Social Media
You choose who to follow. Instagram increasingly chooses what you see.
Instagram has never charged its customers a cent, and its measured satisfaction is above its category average — higher than Facebook's. Neither fact is in dispute here, and both are credited in full below. What this investigation documents is a fourteen-year transfer of the product's central decision — what appears in front of the customer — from the customer's follow graph to Meta's recommendation systems. The original proposition was a chronological stream of photos from accounts the customer deliberately chose. By Meta's own account to investors, more than half of what customers now see on Instagram is selected by AI Company disclosure; by Meta's own account to a federal court, friends' content now accounts for roughly 7% of time customers spend viewing content on Instagram Meta's court filings. The transfer was executed gradually, announced in fragments, and softened by real free value — but no step of it has ever been durably reversed, and the one control that would let the customer decline it — a non-algorithmic, followed-only feed — has never been allowed to persist past an app restart. Established
CFS = CVI − CHI (76 − 70 = +6), placing Instagram in the index's finely balanced band — deeper inside it than Facebook (+13), the only other assessment in that band. Scores were calibrated against the Facebook assessment, the closest comparable in the index: Instagram's manipulation and extraction evidence is stronger (a preliminary regulatory finding against its core engagement design, and per-user ad-load management described in Meta's own statements and trial record), while its delivered value is comparably strong. Assessed under CHI/CVI Methodology v2.0. Overall CHI confidence: A− — unusually high, because the load-bearing evidence is Meta's own statements. Overall CVI confidence: B+. Methodology →
The defining finding
Why does a feed of the people you follow exist — and reset every single time you close the app?
Since March 2022 Instagram has offered a strict chronological feed of followed accounts, called Following — but it cannot be set as the default, and each new session opens back into algorithmic Home. Established Full mechanics below.
You told Instagram which feed you wanted. Instagram remembers — until you close the app. That is a narrower, harder accusation than "the algorithm is bad", and it is the one the evidence supports.
The product contract · 2012 vs 2026
Same name. Same price. Different product.
This comparison is not nostalgia, and it is not a claim that the old product was better. It is a record of what the product proposition was when customers committed to it, and what it is now — assembled almost entirely from Meta's own announcements, earnings calls and court filings. Meta itself litigated the change as fact, and won with it.
- Accounts you deliberately followedThe follow graph was the product. Nothing else decided what you saw.
- Reverse-chronological orderingEvery post from every followed account, newest first.
- PhotosA photo-sharing app, by its own description.
- No advertisingAds did not exist until November 2013, and arrived as "an occasional ad" from "a handful of brands".
- Recommendations kept separateDiscovery lived in an opt-in Explore tab. The feed was yours.
- A standalone company"Committed to building and growing Instagram independently" — Meta, April 2012. Company statement
- Majority AI-recommended content"More than 50% of the content people see on Instagram is now AI recommended" — the last official figure, April 2024. Company disclosure
- Engagement-predicted rankingHome is ranked by what the model predicts will keep you there; "popularity" is a named top-tier signal in Instagram's own 2023 explainer.
- Reels and entertainment emphasis"We're no longer a photo-sharing app" — Adam Mosseri, June 2021. In 2025 the app began reorienting "around DMs, Reels, and recommendations".
- Advertising across every major surfaceFeed, Stories, Explore, Reels, search, other users' profiles — with ad load managed per customer (see below).
- Creator and commercial contentDistribution restructured around creators and recommendations; friends now account for roughly 7% of time spent viewing Instagram content, by Meta's own trial figures.
- Full Meta integrationOne privacy policy since 2022; Accounts Center; AI training on public posts; Meta AI embedded in search and DMs with no removal option.
- Controls that sit outside the defaultFollowing, Favorites, snooze, reset — real controls, none of which persist or bind the defaults.
The two numbers
Meta's own measurements of the transfer.
The two headline statistics on this page come from Meta itself — one from an earnings call, one from Meta's own filings in federal court. They use different denominators and are shown separately for that reason. The first measures share of content seen; the second measures share of time spent. Combining them into one line would be more persuasive and less true.
Content seen on Instagram that is AI-recommended from accounts the customer does not follow.
Mark Zuckerberg, earnings calls. Q2 2022: about 15% of Facebook's feed and "a little more" of Instagram's, with a stated intention to "more than double" it. Q1 2024: "For the first time ever, more than 50% of the content people see on Instagram is now AI recommended" — verified verbatim against Meta's own transcript for this page. It is also the last official figure; Meta has not published an update since. Company disclosure.
Time on Instagram spent viewing content from friends.
Meta's own figures, advanced in its defense in FTC v. Meta; the court's November 2025 opinion records that friends' content is 7% of Instagram time and that the share fell "by more than a third" in about two years. Court record of Meta's stated position — not an independently audited measurement, and this page uses it only in that capacity. Note the denominator: this is time spent, not share of feed inventory. "Only 7% of the feed is friends" is a different claim and this page does not make it.
Mark Zuckerberg, testifying under oath, April 2025.
In November 2025 the court ruled for Meta, holding that its apps compete in entertainment distribution against TikTok and YouTube rather than in a personal-social-networking monopoly. The FTC has appealed; the ruling is not final.
Meta won its antitrust trial partly by proving the transformation this page documents. To defeat the FTC's market definition, Meta established — with its own composition and time-spent data — that Instagram is no longer principally a friends-and-family network but an entertainment and discovery product. The court accepted that, and Meta prevailed.
Two things must be kept straight about this. First, the court did not rule that Instagram mistreats customers — it ruled on monopoly power, found none proven, and even observed that Meta's apps have continuously improved while remaining free. Nothing here says otherwise. Second, the defense's factual premise is precisely this page's factual premise: the product customers originally signed up for is, by Meta's own successful legal argument, not the product they have. The same shift Instagram presented to customers year by year as making their experience better was presented to a federal court, all at once, as evidence that the friends-first product is gone. Both statements are Meta's. Only one was made to the people using the app.
Pattern one · Algorithmic Replacement — supported, primary
The customer chooses whom to follow. Instagram chooses what the customer actually sees.
Algorithmic Replacement is the gradual substitution of content the customer intentionally chose with content the platform's ranking and recommendation systems select. Facebook established the pattern in this index. Instagram is the sharper case — the original product had no other function than showing you what you asked for, and the replacement is confirmed at every step by Meta's own numbers.
The feed and recommendation controls above share one engineered property: they expire. (Other Instagram settings — Restrict, Hidden Words, notification toggles, and most privacy choices — do persist once set; this finding is specifically about the controls that would let a customer opt out of algorithmic selection.)
Real, and honored for the length of a session.
No persistent default has ever shipped, in any market, in ten years of ranked feeds.
The only way to remove recommendations from Home.
"There's no way to turn off suggested posts permanently." The interface accepts the preference and schedules its expiration.
Erases the model's learned picture of you.
Erase is not off. No off exists.
Pattern two · Advertising Creep — supported, primary
From "an occasional ad" to a per-customer maximization system.
Instagram's advertising history is not just "more ads". It is a documented progression across every surface the product built — including the ones marketed as escapes — ending in a system Meta's own executives describe as managed per user. The strongest evidence in this section was given under oath or stated to investors.
Meta deliberately increased the average ad load on Facebook and Instagram to meet its revenue projections.
Pattern three · Preference Amnesia — supported, primary
The choice is real. It just doesn't stick.
Preference Amnesia — a product that offers a real choice, honors it, and then quietly reverts to the configuration the company prefers — entered the CHI Lexicon through the Facebook investigation, with a strict test: the preference must exist, the customer must express it, the product must override or reset it, and the forgotten preference must systematically benefit the company. Instagram supplies the cleanest corroborating set this index has documented — three independent exhibits, each meeting every limb of the test.
Following exists as first-class UI. The customer selects it; the feed re-sorts exactly as asked; nothing is broken. Then the session ends, and the next open returns to algorithmic Home. It has never been possible to set Following as the default. Every forgotten selection returns the customer to the surface where recommendations and ads live. Established
Plainly: you told Instagram what feed you wanted. Instagram remembers until you close the app.The only control over Suggested Posts in Home is a snooze that lapses after 30 days — indefinitely, by design. This is not a setting that fails to save; it is a setting engineered to un-save itself, every month, forever. There is no permanent version at any price in any market. Established
Plainly: "no" is accepted — with an expiration date the customer didn't set.In early 2024 Instagram limited recommendations of political content by default for all customers, announced in what reporting called "a little-noticed February blog post", reversible only through a deeply buried setting. In January 2025 the limit was reversed — political content phased back in — again without asking. Neither the imposition nor the removal was initiated by any customer; both matched Meta's corporate posture of the moment. Established
Plainly: the setting had two owners, and neither of them was you.Facebook remains the type specimen — seventeen years of a chronological view that could never be made the default. Instagram is recorded as corroborating the pattern in stronger form: the resets are faster (every session), the expiry is explicit (30 days, on a timer), and one preference was flipped in both directions by the company itself. All three forgotten preferences resolve in favor of the surfaces where Meta's inventory lives. Open the full Lexicon entry →
One qualification, kept deliberately. Instagram has stated a reason for the session reset — internal research showing users are "more satisfied" with ranked feeds — and the 2023 Science experiment gives that claim independent partial support. What no research finding explains is why satisfaction with the default requires forbidding persistence of the alternative. See the divergence section below for the full argument. Analytical inference
Pattern four · Engagement Harvesting — supported, with caveats carried
The wellbeing toolkit, and what happened to it.
An engaging product is not hostile. The CHI-relevant showing is architecture optimized for engagement against the customer's expressed interest — and Instagram's record contains a documented arc in which tools shipped as wellbeing gestures were converted into engagement surfaces or quietly weakened when growth stalled.
A genuine stopping cue, launched in the time-well-spent era. Instagram's stated goal: time on the app should be "intentional, positive and inspiring".
The stopping cue becomes a doorway: past "all caught up" now lies an infinite recommendation stream. The same day, an Instagram executive said the goal was to "make it clear when you're all caught up so you can decide how you want to best use your time." Company statement
Launched days before Adam Mosseri's Senate testimony. Real, and inert by default.
In the quarter after Meta's first-ever daily-user decline, the self-limit tool's floor was tripled, with notifications nudging existing users to raise their limits. Meta's explanation: "We changed the 'daily limit' options to avoid sending people multiple notifications at the same time." Company statement The timing is documented; the motive is an inference.
The European Commission preliminarily found that "the addictive design of Instagram and Facebook" — infinite scroll, autoplay, push notifications, engagement-oriented recommender systems — breaches the Digital Services Act's risk-mitigation duties, with required remediation including default-off autoplay and infinite scroll. This is a preliminary finding, not a final decision. Meta disputes it: "We disagree with these preliminary findings." Exposure runs to 6% of global turnover if confirmed. Preliminary regulatory finding · pending
Teen safety · Three strata, kept separate
What Meta knew, what science shows, and what courts have found are three different questions.
This is the section of the Instagram record most often flattened into a slogan, in both directions. CHI's approach is to keep the three evidence strata separate, because they are not equivalent and do not substitute for one another. "Meta's internal research found X" and "science proves Instagram causes X" are different propositions, and this page never merges them.
Stratum 1 · Internal research
Real documents · not causal science · interpretation disputed
Leaked 2021 slides include "We make body image issues worse for one in three teen girls." The slides are authentic and entered the congressional record. They are survey research, not causal measurement — and Meta's rebuttal is on the record too: the slide "relied on input from only 40 teens", and on 11 of 12 well-being issues, teen girls who struggled said Instagram made things better rather than worse. Both the slides and the rebuttal are documented; the interpretation is contested. Leaked internal documents · disputed
Stratum 2 · Independent science
Genuinely contested field · modest average effects
The strongest causal work (a 2022 American Economic Review study of Facebook's staggered college rollout) found meaningful increases in depression and anxiety, concentrated in social-comparison-prone students; other prominent research finds digital-technology use explains a trivial share of adolescent well-being variance. Deactivation experiments show modest average effects, larger for vulnerable subgroups. Neither "Instagram is safe" nor "Instagram causes the teen mental-health crisis" is scientifically established, and public-health advisories say the evidence is insufficient to conclude social media is sufficiently safe. Peer-reviewed, contested
Stratum 3 · Courts and regulators
Moving in real time · nothing final
A New Mexico jury found Meta liable in March 2026 ($375M), and in August 2026 the court added a $567M abatement fund — $942M total; Meta says it will appeal, and the verdict is at trial level, not final. A federal trial brought by four state attorneys general — California, Colorado, Kentucky and New Jersey, part of a broader 29-state effort — opened in Oakland this week: jury selection began 12 August 2026 and opening statements are scheduled for 18 August. This trial presents core platform-harm claims from that smaller group alongside broader children's-data claims; the wider 29-state coalition and multidistrict litigation extend beyond it. As of this page's build date, nothing has been decided. Independent testing of 47 advertised teen-safety features rated 30 substantially ineffective; Meta calls the report "misleading" and "dangerously speculative". Trial-level verdict on appeal · trial pending
Credited in full, alongside all of the above: Teen Accounts (September 2024) made minors' accounts private by default with messaging limits, sleep mode and parental supervision — genuinely protective defaults, whatever pressures produced them — and the safety toolkit below is one of the largest in social media. What this section evidences is business incentive and design architecture under active legal test, not a settled scientific harm claim. It is scored, cautiously, within Behavioral Manipulation, with the contested science treated as a ceiling on the score rather than a floor.
Pattern five · Data & Tracking Creep — supported
Fourteen years of integration. The brakes came off at the end.
A distinction first: operating a service, securing accounts, and personalizing a feed all process data, and none of that is hostility. This section tracks something narrower — the movement of Instagram's data practice away from anything the customer chose, and the removal of the controls that pushed back. Collapsing every form of processing into "surveillance" would be the easy mistake here, and this page does not make it.
Pattern six · Choice Fragmentation — supported
Many controls. One architecture. The architecture is not on the menu.
Instagram's settings count has grown every year, and most individual controls work exactly as advertised. The finding is the shape of the whole: granular controls proliferate while the four defaults that carry the business model — algorithmic Home, recommendations on, personalized ads on, data flowing — have never once been customer-settable, anywhere.
- View a chronological Following feed Per session — resets on every app open.
- Pick up to 50 Favorites Chronological, and boosted in Home.
- Snooze Suggested Posts For a maximum of 30 days at a time.
- Reset the recommendation model Which immediately begins re-learning.
- Mark "Not interested", hide ads per advertiser and topic
- Use Hidden Words, Restrict, Limits, Close Friends, comment controls A genuinely strong anti-harassment stack.
- Set per-type notification toggles
- Download their data; delete their account in-app In-app deletion since 2022 — required by Apple's App Store rules.
- EU only: refuse profiling on some surfaces; pay for ad-free; keep Facebook and Instagram data separate
- Set any non-algorithmic feed as the default
- Permanently remove recommendations from Home
- Cap or reduce advertising volume
- Opt out of behavioral advertising in the US
- Stop the app opening into an algorithmic surface
- Remove Meta AI from search and DMs
- Sever off-platform activity from their profile Since the 2026 disconnect-tool removal (EU excluded). Reported
- Opt out of AI training on public posts (US), or of AI chats informing ads
- Reach a human being without paying — and often when paying. See below.
The controls are also scattered — split between Instagram settings and Meta's Accounts Center, relocated by three navigation redesigns since 2020 — and the European Commission's October 2025 preliminary finding says even the reporting and appeal flows use "dark patterns": "unnecessary steps", designs that are "confusing and dissuading", and appeals that don't let customers submit evidence. That finding is contested and not final. Preliminary regulatory finding · pending Scored under Customer Restriction and Trust & Transparency.
Support & Meta Verified · The recourse vacuum
Three billion customers. The human is behind a paywall — and often not even there.
The question this section answers carefully: did Instagram monetize solutions to problems that arise inside its own ecosystem — account takeover, impersonation, wrongful bans, the inability to reach anyone? The sequence is documented. Intent is not, and is not asserted.
No phone or email route to a person has existed at any point in Instagram's history. In 2021, reporting documented locked-out users buying $299 Oculus headsets because Meta's hardware division had human support that could escalate account recovery. Customer reports, verified phenomenon
Its advertised benefits: identity verification, "proactive account monitoring" against impersonation, and "access to a real person for common account issues". Impersonation and takeover are platform-native harms; human access had never been free. Company announcement This page does not claim support was degraded in order to sell it — no evidence establishes that.
Account-takeover complaints in New York rose from 73 (2019) to 783 (2023) — up roughly 1,000% — with 128 in January 2024 alone. The AGs called Meta's failure to help "unacceptable" and linked the surge to layoffs concentrated in security and integrity teams. A letter, not an enforcement action. Regulator correspondence
From mid-2025, waves of customers were wrongly suspended — some flagged for child sexual exploitation, the most stigmatizing accusation a platform can make — and met AI-driven appeal loops with no human contact. Hundreds contacted the BBC; reinstatements clustered after journalists asked. Reporting found human review effectively gated behind Meta Verified — and paying subscribers describing canned responses and months-long loops. Meta acknowledged only a "technical error" scoped to Facebook Groups and has published no root cause for the Instagram flags. Investigative reporting
December 2025, launching a support hub with an AI assistant: "We also recognize that support hasn't always met expectations." February 2026: a Korea-based customer-service center, conceded under regulatory pressure after the ban wave — an acknowledgment that no human channel existed. Human support for free customers is still not promised. Company statements
Regulatory & litigation record · Status as of 14 August 2026
The legal record, with the labels kept on.
Instagram's 2025–2026 legal environment is moving quickly, and the difference between a preliminary finding, a settlement, a trial-level verdict and a final ruling is the difference between an accusation and a fact. Each status below is exact, and none of the pending matters is treated anywhere on this page as decided.
Settlements resolved without admission are identified as such every time they appear on this page; preliminary findings are never described as violations; and the two matters most likely to change this assessment — the EC's preliminary addictive-design finding and the Oakland trial — are flagged as open in the verdict below.
Instagram's case for Instagram
Why CVI is 76.
Assessed independently of the hostility investigation, and given real weight: Instagram's counter-ledger is stronger than most companies' in this index. If this page only convinces you that Instagram extracts, it has failed.
Sixteen years with no mandatory monetary price, at three billion monthly users — photo and video sharing, hosting, messaging and discovery delivered identically to the poorest and richest customers on earth. Fact
Measured satisfaction is above the category, not below it: ACSI 76/100 in both 2024 and 2025 — above the social-media average of 74 and well above Facebook's 70. Whatever this page documents, customers rate the product highly, and the score rose after the 2023 photo rebalancing and control features shipped. Independent survey index
The best causal evidence partially supports Meta's core ranking claim. In the 2023 Science election study, users switched to chronological feeds saw roughly 22% more untrustworthy-source content on Instagram, with no measurable improvement in well-being or polarization. See the divergence section below. Peer-reviewed
Stories, DMs with video calling, Close Friends, Collabs, Notes, Broadcast Channels — a decade of genuinely valuable free capability, much of it now the product's most-used surface.
One of the largest anti-harassment stacks in social media: Restrict (2019), Hidden Words (2021), Limits, comment controls and filters, Mute, pinned comments — tools that work as advertised and persist once set. Fact
Real democratizations: link stickers opened to all customers in October 2021, removing the 10,000-follower gate; like-count visibility became the customer's choice in 2021; auto alt-text (2018), captions and translation serve accessibility at scale.
Teen Accounts (2024): private by default, messaging limits, sleep mode, parental supervision — later defaulted to PG-13-level content. Credited in full; effectiveness disputes are documented separately above. Plus nudity protection defaults, sextortion campaigns, and the Take It Down partnership. Fact
Small-business and creator infrastructure with no listing fee: storefronts, product tags, branded-content tooling, and distribution that built real businesses — partially offset by the 2022–23 commerce and bonus whipsaw, which is why this item is not scored higher.
A documented record of responding to revolt: the 2022 full-screen rollback, the 2023 "overfocused on video" rebalancing, separate Threads deletion (2023), the grid-reorder concession (2025), and the Muse AI tool withdrawn within 72 hours (2026). The company listens, publicly and repeatedly; the responsiveness is credited here, while the resumption that typically follows is documented above. Fact
Deductions applied, and why CVI is not higher: the support vacuum and 2025 wrongful-ban wave sit inside Trust, Safety & Reliability (9/15); E2EE was removed from DMs in 2026 while Messenger kept it; and recommendation-driven commercial saturation degrades the core experience the product is scored on.
CVI 76/100 against CHI 70/100 produces CFS +6 — the finely balanced band, and the narrowest gap in the index. Overall CVI confidence: B+.
The divergence · Where this assessment is genuinely hard
The ranking may work. The customer still can't choose.
Instagram is not Facebook with different screenshots. Its distinctive tension is that the recommendation system may genuinely serve customers better content while genuinely removing their control over what the product shows them — and the evidence for both halves is strong. A page that resolved this tension in either direction would be overclaiming. It is presented here unresolved, because it is the finding.
The case that the ranking works
Content quality
The 2023 Science experiment: chronological feeds increased untrustworthy-source exposure on Instagram by ~22% and improved nothing measurable about well-being or polarization. Peer-reviewed, with data access Meta did not get to veto.
ACSI satisfaction rose to 76 and stayed there — above the category average. Time-spent growth is consistent with customers finding recommended content genuinely engaging, not merely tolerating it.
Meta's claim that ranked feeds serve users better than raw chronology is, on the best available evidence, partially vindicated.
The case that control was taken
Customer agency
No persistent non-algorithmic default has existed since 2016, anywhere. The Following feed resets every session; the snooze expires monthly by design; the reset cannot turn the system off; and by 2025 the app began opening directly into the pure recommendation surface.
Every durable control that exists — chronological return, non-profiled EU feeds, ad-free EU option, in-app deletion — traces to a Senate commitment, a statute, an app-store rule, or a revolt. None arrived as voluntary product judgment.
The transfer of control is fully established, on Meta's own numbers.
Instagram wins the argument it refuses to let customers settle for themselves. Meta holds real evidence — internal testing, and now independent science — that ranked, recommendation-heavy feeds outperform chronology on engagement and on some quality measures. A company confident in that superiority could afford a persistent off switch and let the results speak. Instead, for ten years, the customer's feed and recommendation preferences have been engineered to expire rather than persist — while Meta told investors, precisely and quarterly, what the defaults are worth in engagement and inventory. Customer control and content quality are not the same question. Instagram's history answers the second and forecloses the first — and that choice, not the algorithm, is what this page scores. Analytical inference, built on established facts
Chronology · The load-bearing dates
Fourteen years in eighteen rows.
Associated patterns · Final state
The patterns materially relevant to this verdict.
Primary
Supported
Qualified and partial
Falsification · What the evidence does NOT support
The investigation killed more claims than it confirmed.
Instagram may be the most folklore-encrusted product in this index. Each claim below was investigated and failed — and publishing the failures is the price of being believed about the findings above.
No capability has ever been demonstrated; the one commercial "active listening" pitch that circulated collapsed under scrutiny and was retracted. The accurate kernel is less cinematic: non-audio tracking is powerful enough to produce the eavesdropping feeling — which is an argument about the data perimeter above, not about microphones.
No fixed percentage cap exists, and Instagram has denied one on the record. Median reach per follower genuinely sits well below 100% under ranking — the experience is real; the number is invented. Note this is a different claim from the 7% time-spent figure above, which is Meta's own and means something else entirely.
Undisclosed suppression states did exist: Account Status (2022) revealed "ineligible for recommendation" flags, and 2024's political-content limit was silent category-wide demotion. The mechanism was real and undisclosed; the folk theory of secret individualized bans at scale remains unevidenced.
None of these rules appears in Instagram's published ranking explainers, no penalty mechanism is on record, and several have been debunked by Instagram directly. Early engagement is a real signal; the precise formulas circulating in creator folklore are invented precision.
Instagram denies ranking differences by account type, and the investigation located no evidence of any. Recorded here so the absence is on the record.
No advertising exists inside Instagram DMs or broadcast channels as of this build. Given the fourteen-year surface progression documented above, the claim is understandable as a prediction. It is not a fact, and this page does not treat directionality as destiny.
Evidence & methodology
- Company disclosureMeta Q1 2024 earnings call transcriptSource of "For the first time ever, more than 50% of the content people see on Instagram is now AI recommended." Verified verbatim against the transcript for this page.
- Court recordFTC v. Meta — Memorandum of Opinion, 18 November 2025Primary source for the ~7% friend-content time share, the finding that Meta "individually calibrate[s] each user's ad load", and the court's account of the shift to AI-recommended content. Judgment for Meta; under FTC appeal, not final.
- Company disclosureMeta Q4 2025 earnings call transcriptSource of the ad "redistribut[ion] across users and sessions" and "grow the overall level of ad load" statements, shared with the Facebook assessment.
- RegulatoryEuropean Commission — preliminary finding: the addictive design of Instagram and Facebook breaches the DSA, 10 July 2026Preliminary and contested; named mechanics include infinite scroll, autoplay, push notifications and engagement-based recommenders. URL re-verified for this build.
- RegulatoryEuropean Commission — preliminary DSA findings on Meta and TikTok (dark patterns in reporting/appeal flows; researcher access), 24 October 2025Preliminary and contested; not a final decision.
- RegulatoryEuropean Commission — €200M DMA non-compliance fine on Meta's pay-or-consent model, 23 April 2025Adjudicated at Commission level; under appeal.
- RegulatoryIrish Data Protection Commission — decision in Instagram inquiry (€405M), September 2022Default-settings findings concerning minors' data; under appeal.
- Regulator correspondenceMultistate letter to Meta on account takeovers — 41 attorneys general, 5 March 2024Primary source for the complaint-volume figures and the "unacceptable" characterization. A letter, not an enforcement action.
- Peer-reviewedGuess et al. — How do social media feed algorithms affect attitudes and behavior in an election campaign? Science, 2023The chronological-feed experiment: less time spent, more untrustworthy content on Instagram, no measurable well-being or polarization change. Used at full weight as counterevidence.
- Company disclosureInstagram Ranking Explained — Instagram, May 2023Instagram's own description of per-surface ranking, including popularity among top-tier Feed signals — read against the 2016 "not what we're optimizing for" statement.
- Company disclosureTesting Meta Verified — Meta newsroom, February 2023Launch language for the paid tier, including "access to a real person for common account issues".
- Original journalismInstagram confirms test of "unskippable" ads — TechCrunch, 3 June 2024The ad-break test with a countdown timer that blocks scrolling; later formalized in the EU "less personalized ads" tier per Meta's own November 2024 announcement.
- Trial reportingMeta, 29 states head to court in biggest test yet of youth social media litigation — Reuters via Claims Journal, 12 August 2026Status of the Oakland trial as of this build: jury selection began 12 August; openings scheduled 18 August; nothing decided. The trial itself is a federal case brought by four states (CA, CO, KY, NJ) within the broader 29-state effort. Re-verified on build day.
- Court reportingNew Mexico court orders Meta to pay nearly $1 billion — Forbes, 7 August 2026The $375M jury verdict plus $567M abatement fund ($942M total); Meta: "We disagree with the ruling and will appeal." Trial-level, not final.
This page is a synthesis of a completed evidence dossier covering six parallel research streams with cross-stream consistency checks and a 52-entry evidence ledger, each stream instructed to attempt falsification as well as confirmation.
Evidentiary status is preserved throughout and never upgraded for readability: filed litigation material is never presented as a finding, settlements without admission are labeled every time they appear, and preliminary EC findings are labeled preliminary at every mention alongside Meta's quoted disputes. The two Meta-sourced headline statistics keep their denominators — share of content seen (earnings call) and share of time spent (court record) — and are never combined into a "share of feed" claim.
Independently re-verified on this build's date (14 August 2026): the Oakland trial's status and jury-selection date; the New Mexico totals ($375M jury verdict, $567M abatement fund added in August, appeal announced — correcting the dossier's earlier bench-award date); the July 2026 and October 2025 EC press releases; and the Q1 2024 "more than 50%" quotation, checked verbatim against Meta's transcript.
Omitted rather than softened: any Instagram-specific ad-load percentage series, Meta Verified subscriber counts, the unverified "race to the bottom" quote, and per-category notification defaults. The July 2026 disconnect-tool removal rests on technology-press reporting and is labeled as such throughout. CHI 70/100, CVI 76/100 and CFS +6 are settled values under Methodology v2.0, calibrated against Facebook; they are not a range.
Final verdict
CONCERNING — THE FEED CHANGED OWNERS
Instagram never charged its customers a cent. Instead, over fourteen documented years, it repossessed the feed. The product that promised photos from people you follow became a system in which — by Meta's own numbers — the majority of what you see is chosen by the machine, friends now account for roughly 7% of time spent viewing Instagram content, and advertising load is tuned per customer to what you'll tolerate. The feed choice that would let you refuse the machine resets every time you close the app, and the only control over recommendations lapses on a 30-day timer. None of this was hidden; nearly all of it was announced to investors as it happened. It was simply never offered to customers as a persistent choice.
What customers got back is real, and this page credits it in full: a free, polished, above-average-rated product at three billion users, a genuinely strong safety toolkit that largely does persist once set, a documented record of retreating under revolt — and independent science suggesting the ranked feed may serve better content than the chronology it replaced. That last point is why this assessment is not a prosecution brief. The issue is not algorithms, or advertising, or any single change. It is the accumulation: algorithmic replacement, recommendation growth, advertising expansion, engagement optimization, data integration — and feed and recommendation controls engineered not to persist. Control flowed overwhelmingly one way; some of it returned after backlash, while the most durable structural protections — the chronological option, the EU's non-profiled feeds, ad-free access, in-app deletion — arrived only under regulatory, statutory or platform-rule pressure.
CFS +6 is the narrowest gap in the index — narrower than Facebook's. The open matters flagged above, particularly the European Commission's preliminary addictive-design finding and the Oakland trial, could move this assessment in either direction within months. The evidence will decide; that is the method.
You still choose who to follow.
When did you last choose what you saw?