What Google, Meta, Amazon and TikTok Actually Know About You

Given versus concluded: the two halves of your file

Think of your data file at any large platform as a folder with two sections. The first section is everything you consciously handed over: your name, your email address, the search you typed, the video you played, the item you bought. You knew you were doing those things. The second section is everything the platform worked out from watching you do them: your likely income bracket, your political leanings, whether you're going through a life event, how impulsive you are when shopping at night. You did not hand that section over. It was built.

Most privacy conversations stop at the first section. That's understandable, because the first section feels like the obvious problem. But the second section is where the commercial value lives. Advertisers do not pay a premium to reach "people who searched for running shoes." They pay to reach "people our model predicts will buy running shoes in the next two weeks, who also respond to scarcity messaging, and whose household income sits above a certain threshold." That targeting comes from inference, not from what you typed.

Under UK GDPR, you have the right to access both halves. Article 15 covers not just the raw data but the logic used to process it, the purposes it serves, and any inferred attributes that form part of a profile. The platforms provide tools to exercise this right. Most people never open them. This piece is about opening them.

One practical note before we start: what you see in these tools reflects what the platform is willing to surface through a consumer-facing interface. A formal Subject Access Request, submitted in writing to the company's data protection team, can go further. The ICO's guidance on your data rights explains how to escalate if a platform's response feels incomplete.

Google: My Activity, location history and the ad centre

Google's file on you is probably the longest of the four, simply because the surface area is so large. Search, YouTube, Maps, Gmail, Chrome sync, Android location, Assistant queries, Play Store downloads. Each service feeds a separate data stream, and those streams converge into a single profile.

The best starting point is My Activity. This shows your timestamped history across Google products: every search, every YouTube watch, every Maps route. It's organised by date and by product, and you can filter to see just one service at a time. Scroll back a few months and the picture becomes vivid quickly. The volume alone is instructive.

Location history lives in a separate tool called Timeline (formerly Google Maps Timeline). If location reporting has been enabled on your Android device or through the Maps app, this shows a map of where you've been, often with inferred labels: "you visited this gym," "you were at this hospital." Google uses this data to improve its services but also to build what it calls "store visit" signals for advertisers, allowing a retailer to measure whether someone who saw their ad later walked into a physical shop.

The inferred layer is most visible in the Ad Settings centre. This page lists the interests and demographic attributes Google has assigned to your account: age bracket, parental status, homeowner status, and a long list of topic interests ranging from the obvious ("cooking") to the unexpectedly specific. These are the categories advertisers can select when targeting you. You can remove individual categories here, though Google will simply re-infer them from future behaviour unless you also reduce what you share.

For the full picture, Google Takeout lets you download an archive of your data across all Google products. The archive can be enormous, so it's worth selecting specific products rather than downloading everything at once.

Practical takeaway: Open Ad Settings now and look at the demographic attributes listed under "How your ads are personalised." The gap between what's listed there and what you remember consciously telling Google is a reasonable proxy for how much has been inferred.

Meta: the Accounts Centre and the off-app activity feed

Meta's data infrastructure spans Facebook, Instagram and WhatsApp, and the central access point is the Accounts Centre. From there you can manage your connected accounts and reach the data download tool, which produces a structured archive of your posts, messages, reactions, searches, and the ads you've seen and interacted with.

But the part most people don't know about is the Off-Facebook Activity tool, now labelled as "Your activity off Meta technologies" within the Accounts Centre. This is genuinely striking. It shows a list of third-party websites and apps that have sent data about your behaviour back to Meta, even when you weren't using Facebook or Instagram at the time. A retailer's website with a Meta Pixel embedded in it, a news site, a health app, a booking platform: if they use Meta's advertising or analytics tools, your visits are reported back. Meta uses this data to build a more complete picture of your interests and to close the loop on whether its ads led to purchases elsewhere.

The inferred profile itself is visible through the ad preferences section of the Accounts Centre. Meta lists the interests it has assigned to you, the advertisers who have uploaded contact lists that matched your account ("Custom Audiences"), and the broader demographic and behavioural categories used to target you. Researchers at the Norwegian Consumer Council documented in detail how granular these categories can become, including inferences about political views, relationship status transitions, and financial behaviour.

What makes Meta's inference engine particularly sophisticated is the combination of on-platform signals (what you like, share, pause on, comment on) with off-platform signals (the websites you visit, the apps you use) and social graph signals (who your friends are and what they do). None of those three inputs alone would produce the same result. Together, they produce a profile that's considerably more detailed than most people expect.

The most revealing part of your Meta file isn't your posts or your messages. It's the list of third-party websites that have been quietly reporting your visits back to Meta for years, building a picture of your life that extends well beyond anything you did on Facebook or Instagram.

You can clear your off-Meta activity history and instruct Meta to disconnect future off-platform data from your account. This doesn't stop the data being collected at source; it just means Meta can't tie it to your specific profile. The Meta help page on off-Facebook activity explains the controls in plain language.

Practical takeaway: Find "Your activity off Meta technologies" in the Accounts Centre and look at how many businesses appear on that list. Then check the ad preferences section to see what interests Meta has inferred. The contrast between the two is informative.

Amazon: purchases are the least of it

Amazon knows what you bought. That part is obvious. What's less obvious is the breadth of the behavioural data that surrounds each purchase: what you searched for before buying, what you looked at and didn't buy, how long you spent on a product page, which items you added to a wish list and left there, which reviews you read, and how your browsing patterns shift across seasons and life events.

Then there's the ecosystem beyond the shop. If you use Alexa, Amazon holds a history of your voice queries. If you use Amazon Prime Video, it holds your viewing history and the inferences it draws from it. If you use Amazon's advertising network (which serves ads on third-party apps and websites), it builds a profile of your interests from your shopping and browsing behaviour that extends into those external environments. Ring doorbell footage, Kindle reading habits, Fresh grocery orders: each Amazon product feeds a separate data stream, and Amazon's advantage is that it can connect purchase intent directly to purchase behaviour in a way that other platforms can only approximate.

Amazon's data request tool lets you download your data across its services. The categories available include browsing history, search history, Alexa interactions, Prime Video watch history, and advertising preferences. The advertising preferences section, like Google's and Meta's equivalents, shows the interest categories Amazon has assigned to your account for ad targeting purposes.

There's also a simpler starting point: your browsing history page, which shows everything you've viewed on Amazon. Many people are surprised to find items they browsed years ago still recorded there. This history is used to personalise recommendations and, if you're opted into personalised ads, to inform targeting across Amazon's advertising network.

One thing worth understanding about Amazon specifically: because it operates at the point of purchase rather than earlier in the journey, its inferences about commercial intent are particularly valuable to advertisers. Knowing someone searched for "cordless drill" on Amazon is a much stronger buying signal than knowing they searched for the same term on a general search engine. That specificity is what makes Amazon's advertising business significant even relative to its retail operation.

To understand more about how the data Amazon collects feeds into the broader market for consumer profiles, the piece on how your data is collected and sold covers the data broker economy in detail.

Practical takeaway: Visit your Amazon browsing history and clear any items you're comfortable removing. Then request your data download and, when it arrives, open the advertising preferences file to see which interest categories Amazon holds for you.

TikTok: the inference engine and your data download

TikTok's approach to inference is worth understanding on its own terms, because its method is somewhat different from the other three. Google, Meta and Amazon have years of explicit behavioural data to work from: things you searched, things you liked, things you bought. TikTok's primary signal is much simpler and, it turns out, extremely powerful: how long you watch each video, and whether you rewatch it.

You don't need to like a video for TikTok to learn from it. You don't need to share it or comment on it. Simply pausing on a video for longer than average tells the algorithm something. Rewatching a clip tells it more. The "For You" feed is built almost entirely on this implicit engagement signal, which means TikTok is inferring your interests and emotional responses from behaviour you may not even have been conscious of performing.

Researchers studying TikTok's recommendation system have noted how quickly it can identify sensitive interests, including mental health concerns, political sympathies, and sexual orientation, from watch-time patterns alone, often within a relatively small number of sessions on a new account. TikTok's own privacy policy acknowledges that it infers attributes about users from their activity. The advertising platform allows targeting by inferred interest categories, age, gender, location, device type, and behavioural segments.

To see your data, go to your TikTok profile, tap the menu, then Settings and Privacy, then Privacy, then Personalisation and Data. The "Download your data" option produces a file that includes your browsing history, video watch history, liked videos, comments, direct messages, and the interest categories TikTok has inferred. The file arrives in JSON or text format, which isn't the most readable, but the interests list is worth finding. It often contains categories that feel accurate in a slightly unsettling way, precisely because they were inferred from watching behaviour rather than anything you explicitly declared.

TikTok is also a useful case study in how companies track you online more broadly, because its tracking extends beyond the app: the TikTok Pixel functions similarly to Meta's, allowing third-party websites to report user behaviour back to TikTok's advertising infrastructure.

Practical takeaway: Request your TikTok data download and, when it arrives, look specifically at the inferred interests list. Compare it to what you'd have said your interests were if someone had simply asked you. The divergence, if there is one, is the inference engine at work.

Why the inferred file is the valuable one

It's worth being precise about why inference matters commercially, because "they know a lot about you" is too vague to be useful.

The advertising market works on probability. An advertiser selling, say, private health insurance doesn't want to reach everyone in the UK. They want to reach people who are likely to be considering private cover right now, who have the income to afford it, and who respond to particular kinds of messaging. None of that can be derived from a name and an email address. All of it can be derived from a sufficiently rich behavioural profile.

Inferred attributes are more valuable than declared ones for a specific reason: people don't always tell the truth when asked directly, but their behaviour is harder to fake. Someone who says they're not interested in luxury goods but consistently browses premium products at night is a more accurate signal than a survey response. The platforms know this, which is why they invest so heavily in inference models rather than simply asking users to fill in preference forms.

The inferred file is also the part most likely to contain errors. Because inference is probabilistic, it's wrong sometimes. You might be assigned to an interest category because you watched one video on a topic that a family member chose. You might be flagged as a homeowner because you searched for mortgage information on behalf of a friend. Under UK GDPR, you have the right to request correction of inaccurate personal data, including inferred attributes. The ICO's guidance on the right to rectification explains how this works in practice.

The inferred profile doesn't stay with the platform that built it, either. Parts of it flow into the data broker economy, where profiles are aggregated, bought, and sold. That's a longer story, but the short version is that the file you see in a platform's privacy centre is not necessarily the only place that file exists.

The four ten-minute checks, back to back

Here are four focused checks you can run in sequence. Each takes roughly ten minutes if you stay focused and don't go down every rabbit hole (there will be rabbit holes).

• Google: Open Ad Settings at adssettings.google.com. Read the demographic attributes and the interest list. Then open My Activity and filter to Search. Pick any week from six months ago and read what you were searching for. Notice the gap between those two things.
• Meta: Open the Accounts Centre and find "Your activity off Meta technologies." Count the number of businesses listed. Then go to ad preferences and read the interest categories. Clear your off-platform history if you want to disconnect it from your profile going forward.
• Amazon: Visit your browsing history at amazon.co.uk/gp/history. Scroll back as far as it goes. Then visit the advertising preferences section of your account settings and read the interest categories listed there.
• TikTok: Request your data download via Settings and Privacy. While you wait for it to arrive (it can take up to 30 days, though often much faster), go to Settings, then Ads, and look at the "Ad interests" section for the categories currently assigned to you.
• After all four: Note which inferred attributes surprised you, which felt accurate, and which felt wrong. Inaccurate attributes can be challenged under your right to rectification.
• If you want to go further: Submit a formal Subject Access Request to any platform whose response feels incomplete. The ICO's template letters are a useful starting point.

What seeing your file changes about how you use these services

People who have read their own data files tend to describe a similar experience: not panic, but a shift in perspective. The services feel less neutral afterwards. Not sinister, exactly. Just more legible.

That legibility is genuinely useful. Once you know that your YouTube watch history is feeding your Google ad profile, you might be more deliberate about what you watch while logged in versus logged out. Once you've seen the off-Meta activity list, you might think differently about which websites you visit while your Facebook session is still open in another tab. These aren't dramatic behaviour changes. They're small calibrations made possible by understanding what's actually happening.

Some people decide, after reading their file, that they're broadly comfortable with the exchange: personalised services in return for a detailed profile. That's a legitimate position, and it's more meaningful when it's informed. Others decide they want to reduce the profile without abandoning the services, which is possible to a meaningful degree through the controls each platform provides. A few decide they want to use privacy-preserving alternatives for some activities, such as a search engine that doesn't build a profile, or a browser that limits cross-site tracking by default. The Mozilla Foundation's privacy resources are a good starting point for the browser question.

What changes most, for most people, is the feeling of passivity. Before reading your file, data collection is something that happens to you, invisibly, in the background. After reading it, you have a much clearer sense of the mechanism. You know what's being collected, where to look at it, and what rights you have over it. That knowledge doesn't require you to do anything dramatic. It just means you're no longer operating entirely in the dark.

The platforms are not going to stop building these profiles. The advertising model that funds free services depends on them. But the legal framework, the tools, and the right to see and challenge your own file: those exist, they work, and they're available to you right now. Using them is not a protest. It's just knowing what's in your folder.