Checked 2026-08-27: at least six major platforms put their conclusions about you inside your own data export, under names like Inferences.json, personalization.js and Ad_Targeting.csv, and two more show them on screen without handing you a file. This page lists every file name, where it sits, and which export tier you have to pick to get it.
Both are free under CC BY 4.0, and both carry a permanent id on every row, so you can cite one platform instead of the whole page. Cite the index as: Pupul, The inference file index, pupulcorp.com/knows, checked 2026-08-27. Cite one row by its id, for example pupulcorp.com/knows#inference-spotify.
If one of these files is already on your machine, the free Count opens it in this browser and prints what it finds. Nothing is uploaded, and there is no account or email field.
An inference file is the part of a platform's data export where the platform writes down its own conclusions about you: not what you did, but what it has decided you are.
Your watch history is what you did. The interest categories, the inferred age band, the advertiser audiences you were placed in: those are what the platform concluded. Most guides tell you how to download your data. Almost none tell you which file inside the zip holds the conclusions, and on several platforms that file only exists in an export tier most people never request.
| Platform | Inference file | Exact file or section | Tier you must request | What it holds | Status |
|---|---|---|---|---|---|
| Spotify # | Yes | Inferences.json | Technical log information, a separate request from Account data | The marketing segments Spotify has filed you under | reported |
| LinkedIn # | Yes | Inferences file plus Ad_Targeting.csv | The larger data archive only, not the fast file selection | Conclusions LinkedIn drew about you, plus the targeting attributes advertisers can select | reported |
| X # | Yes | data/personalization.js | Full archive | Assigned interests, inferred age band, gender and languages, and the advertiser audiences you were added to, by company name | reported |
| TikTok # | Yes | Ad interest categories inside user_data_tiktok.json | All data, not the Activity-only selection | The advertising interest categories you were placed in, and off-TikTok activity other companies reported | reported |
| Instagram # | Yes | Ads interests and topics files | All information, or the Ads information selection | The interest topics Meta uses to pick ads for you. Meta renames these folders often | reported |
| Amazon # | Yes | Advertising audience files | The Advertising category, or Request All Your Data | The advertising audience segments you sit in | reported |
| Google # | View-only | No file in Takeout for most accounts | None: read it on screen at My Ad Center | Interest topics and demographic guesses behind your ads | reported |
| ChatGPT # | View-only | Saved memories under Settings, then Personalization | The export centers on conversations.json | What ChatGPT saved about you as memory. What the zip includes has changed over time | reported |
| Claude # | None found | conversations.json only | Full export | Conversations with timestamps. No separate file of conclusions found in reports | reported |
| Netflix # | None found | Activity logs only | Full data request | Viewing activity and account records. No separate file of conclusions found in reports | reported |
| Grok # | None found | prod-grok-backend.json only | Full export from accounts.x.ai | Your grok.com conversations. Grok used inside the X app lands in the X archive instead | reported |
Each platform name links to the export guide with the exact taps to request the right tier. The # beside each platform links to that one row, so a claim about one platform can be cited on its own. Rows last checked 2026-08-27.
The useful unit here is one platform, not eleven. Every row has a permanent id, and the id is the link.
The claim that Spotify keeps Inferences.json behind the Technical log information request is
pupulcorp.com/knows#inference-spotify, and it will still be that tomorrow.
id column
and a row_url column holding the full link, so a tool and a reader land on the same row.Every row above is currently labeled reported: it comes from the platforms' own documentation and from consistent public reports, checked 2026-08-27. When we confirm a row by opening the file inside an export of our own, that row changes to verified and carries the date we opened it. A row we cannot confirm stays labeled reported or comes out. If a platform stops shipping a file, that is worth knowing too, and the row will say so with a date.
The count, rather than the promise: 11 rows reported, 0 verified. We have not opened
any of these files in an export of our own. The same count is in the JSON, under
status_counts, so a tool reading the data sees it too.
The free Count runs in your browser, reads your export locally, and shows you the interest categories and audiences inside it. No account, no email field, no upload.
The full Reading is 6.99 a week or 69.99 a year. No free trial. There is a Refund button in your dashboard for 7 days, one tap, no reason required. The Count stays free forever either way.
An inference file is the part of a platform's data export where the platform writes down its own conclusions about you: not what you did, but what it has decided you are. Spotify's Inferences.json and X's personalization.js are the clearest examples reported today.
Usually because it lives in a tier you did not request. Spotify keeps it in Technical log information, LinkedIn in the larger archive, TikTok in the All data selection. The table above names the tier per platform, and each guide gives the exact taps.
Mostly no, as reported 2026-08-27. Google's inferences are shown on screen at My Ad Center rather than shipped as a file, though many accounts can include a My Ad Center selection in Takeout.
Yes. The page, the CSV at /data/inference-files.csv and the JSON at /data/inference-files.json are licensed CC BY 4.0. Cite the whole index as: Pupul, The inference file index, pupulcorp.com/knows, checked 2026-08-27. To cite one platform, use that row's permanent id, for example pupulcorp.com/knows#inference-linkedin.
No. The count is 11 reported and 0 verified. Every row comes from the platforms' own documentation and consistent public reports, checked 2026-08-27, and we have not opened any of these files in an export of our own. A row changes to verified only on the day we do, and it carries that date.
One page per file. Each is the same browser local tool scoped to a different export, it reads the files named above on your own device, and none of them ask for an email.
Same shape every time: the exact taps, the reported wait, the real size, what is inside, and what is missing.