You Are Not the User

The other day I was on a gym treadmill doing cardio and doom-scrolling YouTube Shorts, the kind of trance where your legs keep moving and your thumb keeps swiping, when an ad cut through the feed. Upload your voice and your client work, it said, and an AI twin trained on you will answer enquiries while you sleep. It markets itself as the first marketplace where professionals build a twin that sells their services around the clock. The categories run to a hundred, from accountant to architect. viame The promise is the same one half a dozen startups are running this year. They are not wrong that you need a twin. They are wrong that uploading your corpus to a fourth landlord fixes agency when larger platforms already built one from the content you gave away.

LinkedIn, Meta, Google, and X already shipped that twin, months before the marketplace existed. Every platform twin has three roles. You supply the corpus. The platform hosts and trains the model. The paying customer, a recruiter, advertiser, or seat-holder, gets the useful output. You are not in the third seat. You are not the customer. You are the product the customer rents. A platform, in this sense, is the company that hosts your professional content and your attention. An AI twin is a model built from a specific person's content, voice, and judgement. The twin works. It just does not work for you.

You Are Not the User

The Social Dilemma, a Netflix documentary from 2020, dramatised the economics years before the twin marketplaces arrived. Actors played engineers inside a teenager's phone, tuning engagement dials until he was primed for an ad auction. Tristan Harris called it an avatar voodoo doll on the other side of the screen. social-dilemma You were never the user in that scene. You were the doll. The recruiter, the advertiser, and the platform's own growth team are the users now. You are the training set. "You are not the user; you are the product" has been attributed to Seth Godin, the bestselling author and marketing expert. godin-choke-point LinkedIn and Meta did not invent the economics. They scaled them on a professional corpus you already supplied.

The same Google that served those Shorts can now link your watch history, Gmail, and Workspace files to Gemini, and train on summaries from that material when your activity setting stays on. google-gemini-connected X shares public posts with Grok unless you find the opt-out under Privacy and safety. ars-grok-x LinkedIn and Meta switched generative training defaults on outside a few designated privacy regions. linkedin-help-gai You train it. They query it. They bill for it. You never see the invoice. You may get reach. You do not get a revenue share.

They Already Have the Corpus

The advice to post consistently and build your brand is not wrong. It is incomplete, in a way that benefits one side of the relationship far more than the other. LinkedIn's help pages describe a setting called Data for Generative AI Improvement. It is switched on by default for members outside the EU, EEA, Switzerland, and the UK. It trains generative models on your profile data and on the posts, comments, and articles you publish. You have to find the toggle yourself if you want it off. USA Today reported the same mechanism within days of its 2024 rollout. LinkedIn opted members outside those designated regions in without asking them first. The fix was a setting buried under Data Privacy. Nobody had put the question to them first. usatoday-linkedin-ai

Meta runs the identical play at a larger scale. From April 2025, Meta began training its generative models on public posts, comments, and photo captions. Adults in the EU had already shared that material. meta-europeans Meta also used their conversations with Meta AI. It notified members afterwards and offered an objection form. It did not ask first. reuters-meta-eu It is the same loop I traced through the thumb-to-ledger economics of a feed. That feed predicts, ranks, and retrains on attention it did not have to bid for. The ledger there was your scroll time. The ledger here is your CV.

The Second Harvest

Open LinkedIn's Premium tier and click Write with AI on your own About section. The assistant drafts a personalised summary. LinkedIn says it incorporates insights from analysing millions of top LinkedIn profiles. linkedin-write-ai Your profile is one of the millions feeding someone else's suggested paragraph. Their profile is feeding yours. Nobody asked either of you a separate question before the pooling started. The tool does not tell you whose sentences it borrowed to write yours, and there is no way to ask it.

That pooled corpus is not a hobby project sitting quietly in a lab. In late April 2026, Reuters reported that LinkedIn's agentic AI hiring products were on track to generate $450 million in sales in the coming year. Reuters put the network at more than one billion members. reuters-linkedin-agents That is the same registry logic behind Meta buying the AI agent directory that mattered before most builders noticed the window had closed. There, the directory was agents. Here, the directory is every profile you have ever filled in. An assistant reads it for whoever pays the seat fee, not for you.

Answer Engines Do Not Send Invoices

I nearly used Write with AI on my own About section. I opened the tool and read the first draft it offered. Then I closed it, because the sentence sounded like every other consultant's profile trained on the same millions, mine included. I know a thing or two about design. I have led product and design programmes for Fortune 100 brands, been a Wipro partner, shipped telematics at Aviva, and advised boards after a Chicago Booth MBA. That is not a brag list. It is the point. Twenty years of lived experience, pains, gains, frustrations, frameworks, and arguments with sceptical CFOs do not survive pooling with millions of strangers. Platforms want sameness because sameness scales. An experienced professional's value is the opposite: judgement from a specific vantage point, not a smoothed average of everyone else's headline. I still catch myself, most weeks, drafting a new idea as a LinkedIn carousel before the longform essay on my own site, because the carousel gets read that afternoon. I know which layer I own. I reach for the rented one anyway when the deadline is close.

That habit still costs someone else. Somewhere in LinkedIn's hiring pipeline is a candidate I will call Tunde. His profile was one of the 62 per cent fewer profiles a recruiter had to review by hand, after that recruiter switched on Hiring Assistant. LinkedIn says the same tool saves four hours per open role. It lifts InMail acceptance by 69 per cent. linkedin-hiring-assistant Tunde never sees the sub-agent's read of him. He never gets a citation for the paragraph it wrote about his fit for the role. He is not the customer LinkedIn is billing for that efficiency. The recruiter is. The recruiter's employer pays an invoice Tunde never even learns exists.

The Flat You Improved

Picture a flat you rent, not own. You spend two weeks on your knees in it anyway. You fill the cracks, sand the skirting, paint the kitchen a colour the landlord would never have chosen for it. Your knuckles carry the small cuts from the Stanley knife for a week afterwards, a receipt for effort that nobody records anywhere official.

Then you move out. The landlord relists the flat as recently renovated and raises the rent by a fifth. Those two weeks on your knees became somebody else's asking price, and the new tenant will never learn your name. The receipts belong to the person who owned the walls, not the person who held the paintbrush. Caring about the paint job does not change whose name is on the deed.

The Strongest Case for Posting Where the Room Already Is

Give the objection its full weight before answering it. LinkedIn has more than one billion members. My own site runs a grounded answer surface, Ask. It sees roughly five hundred sessions a month by my own published account. Search Finds Pages. Chat Finds Answers. Reach is a genuine advantage, not a rhetorical one. An unread twin with perfect citations helps nobody at all. Posting where the room already holds a billion people in it is not vanity. For most careers, most of the time, it is close to the entire game.

That argument holds only as long as the room keeps sending the visitor to you. Agarwal and Sen ran a randomised field experiment on Google's AI Overviews, reported by PPC Land in July 2026. Publisher clicks fell 39.8 per cent. Zero-click searches rose 34.5 per cent. More than a third of queries now end without anyone visiting the site that wrote the answer. ppc-land-ai-overviews The platform kept the reader. The publisher lost the traffic. Reach on somebody else's feed is not yours to keep once the feed's own AI answers first.

The Twin I Control

I built the twin the platforms withhold. Call it a retrieval twin if you prefer. The point is you hold the keys. On my website it has three parts: Ask, a public `llms.txt`, and a hand-written retrieval brief behind each essay. A language model can cite that brief rather than guess at the answer. That habit started with training a model on my own writing instead of a generic one, and it has not stopped since. The corpus is mine: how I frame a job to be done before anyone opens Figma, when I kill a project six months in because research invalidated the assumption, what I learned shipping navigation for EE or telematics for Aviva, and what I would tell a CEO at two in the morning. A platform twin averages that judgement away. Mine is supposed to sound like me. You do not need a bespoke answer engine on day one: a domain you hold, a public `llms.txt`, and one page that cites its sources already beats a third-party twin marketplace. You will not see a revenue share when LinkedIn's twin of you screens a candidate. A twin you control can answer enquiries on your terms, send the curious to work you actually price, and retain clients who came for your judgement, not the platform's average. Agency is who can query your model, what it may say about you, whether the reader gets a citation, and whether you can audit or revoke the answer. I can read every source Ask cites. I can change the brief when my view shifts. Nobody at LinkedIn can deprecate the feature tomorrow or fold my posts into a training default I never separately agreed to.

LinkedIn, Meta, Google, and X already built your twin from content you supplied under a default you never read, and they bill other people to query it. Owning your professional identity in 2026 means building and controlling a twin on a domain you hold rather than uploading your corpus to a fourth landlord—one that shares your judgement instead of pooling it into the mean. Tunde will never see the query that screened him out, or a penny of the $450 million LinkedIn aims to bill for profiles like his.