory authority is Ireland's Data Protection Commission. Its statement of November 7, 2025 has two halves, and each one reads very differently on its own.
The first half: "The DPC has not approved, or found compliant, LinkedIn's use of users' personal data for generative AI model training."
The second half, the next sentence in the same paragraph: "However, the additional measures implemented by LinkedIn have sufficiently addressed the DPC's concerns such that further regulatory intervention is not considered necessary at present."
Quoting either half alone misrepresents it. The DPC did not bless the processing and did not stop it. It extracted changes first: better transparency notices, a reduction in the scope of the data used and the period it was drawn from, stronger measures for under-18s, filters on sensitive material shared in certain Pages and Groups including trade union content, plus fuller risk documentation.
The DPC also required LinkedIn to compile a report within five months of the commencement of processing, on how those safeguards worked. Five months from November 3, 2025 fell in early April 2026. The DPC's AI Insights Report of September 2026 does not disclose the result, and a search of its published statements on September 29, 2026 turned up no outcome. The deadline passed quietly.
Ireland was not the only regulator. The Dutch data protection authority published its own warning on September 24, 2025, flagging the default-on design and urging members to change the setting before November 3.
The Microsoft question, answered precisely for Europe
LinkedIn's privacy policy defines Affiliates as companies controlling, controlled by or under common control with LinkedIn. Microsoft Corporation and its subsidiaries are among the examples it lists. That is why "LinkedIn and its affiliates" appears throughout.
The regional wording is where it gets specific. In the EU, EEA and Switzerland section, the phrase is "use member data and content." In the UK and the Canada and Hong Kong sections, it is "use and share member data and content." Those two sections also carry a line the European section does not: "This setting also controls whether LinkedIn can share this member data and content to our Affiliate Microsoft for their model training activities." For Europe, the help page for the setting states instead: "if you are in the Designated Countries your personal information will not be made available to certain LinkedIn affiliates (such as Microsoft) for training generative AI models without further notice."
That wording is a real difference from the other regions, and it promises less than a quick read suggests. It says certain affiliates and names Microsoft only as an example. LinkedIn's generative AI FAQ states the same commitment without the word "certain." A commitment made "without further notice" can be changed by giving notice. A narrower exposure, not a guarantee.
What this changes for an executive profile
Nothing here argues for a thinner profile. A profile too thin for a model to read is usually too thin for a buyer. Profile views with no conversations is not a privacy problem. It is a clarity problem.
Two decisions do change. The first is what goes into group discussions, because group activity and group messages are in scope while private messages are not. The candid industry group is now a different kind of room than the inbox.
The second is unverified claims. A profile written to impress an algorithm ages badly, and tactics that borrow authority age worse once that text is training data rather than a page that can be quietly rewritten. Opting out does not affect training that has already taken place, so an edit today does not reach what a model already read.
Frequently asked questions
Does turning off Data for Generative AI Improvement delete data already used for training? No. LinkedIn states that opting out does not affect training that has already taken place. The Irish DPC noted in September 2026 that with the current state of the art, removing training data from a trained model can be very difficult.
Are members in the EU exempt from LinkedIn AI training? No. LinkedIn's help page states that on November 3, 2025 it started using data from members in the EU, EEA, Switzerland, Canada and Hong Kong to train content-generating AI models. Guidance claiming a European exemption describes the position before that date.
Does the setting stop LinkedIn using profile data for the feed or for safety systems? No. It covers content-generating models only. Objections to training for personalization, security, safety or anti-abuse models go through the separate LinkedIn Data Processing Objection Form.
Are private LinkedIn messages used for AI training? LinkedIn states that private messages, including InMail and Inbox messages, are not used. Group activity and group messages are listed as in scope.
The deadline was never the story
November 3 looked like the decision point. It was not. The decision is what a senior professional is willing to publish on a permanent, machine-read record, in a system where the regulator has declined to certify the processing as compliant and declined to stop it.
The toggle is worth setting. It is a floor, not a fix. The profile is still the asset, and it should still be written for the people reading it.