Think about the last case study your firm posted on LinkedIn. A client win, the approach that got the result, maybe a comment thread underneath where your team added detail. Since late 2025, that post has a second life. Unless the person who wrote it has switched off a setting buried in their privacy menu, LinkedIn can use it to train the generative AI models it builds for itself and for Microsoft. The same goes for every staff profile, every job application handled through the platform, and every group discussion your people join. The opt-out exists, but it sits on individual accounts, and it arrived switched on by default. That combination makes it a business decision wearing the clothes of a personal privacy setting.
What is LinkedIn’s AI data use opt-out?
LinkedIn’s opt-out is a setting called Data for Generative AI Improvement, found under Settings & Privacy, then Data privacy. It is switched on by default, which means LinkedIn and its Microsoft affiliates can use your profile, posts, job applications, group activity and feedback to train content-generating AI models. Turning it off stops future training only. Data already used stays used.
The change was announced on 18 September 2025 and training began on 3 November 2025 for members in the EU, EEA and Switzerland. The UK took a different route. The Information Commissioner’s Office raised concerns in 2024, LinkedIn suspended training on UK data while the two engaged, and it has since resumed with clearer notices, a simpler objection route and a longer window to act. The ICO says it will keep monitoring.
What goes in is broad. Profile data, including work history, skills and recommendations. Job-related data such as CVs and screening answers. Posts, articles, comments, poll responses and group activity. In some regions the training corpus can reach back to 2003. Private messages, passwords and payment details are excluded, and LinkedIn says under-18 data is kept out. Nearly everything else professional you have ever published there is in scope.
Why does it matter for your business?
Staff LinkedIn activity is organisational data in everything but name. Profiles list clients, projects, specialisms and career history. Posts describe methods and results. Under the new default, all of it can feed models that generate suggestions for other users, including your competitors. And because the setting lives on individual accounts, there is no company-level control. Your stance only exists if you communicate one.
Consider what a twenty-person consultancy publishes in a normal year. Named specialisms on every profile. Case studies describing how engagements were approached. Endorsements that map who is good at what. Screening questions that reveal how the firm hires. Fed into a model, that material can shape the suggestions LinkedIn serves to every other firm in your niche, from post structures to job descriptions. The distinctiveness you built in public becomes pattern data.
The limits of the opt-out sharpen the point. It covers content-generating AI only, so personalisation, security and anti-abuse processing continue regardless. It cannot stop your material entering training when someone else reposts it. And it does nothing about history. For a firm, encouraging staff to opt out caps future exposure rather than recovering anything, which is why guidance on what gets posted matters at least as much as the toggle itself.
Where will your firm actually meet it?
Three places, mainly. Recruiting, where LinkedIn’s AI features now sit inside candidate search, matching and outreach. Marketing and business development, where staff posts and case studies become training material. And staff onboarding, where new hires arrive with the default switched on and no idea the firm has a view. Each is an ordinary activity that now carries a data decision inside it.
Recruiting is the sharpest edge. LinkedIn’s Recruiter platform includes AI candidate matching, conversational search and AI-assisted outreach, and its Hiring Assistant agent claims to automate up to 80 per cent of the pre-offer workflow. Early adopters include Siemens, Canva and AMS. LinkedIn reports that AI-assisted outreach gets a 44 per cent higher acceptance rate and responses 11 per cent faster, with AI search producing 18 per cent higher candidate acceptance. Those numbers explain why the platform wants training data this badly, and why a small firm that hires through LinkedIn is already inside the loop.
Your own obligations sit alongside. When you run candidates through LinkedIn’s AI features, you remain a data controller for your recruitment activity under UK GDPR. The ICO expects a lawful basis, transparency with candidates, and a data protection impact assessment where AI-driven screening could significantly affect people. None of that requires heavy infrastructure at five to fifty staff, but it does require being able to explain what the tools are doing with candidate data.
When should you opt out, and when is it fine to leave it on?
Opt out as the default stance if your staff profiles and posts carry client detail, methodology or anything a competitor would find useful. Leave it on where the firm gains real value from LinkedIn’s AI features and staff share little beyond generic content. Either way, decide deliberately, tell the team, and write the decision down. Inconsistent individual settings are the worst outcome.
The stance splits broadly by sector. A firm handling legal, financial or health matters, where client confidentiality is the product, has little to gain from feeding LinkedIn’s models and should recommend opt-out to all staff. A marketing or design agency that lives on LinkedIn visibility and uses the AI writing and recruiting features may reasonably leave the setting on for some roles while asking fee-earners with sensitive client exposure to switch it off.
Then make it real on Monday. Write a short stance statement, a paragraph is enough. Issue a one-page instruction with the exact path, Settings & Privacy, then Data privacy, then Data for Generative AI Improvement, and a reminder that the switch does not undo past use. Add the check to onboarding so new hires make the decision with the firm’s view in front of them. Record who was told and when. The ICO’s accountability principle rewards exactly this kind of modest, documented decision-making.
How does this fit into wider AI governance?
LinkedIn is one instance of a wider pattern. Public platforms increasingly treat member content as training data by default, employees bring their own AI tools to work, and regulators expect organisations to have safeguards in place before personal data feeds any model. Treat the LinkedIn setting as the entry point for a broader conversation about how your firm handles AI and data.
The comparison across platforms is instructive. Meta offers US users no simple toggle to keep public posts out of AI training. X lets users switch off data sharing with its Grok models. Snap’s My Selfie feature is opt-in only, which shows default-off is perfectly feasible and that LinkedIn’s default-on was a commercial choice. Meanwhile Microsoft and LinkedIn’s own Work Trend Index found 75 per cent of knowledge workers using AI at work and 78 per cent of AI users bringing their own tools, so your team is almost certainly feeding several models already, with or without guidance.
That is the real value of this episode for an owner-managed business. One visible, well-documented platform change gives you a concrete reason to set expectations about AI and data generally, before the EU AI Act’s treatment of employment AI as high risk starts shaping UK practice too.
This needs no committee. A founder-level decision, one page of guidance, and a recurring fifteen-minute check on whether the policy landscape has moved will cover it. The firms that handle it well will be the ones whose staff can say what the firm’s position is and why, to a client, a candidate or a regulator. If you want help setting an AI governance baseline that fits a firm your size, book a conversation.
