Technology
Amazon
Amazon's core retail relationship is the strongest value case in this index. The hostility lives one layer beneath it.
Two things are true about Amazon at the same time, and this page exists because neither one cancels the other. On the surface — the part almost every customer actually touches — the exchange has improved for twenty-one years: an independent study finds Amazon the lowest-priced major US online retailer for the ninth consecutive year, the two-day promise of 2005 is a same-day norm in 2026, returns are close to frictionless, and the membership fee has tracked inflation. Independently measured One layer down — inside the Prime media bundle, inside devices customers bought outright, inside digital libraries they paid for title by title — Amazon has spent 2022 to 2026 changing what an already-paid price buys, and charging to restore parts of what it used to include. Amazon-primary The original exchange was: pay once and the friction disappears. The later exchange, in seven product surfaces, is: keep paying to keep what you had.
CFS = CVI − CHI (88 − 67 = +21), the top of the band the methodology calls Fair — which is explicitly not customer-friendly territory. Amazon ties Google / Alphabet exactly at CHI 67; its CVI 88 is behind only Canva and Zoom, both at 89. Scores are settled analytical values under Methodology v2.0 and are not presented as a range. Overall CHI confidence: High on the aggregate and on four of five dimensions, Moderate-High on Customer Restriction, where the 14 August 2026 arbitration term was five days old at scoring and its practical effect is unmeasured. Overall CVI confidence: High. Methodology →
The defining finding
Ring exempted the customers who were already locked in — and that is the tell.
When Ring moved Home/Away modes and Alarm app arming behind a subscription on 29 March 2023, it applied the change to new accounts and grandfathered existing Alarm owners: “If you did not own an Alarm, these features will now require a subscription.” Company statement A company that exempts its embedded users from a new term is disclosing its own judgment about how that term would land with people who had already wired the hardware into their homes. Analytical inference
CHI reads this as evidence of dependency sensitivity, not of malicious intent. It is a documentary tell, and it is also a genuine mitigation — which is why Ring did not push Revenue Extraction higher.
The core contradiction · Read this before the evidence
Two layers. One company. Only one of them is the finding.
Almost every argument about Amazon goes wrong in the same place: it treats “Amazon” as one relationship. It is at least two. The layer customers touch daily has improved, measurably, for two decades. The layer underneath it — the media tiers, the devices, the libraries — is where every core exhibit on this page lives. A reader who loves Amazon's delivery will reject a CHI of 67 unless this distinction is made first; a reader who dislikes Amazon will over-read it for the same reason.
Layer one — the core retail relationship
No CHI finding · Carries CVI 88Buy something, have it arrive fast, send it back if it is wrong, at a price independently measured as the lowest among major US online retailers. This layer passed the Dependency-Principle test outright: Amazon would plausibly offer a brand-new customer today's terms today, because they are better than the terms it offered at launch. Customers who stay because Amazon remains better are scored as normal retention, not as captives.
- Price — ~14% advantage, nine straight years
- Speed — 13B same/next-day items in 2025
- Selection — 300M+ Prime-eligible items
- Returns — 10,000+ box-free drop-offs
- Satisfaction — ACSI 82–84, tied first
One layer down
Layer two — the bundle, the hardware, the library
Every core exhibit · Carries CHI 67Prime media benefits, purchased devices and purchased digital content. Here the chronology runs the other way: the customer commits, switching costs accumulate, and then Amazon changes the exchange. All six exhibits below sit in this layer. None of them sits in layer one — and that is the single most important fact about this assessment.
The qualifier this page carries everywhere: Amazon exploits dependency in the media tiers, devices and digital libraries, 2022–2026. It is not accurate to say “Amazon exploits dependency” unqualified — the retail core is the counter-example, and it is the part most customers actually experience. Scoring decision
Primary exhibit · Advertising Creep, type specimen
Thirteen years ad-free. Then ads, at the same price. Then a more expensive way out.
This is the defining Amazon exhibit and the reason Revenue Extraction sits at 17/25. Read it as a sequence rather than a list: at no point did the headline price of Prime go up. What changed was what the headline price contained — and then what it cost to get the old version back. All five steps are Amazon-primary or independently measured
Streaming enters the membership as a benefit of the shipping subscription. It stays ad-free for thirteen years, while Prime grows past 200 million members worldwide.
Announced 22 September 2023 (“limited ads… We're not making changes in 2024 to the current price of Prime membership”), live in the US on 29 January 2024. Ad-free viewing becomes an additional $2.99/month. The migration email to existing members was headed “No action is needed” — wording a Munich court held misleading in December 2025 (under appeal). Amazon announcement
Not announced. The tier customers were moved into by default quietly lost picture and sound capability it had the month before. Strongly supported
Reported by Adweek in June 2025 as a move from approximately 2–3.5 minutes per hour to approximately 4–6 minutes per hour. Not denied by Amazon. Amazon's ad load remains mid-pack against other major streamers. Trade reporting, not denied
With the launch of Prime Video Ultra on 10 April 2026, the ad-free surcharge goes $2.99 → $4.99 (+67% in 27 months) and 4K resolution is relocated out of the base tier. Amazon's stated rationale — that this “aligns with other major streaming services” — is an ex post comparison: no streaming service has ever launched by charging for the ad-free version of a benefit it already included. Amazon announcement
That four-step mechanism — not the ads on their own — is what CHI is scoring. It is the cleanest instance in the corpus of advertising inserted into a paid, ad-free benefit with the paid exit subsequently repriced upward, which is why Amazon is recorded as the type specimen for it. Patterns: Advertising Creep (defining) · Access Downgrading (supporting, for the Dolby and 4K removals).
Counterevidence — carried at full weight
- The Prime fee did not move. $139/year through August 2026; monthly flat since 2022.
- Content investment rose sharply. Video and music spend went from $13.0B (2021) to $22.4B (2025), with TNF, NBA and NASCAR rights acquired and no Prime fee change.
- The US class action was dismissed with prejudice in July 2025, the court treating the change as “a change in subscription benefits as opposed to a price increase.”
- Dolby Vision was restored to the base tier in April 2026 — a genuine, if partial, reversal. 4K remains gated.
- German litigation is split and non-final. A Munich court held in December 2025 that introducing ads without consent was unlawful (under appeal); a Bavarian appellate court rejected a consumer-group damages claim in July 2026.
- An exit exists at every step. A paid exit is a poor exit and is scored as one — but it is not the absence of one.
Prime Video also carries its own product-level CHI assessment on this site, scored on the streaming-category comparison set rather than at company level: Prime Video — CHI 58 →
Primary exhibit · Adjudicated · Fixed under pressure
The strongest single item in this file — and it no longer exists.
The “Iliad” flow
Amazon's internal name for its Prime cancellation path was Iliad. Cancelling ran to four pages, six clicks and fifteen options. Enrollment pages took billing information before disclosing the terms of the subscription, and the link declining Prime read: “No thanks, I do not want fast, free shipping.” Court record
On 18 September 2025 a federal judge ruled on partial summary judgment that Amazon's Prime enrollment violated the Restore Online Shoppers' Confidence Act as a matter of law on the disclosure-before-billing issue — “no reasonable jury could find in favor of Amazon.” Days into trial, on 25 September 2025, Amazon settled the FTC's enrollment and cancellation case for $2.5 billion: a $1 billion civil penalty and $1.5 billion in refunds to roughly 35 million customers, plus a court-ordered redesign with an external monitor. Adjudicated in part; settled without admission
An actual judicial finding, a named internal flow, and personally bound executives. Nothing else in the CHI corpus has all three.
Fixed under pressure
CHI's methodology is explicit here, and it cuts both ways. The current experience gets full credit regardless of why it changed. The US flow was redesigned in April 2023; the EU was fixed with a two-click cancellation in July 2022, before any order existed, which is a genuine voluntary-repair data point. Since September 2025 a stipulated order mandates a clear decline button, cancellation as easy as sign-up, and an external monitor. Prime is easier to cancel today than at any point in its history, and this page credits that without discount. Current state
And compulsion does not retroactively erase the conduct. The pressure — the FTC suit, the ROSCA holding, the $2.5B order — is recorded as context, not charged as an extra penalty. Amazon settled without admitting wrongdoing and maintains it has “always followed the law”; neither that statement nor the compulsion adds to the score.
Net effect: Behavioral Manipulation sits at 15/25 rather than the 18–19 the historical conduct alone would support.
No reader should leave this page believing the Iliad flow still exists. It does not. The $2.5 billion order and the ROSCA holding are not softened either — both are on the record above, in the words the court and the FTC used. Patterns: Exit Resistance (host pattern for the cancellation architecture) · Choice Illusion (the fifteen-option enrollment path and the decline link) · Intellectual Disrespect (narrow — the decline-link wording only).
Primary exhibit · Paywall Creep
The catalogue got fifty times bigger. The thing you could do with it got smaller.
What was announced
On 1 November 2022 Amazon expanded the Prime Music catalogue from roughly 2 million songs to 100 million. The announcement is accurate and the expansion is real. Amazon press release
“Prime members can shuffle play any artist, album, or playlist.”
What also happened, in the same sentence
General on-demand playback was removed. The included benefit became largely shuffle-based, with on-demand play relocated to Music Unlimited. Amazon told Fox Business in January 2023 that it had chosen the larger catalogue “even if it meant a primarily shuffle-mode experience.” Company statement
Catalogue expansion and functional narrowing, delivered as one announcement.
| Year | 2022 | 2023 | 2024 | 2025 | 2026 | Change |
|---|---|---|---|---|---|---|
| Prime-member monthly price | $7.99 | $8.99 | $9.99 | $10.99 | $11.99 | +50% |
Scroll table →
The distinction the page is drawing is between a bigger catalogue and a narrower benefit. Both are true; only the second is a CHI finding. The scored harm is that a capability included in a paid membership was moved to a paid upgrade whose price then rose by half in four years. Pattern: Paywall Creep. The framing — a removal announced as an expansion — is recorded as Intellectual Disrespect in its narrow sense only.
Counterevidence — carried at full weight
- The catalogue really is 50× larger, and for many listeners that is the more valuable trade.
- All-Access playlists returned to on-demand in October 2023 — a partial walk-back.
- Ad-free podcasts and HD audio remain included at no extra charge; HD has been free since 2021.
- No advertising has been added to Prime Music in the US. Ads appeared in India only, in June 2026, and this page does not extend that to other markets.
Primary exhibit · Paywall Creep · Rentalization
The doorbell you bought. The modes you now rent.
Ring is the hardware case: functions of a device the customer owns outright were moved behind an active subscription. It is also where this page has to be most careful with arithmetic, because the two available baselines give very different numbers and only one of them is usually quoted.
| Date | Price | New capability added? | From 2015 baseline | From Jul 2022 baseline |
|---|---|---|---|---|
| 2015 (baseline) | $3.00 | — | — | — |
| July 2022 | $3.99 | Yes — 180-day event history | +33% | baseline |
| March 2024 | $4.99 | No | +66% | +25% |
Scroll table →
Stating the baseline, because the common version of this claim is wrong.
Ring Basic is +66% from the 2015 baseline of $3.00, and +25% from the July 2022 price of $3.99. A “66% since 2022” framing appears widely. It is arithmetically wrong, it mixes the two baselines, and this page does not use it. Two of the three rises added no new capability; the July 2022 rise did.
What moved behind the subscription — 29 March 2023
Home and Away modes on cameras, and arming, disarming and notifications from the Alarm app, now require an active subscription for accounts created after that date. Ring's support documentation states plainly: “Compatible Ring subscription required for digital arming/disarming.” Company documentation
Free event history was cut from 60 days to 24 hours, and the plan structure has been reorganised four times between 2021 and 2026.
These are operating controls for hardware the customer already owns, moved into a recurring fee.
The grandfather tell — and why it cuts both ways
Ring exempted existing Alarm owners: “If you did not own an Alarm, these features will now require a subscription.” Company statement, 29 Mar 2023
Read one way, that is evidence of dependency sensitivity: a company exempting the users most embedded in its hardware is disclosing its own view of how the term would survive their scrutiny. Read the other way, it is a real mitigation — the people with the highest switching costs were not charged. Both readings are carried. This page classifies it as exploitation of future dependency, mitigated by the exemption, and presents it as evidence of dependency sensitivity rather than proof of malicious intent. Analytical inference
The mitigation is the reason Ring did not push Revenue Extraction to 18.
Patterns: Paywall Creep · Rentalization · Price Creep, recorded strictly at product level. Price Creep is not a company-wide finding for Amazon and is contraindicated for the Prime fee itself, which has tracked inflation over twenty-one years.
Counterevidence — carried at full weight
- Cloud recording was always paid. The widely repeated claim that Ring moved previously free recording behind a paywall is rejected by this assessment; the real finding is the price path and the 2023 operational paywalls.
- Live View, motion alerts, two-way talk and Alexa integration remain free on every Ring device.
- Existing Alarm owners were grandfathered and pay nothing extra for arming and modes.
- Blink, also Amazon-owned, retains free local storage options, so the household has a first-party non-subscription path.
Primary exhibit · Summary only — full investigation cross-linked
“No plans to add advertising to Alexa.”
Alexa has its own dedicated CHI investigation on this site and is not reproduced here. What follows is the Amazon-level summary: the five movements that bear on the company score, in order. Read the full Amazon Alexa assessment →
Made while Echo hardware was being sold at or near cost. Company statement
Smoke-alarm and glass-break detection that had been a free Alexa feature moved into a subscription: $5.99 at launch, now $7.99 list. Confirmed
“Do Not Send Voice Recordings” was withdrawn by email on 28 March 2025: “we have decided to no longer support this feature.” This is the cleanest post-purchase privacy downgrade in the record — an affordance some customers bought the device partly to have, removed with no compensation and no substitute. Amazon email
The opt-out works — “Alexa, exit Alexa+” returns the device to classic Alexa, and the setting is account-wide. Recorded as supporting evidence, not as a core finding, for exactly that reason. Strongly supported
Sponsored Tiles appear in April 2026; Amazon publishes “Agentic Ads” for Alexa on advertising.amazon.com on 23 June 2026. Andy Jassy had described advertising on 31 July 2025 as “a lever to drive revenue.” Eight years after “no plans,” Alexa carries two advertising products. Amazon-primary
Patterns: Promise Reversal (the 2018 statement) · Advertising Creep (Sponsored Tiles, Agentic Ads) · Paywall Creep (Guard → Emergency Assist) · Access Downgrading (local processing removed). The privacy items are also scored under Information & Privacy 10/15; CHI's Lexicon has no “Data/Tracking Creep” entry, so no such label is asserted here.
Counterevidence — carried at full weight
- Alexa+ is included free for Prime members (February 2026) against $19.99 standalone, after years of documented device losses.
- Classic Alexa remains free, and the Alexa+ opt-out is real and account-wide.
- Voice-recording deletion controls arrived in 2019–20, alongside per-feature opt-outs including “by the way.”
- The Flock partnership was cancelled in February 2026 after public objection.
Not claimed: that Amazon silently switched users' default music services. That hypothesis was investigated, rechecked in August 2026, and found to rest on forum reports with no credible corroboration. It is not asserted anywhere on this page and did not contribute to any dimension score.
Primary exhibit · Rentalization
“Buy now.” Licensed, not sold.
The overclaim here is easy and wrong, so this page states the argument precisely before the evidence. Amazon did not secretly convert owned books into rentals. The legal relationship has been a licence for essentially the whole life of the product. What changed is narrower and better documented: the owner-like affordances around that licence narrowed over time, portability decreased, and the disclosure arrived last. The claim this page makes, stated exactly
The same year, Amazon remotely deleted purchased copies of 1984 from devices, apologised — Bezos called it “stupid, thoughtless, and painfully out of line with our principles” — and settled, accepting limits on future deletions. Company statement · settlement
Withdrawn without announcement. A sharing affordance that made a purchased book behave a little more like a physical one simply stopped appearing. Strongly supported
Notified on 12 February 2025 and effective 26 February — the only sanctioned route to a local backup of a purchased book. With DRM preventing transfer, and account closure meaning loss of the library, the exit cost of leaving Kindle is the library. Amazon notice
Added around the “Buy now” transaction roughly seven weeks after California's AB 2426 took effect — law-driven rather than voluntary, and credited here as a genuine current-state improvement all the same. Regulatory
UK Kindle Terms, last updated 30 June 2026: “Kindle Content is licensed, not sold, to you by Amazon.” Sixteen years after “permanent copy,” under a button that still reads “Buy now.” Amazon terms
Patterns: Rentalization (primary — the ownership-to-licence facet, and the withdrawal of owner-like affordances) · Promise Reversal (supporting — the 2009 “permanent copy” language against the current terms).
Counterevidence — carried at full weight
- Family Library, Prime Reading and Kindle Unlimited all widen access rather than narrow it; Kindle Unlimited has raised its price once in nine years.
- Send-to-Kindle added EPUB support in 2022, which is an increase in portability into the platform.
- Library integration via Libby means the device is not a closed Amazon-only reader.
- The 2009 settlement placed limits on remote deletion, and the US disclosure now exists where it did not before.
Control pairing · The most instructive comparison in the record
Amazon already knows how to sell advertising respectfully. It has done it for fifteen years.
This module exists to stop a lazy reading. The finding is not that advertising on Amazon hardware is inherently hostile — Amazon runs one of the most transparent hardware advertising models any company has shipped. The finding is that the same company, on a different device, chose the opposite structure. Holding the two side by side is what makes the contrast analytically useful.
Kindle “Special Offers”
The transparency counter-model. Stable since 2011.
- Disclosed before purchaseThe ad-supported device is sold as an ad-supported device, at a lower price than the ad-free one.
- Removable for a clear, stable fee$20 on Kindle, $15 on Fire tablets — unchanged for roughly fifteen years.
- Ads never appear inside booksThey occupy the lock screen and the home row, not the reading experience the customer paid for.
- The one exception was reversed in daysA non-removable ad configuration on the Fire HD in September 2012 was withdrawn almost immediately: “We're happy to offer customers the choice.”
Fire TV
Purchased hardware, escalating interface advertising.
- Advertising added after purchaseAutoplaying video advertising on the home interface from November 2023; softened in June 2024 after backlash, but not withdrawn.
- The screensaver became an ad surfaceSponsored imagery overriding the customer's own screensaver from July 2024.
- A programmatic redesign in 2026The home interface was rebuilt with more, and more automated, advertising inventory.
- No ad-free tier at any priceThere is no Special-Offers-style buyout. The customer who wants the 2015 experience cannot purchase it.
Why this pairing carries weight in the scoring. Kindle Special Offers is the strongest transparency evidence in Amazon's favour anywhere in this file, and it materially limits Trust & Transparency at 11/15 rather than higher. Fire TV is the counterpart: the same corporate capability, exercised without the price, the disclosure, or the exit. Pattern: Advertising Creep (secondary surface).
Secondary evidence · Deliberately not a hero exhibit
Discovery as an advertising surface — the part that is established, and no more.
Amazon's search page is the subject of the loudest allegations in the current record and the weakest evidence on this page, and those two facts are related. What follows is limited to what is measured or disclosed. The unadjudicated material is listed underneath, labelled as such, and carried no scoring weight.
What this page does not say about Amazon search.
The FTC and 17 states allege that Amazon degraded its own search results with advertising — the internal “defects” characterisation — and separately allege a pricing algorithm referred to as Project Nessie. Nothing in that complaint has been proven. Amazon denies the claims and calls the “defects” characterisation “grossly misleading.” The case is scheduled for trial in March 2027. Neither allegation is treated as fact here, and neither moved any dimension score. Certified consumer class actions (De Coster, Frame-Wilson) allege Amazon's policies raised prices across the web, with trial set for October 2026; Amazon disputes the claims, and seller-fee pass-through is not presented as established consumer harm.
This section is scored at reduced weight for exactly that reason, and it sits below the six primary exhibits rather than among them. If the March 2027 findings support the intent allegations, this becomes a different section; today it is a measured advertising-density finding and nothing more.
Analytical label, not a Lexicon assignment. The behaviour described here — a platform controlling discovery while selling position within it — is what the CHI research calls commercial steering. It is used as plain-language analysis on this page: there is no “Commercial Steering” entry in the CHI Lexicon, and this assessment does not create one to accommodate Amazon. The scored components sit inside Advertising Creep as a secondary surface, at reduced weight.
Chronology · 2009–2026
Seventeen years, in the order they happened.
Both directions are on this timeline. Amber marks legal and regulatory events, red marks changes to an existing customer relationship, and volt marks improvements — including the improvements that arrived under pressure, which CHI credits in full.
On sequence and causation. The arbitration change lands in the same twelve-month window as the $2.5 billion ROSCA settlement and the certification of a very large consumer class. This page shows that sequence chronologically and draws no inference from it. No motive is asserted, and none is evidenced. Stated deliberately
Customer Restriction · Five days old at scoring
Recourse, narrowed — carefully weighed, deliberately not a hero exhibit.
What the term does
Effective 14 August 2026, Amazon's US Conditions of Use and Prime Terms again require individual arbitration and waive class actions. Verified across concordant sources
No opt-out window is reported. Acceptance is by continued use, following a 60-day notice period. It applies to every US customer, which makes it the broadest single item in the frozen evidence set — and a class waiver is a poor fit for small-dollar, mass-scale harm, which is the shape of nearly every item on this page.
Worth approximately +1.5 points on Customer Restriction: 12–13 without it, 14 with it.
What the term preserves
A small-claims carve-out survives. There is a published mass-arbitration batching procedure under JAMS administration rather than a blocking clause. Pending litigation is excluded, so the certified consumer class actions already on foot are untouched.
Mandatory arbitration is also the US consumer-contract norm: Amazon's 2021 removal was the anomaly, and a return to the norm is a smaller move than pioneering a new restriction would be. Amazon says the change “will offer customers a fast, cost-effective way to resolve disputes while still giving them the option of going to small claims court.”
Material, not transformative — and five days old, with its practical effect entirely unmeasured.
Two deliberate omissions.
Batch-size thresholds are not published here. Secondary sources conflict on the numbers, Amazon's own Conditions of Use page could not be retrieved at build time, and a disputed figure would be worse than no figure. The structure is published; the arithmetic is not.
No motive is inferred. This page does not suggest the change was intended to respond to any particular litigation. It is also recorded in Trust & Transparency as the third documented promise reversal in the file, because the May 2021 removal was publicly framed as letting customers sue in court, and that framing is now void.
Pattern: Responsibility Diffusion — narrowing collective recourse compounds the “not seller / not distributor” posture Amazon maintains against a single-storefront interface. Amazon partly buys that diffusion down: the A-to-z Guarantee, including Amazon-funded $1,000 injury and property coverage, is a real remedy that most marketplaces do not offer.
The control evidence · Equal weight, not a footnote
Why CVI is 88.
This is the third-highest customer value score in the index — behind only Canva and Zoom, both at 89 — and above Google's 87. If this page leaves you thinking Amazon is an extraction machine, it has failed — because the two facts doing the most work below are independent third-party measurements, not Amazon's description of itself, and the framework is required to say so at the same volume as everything above.
Price — the single most load-bearing fact in Amazon's favour. Profitero's Price Wars 2025 study, published 18 November 2025 and measuring more than 10,000 products across 23 retailers, found Amazon the lowest-priced major US online retailer for the ninth consecutive year, at roughly a 14% average price advantage. It rebuts the “captive customers pay more” reading directly and quantitatively, and it is independent of Amazon. Independent study
Delivery. 13 billion items delivered same- or next-day globally in 2025, roughly 70% more same-day items in Q4 2025 year on year, about 100 million customers using same-day, 30-minute delivery through Amazon Now in nine countries, and 4,000+ rural communities served. The 2005 promise was two days; the 2026 metro norm is same day. Amazon-reported, consistent across releases
Returns — the explicit control case. More than 10,000 box-free drop-off points (FedEx Office added March 2026), refund at first scan retained (confirmed December 2025), returnless refunds, a holiday window running through 31 January, and a “frequently returned item” badge that discloses against Amazon's own interest. The claim that Amazon's returns have broadly deteriorated was tested and is not supported. Fact
Prime economics. $139 a year after 21 years — roughly tracking inflation — with the monthly price flat since 2022 and no increase through August 2026. Prime Access is $6.99 for customers on qualifying assistance, Prime for Young Adults is $69, RxPass is $5, and Alexa+ is included with Prime against $19.99 standalone. JPMorgan valued the bundle at approximately $1,430 a year (2025 estimate) against that $139 fee. Prices confirmed Analyst estimate
Satisfaction. American Customer Satisfaction Index scores of 82–84 across 2023–2026, tied first among online retailers in 2026 against an industry average of 79 and Walmart's 77 — independent, longitudinal and customer-reported, measured during the years documented above. ACSI also singles out Amazon's ease of navigation and checkout as an industry standard. Independent index
Remediation, which is better than most companies in this index. The 2009 apology and settlement; the EU two-click cancellation fix delivered before any order; Fire TV ad softening under backlash; full refunds on Halo, Glow and Astro; a free Blink Mini replacement plus a year of service for Cloud Cam owners; commingled inventory ended in March 2026; the Flock partnership cancelled in February 2026. Amazon also self-identified its refund-processing errors and paid roughly $570 million back before any settlement. Fact
Guarantees and enforcement. The A-to-z Guarantee including $1,000 injury and property coverage (September 2021), Amazon-funded and beyond legal requirement; the 2016 incentivised-review ban, ahead of regulators; 200–275 million suspected fake reviews blocked per year; more than $1 billion a year in brand protection; binding CMA undertakings; seller identity display since 2020. Fact Blocking volumes self-reported
Capability added, not only withdrawn. Content and benefit investment rose from $13.0B to $22.4B between 2021 and 2025 with TNF, NBA and NASCAR rights acquired and no Prime fee change; Alexa+ shipped as an LLM assistant across an installed device fleet; RxPass at $5 with same-day pharmacy in 3,000+ cities; accessibility provision across Alexa, Fire TV VoiceView and Kindle text-to-speech.
Innovation that changed industries. Prime (2005) became the template for the modern consumer subscription; Kindle created mainstream e-reading; Alexa created the consumer voice-assistant category; same-day logistics at national scale is the largest physical-engineering delivery in this corpus. Innovation scores 9/10 — not 10, because the abandonment record is real and its compensation has been uneven. Scoring judgment
Deductions applied, and why CVI is not higher. Trust / Safety / Reliability sits at 10/15 because of one adjudicated statutory violation, two stipulated FTC privacy orders, the marketplace-responsibility posture, and documented refund-processing failures spanning 2017–2025 that settled for $309.5M. Feature & Capability is held to 22/25 because Amazon removed capability inside the same window — Prime Music on-demand, Kindle lending and USB download, the Alexa local-processing option, 4K from the Prime Video base tier, Prime Household and Invitee sharing. A dimension measuring capability delivered cannot ignore capability withdrawn. Scoring judgment
CVI 88 against CHI 67 produces CFS +21. CVI was not reduced because CHI is high. The dimensions are independent by design, and Amazon is the clearest demonstration in the corpus of why: the only place the hostility evidence legitimately touched CVI is Feature & Capability, where documented capability withdrawal is a direct measure of capability delivered — a fact about features, not a penalty imported from the other score.
Score breakdown
Why CHI is 67 — and why it is not higher.
67 sits in the upper third of this index and exactly level with Google, and it needs a reason that is not “Amazon is enormous.” The reason is that Amazon operates the broadest post-adoption monetization ladder in the corpus — seven consumer surfaces where an already-paid experience was narrowed and the restored version sold back — and that every rung of it is currently in force. Against that, severity is moderate throughout, the core relationship is clean, and the single most severe item has been repaired. Breadth was not permitted to substitute for severity; the Google precedent, where scale-driven scoring would have produced 75+ and was rejected, was applied directly.
CFS = 88 − 67 = +21, the top of the band the methodology calls Fair — alongside Google (+20), Uber (+18) and Apple (+16), and explicitly short of customer-friendly territory, which begins around Spotify's +27. CFS was derived, not chosen: CHI and CVI were each scored dimension-by-dimension against the frozen evidence before the difference was computed, and the result was not adjusted afterwards. A reader who saw only the CHI would think Amazon is among the most hostile companies here; a reader who saw only the CVI would think it is among the best. Both readers would be right, which is the entire reason this framework reports two numbers.
Anti-double-counting, stated plainly.
Prime Video touches three dimensions and is split rather than repeated: the money taken is counted once in Revenue Extraction, the “No action is needed” framing once in Behavioral Manipulation, and the undisclosed Dolby removal once in Trust & Transparency. The amount taken, the framing used, and the failure to disclose are genuinely distinct harms, not one harm counted three times. Likewise the arbitration term is scored once in Customer Restriction and recorded — not re-scored — as a promise reversal; the Ring grandfathering is counted once as a mitigation; and Amazon's scale, product count, dossier length, regulatory attention and volume of available evidence carried no independent weight at all.
Calibration · Not a normalized head-to-head comparison
Two companies reach 67 by opposite routes.
Amazon and Google land on the same aggregate with almost nothing in common mechanically. The tie is the finding, not an artifact — and the dimension profile, not the total, carries the information. Both pages publish their CHI split on the same v2.0 dimension maxima, so the columns below are directly comparable.
| Dimension | Amazon | |
|---|---|---|
| Revenue Extraction | 14 | 17 |
| Behavioral Manipulation | 16 | 15 |
| Customer Restriction | 15 | 14 |
| Information & Privacy | 11 | 10 |
| Trust & Transparency | 11 | 11 |
| CHI | 67 | 67 |
| CVI | 87 | 88 |
| CFS | +20 | +21 |
Equal totals do not imply equal behaviour
Who pays, and in what currency. Google's structure is two populations and two currencies: end users pay in defaults, attention and lost control on products that are free, while developers, creators and merchants pay in cash. Amazon has one population paying in cash — which is why Revenue Extraction separates by three points. Amazon's adverse conduct is directed at people who already handed over money and then found the money bought less.
Google is higher on all three control dimensions, and correctly so. Its canonical case is a change with no off switch at all. Every Amazon change on this page leaves an exit — pay $4.99, pay $11.99, use the keypad, keep classic Alexa. A paid exit is a poor exit and is scored as one, but it is not the absence of one. Google also has more adjudicated findings, deeper platform control, and a far wider data estate.
Both land at 67. Only one of them is doing it to customers holding a receipt.
Amazon and Apple — two dependency archetypes, and no normalized comparison is published here.
Apple's dependency is architectural: hardware, OS, services, storage, accessories, repair and migration are engineered as one system, and the Apple assessment has direct intent evidence for exit friction. Amazon's dependency is accretive: no single product locks the customer, every layer has a competent substitute — Netflix, Spotify, Kobo, Libro.fm, Google Home, Arlo, Roku, Walmart+ — and the lock is the cumulative unthinkability of moving an entire household at once. Amazon has no intent evidence; it has documentary tells instead, of which the Ring grandfathering is the clearest.
The mechanisms differ in kind, not degree. Apple monetizes by metering — storage tiers, care plans, upgrade cadence, accessory transitions — which is a higher price for more, and it keeps purchased hardware ad-free. Amazon monetizes by re-tiering benefits already bought, converting purchased interfaces into advertising surfaces, and paywalling operational features of installed hardware. Put compactly: Apple charges at the gate; Amazon changes the house after you move in.
Why no side-by-side table. The Apple page currently publishes its CHI dimension split on a five-point severity scale rather than the v2.0 dimension maxima used here, and this site's convention is that a matched-surface panel is required before any comparative superiority claim is published. That reconciliation has not been done, so the distinction above is offered as calibration and not as a head-to-head result. Ecosystem Enclosure was deliberately not assigned to Amazon — assigning it to make the comparison symmetrical would manufacture a second Apple, and Amazon's evidence does not support it. That decision alone holds Customer Restriction two to three points below where an Apple-style reading would have put it.
Associated patterns · Final state, reconciled against the live Lexicon
The patterns materially relevant to this verdict.
Primary
Supporting
Tested and deliberately not assigned
Methodology language kept out of the pattern strip. The Dependency Principle, the Dependency Stack and Fixed Under Pressure are CHI methodology, not patterns. They appear in the reasoning on this page — and Fixed Under Pressure has its own titled treatment in the cancellation exhibit — but none of them is presented as a Lexicon assignment, and no new Lexicon page was created for this build.
Falsification · What the evidence does NOT support
Twelve popular claims about Amazon that did not survive this investigation.
Each of these, had it survived, would have added between two and six points to CHI. Publishing the failures is the price of being believed about the findings above. None of them appears anywhere else on this page, including as insinuation.
Legal status, stated exactly — including the one that would have been the biggest number on this page.
The Luxembourg €746 million GDPR fine is not displayed as a fine anywhere on this page, because it is not one. It was annulled on appeal on 12 March 2026 by the Luxembourg Administrative Court and remitted to the regulator, which confirmed the underlying violations of the legitimate-interest basis and information duties. The confirmed violations informed Information & Privacy at 10/15; the annulled fine did not.
On Prime and ROSCA: a federal judge ruled in September 2025 that Amazon's Prime enrollment violated the Restore Online Shoppers' Confidence Act as a matter of law on the disclosure-before-billing issue; days into trial Amazon settled the FTC's enrollment and cancellation case for $2.5 billion and agreed to a court-ordered redesign. Amazon did not admit wrongdoing and says it has “always followed the law.” This page does not say Amazon admitted trapping customers, because it did not.
On the German Prime Video litigation: outcomes are split and non-final. A Munich court held in December 2025 that introducing advertising without consent was unlawful and the migration email misleading — that ruling is under appeal — while a Bavarian appellate court rejected a consumer-group damages claim in July 2026. A US class action was dismissed with prejudice in July 2025.
Weighting rule applied throughout: adjudicated outcomes above stipulated settlements, stipulated settlements above pending allegations. On that test Amazon's regulatory file is thinner than it looks — one adjudicated statutory violation, with everything else stipulated without admission, under appeal, annulled, remitted, or awaiting trial.
Evidence & methodology
- Company announcementAmazon — Prime Video advertising announcement, 22 September 2023Primary source for “limited ads,” for “We're not making changes in 2024 to the current price of Prime membership,” and for the $2.99 ad-free surcharge. US launch 29 January 2024.
- Company announcementAmazon — Prime Video Ultra, 10 April 2026Source of the $2.99 → $4.99 ad-free repricing, the relocation of 4K, the restoration of Dolby Vision to the base tier, and the “aligns with other major streaming services” rationale.
- Company announcementpress.aboutamazon.com — Prime Music expansion, 1 November 2022Verbatim source for “Prime members can shuffle play any artist, album, or playlist, plus stream a selection of All-Access playlists on demand” — the expansion and the narrowing in a single sentence.
- Company statementRing — subscription requirement statement, 29 March 2023“If you did not own an Alarm, these features will now require a subscription.” The grandfathering exemption and the modes/arming paywall both derive from this statement and from Ring's own support documentation (“Compatible Ring subscription required for digital arming/disarming”).
- Company noticeAmazon — removal of “Download & Transfer via USB,” notified 12 February 2025, effective 26 February 2025The only sanctioned local-backup route for purchased Kindle content. Load-bearing for the Rentalization finding.
- Company termsAmazon — Kindle Store Terms of Use (UK), last updated 30 June 2026“Kindle Content is licensed, not sold, to you by Amazon.” Read against the 2009 terms' “permanent copy” language, recorded in the Yale Journal of Law & Technology.
- Company emailAmazon — withdrawal of “Do Not Send Voice Recordings,” 28 March 2025“We have decided to no longer support this feature.” The cleanest post-purchase privacy downgrade in the record, and the reason Information & Privacy is not lower than 10/15.
- Company statementAmazon — “no plans to add advertising to Alexa,” January 2018Read against Sponsored Tiles (April 2026) and Agentic Ads published on advertising.amazon.com (23 June 2026), and against Andy Jassy's 31 July 2025 description of advertising as “a lever to drive revenue.”
- Court recordFTC v. Amazon (Prime/ROSCA), W.D. Wash. — partial summary judgment, 18 September 2025Holds that Prime enrollment violated ROSCA as a matter of law on disclosure before billing: “no reasonable jury could find in favor of Amazon.” The only adjudicated statutory violation in this file.
- RegulatoryFTC — stipulated order, 25 September 2025$1 billion civil penalty plus $1.5 billion in redress to roughly 35 million customers, a $51 per-customer cap, a mandated clear decline button, parity cancellation, and an external monitor. Settled without admission of wrongdoing.
- Court recordLandgericht München I, 16 December 2025 · Bayerisches Oberstes Landesgericht, July 2026Opposite conclusions on Prime Video advertising in Germany: the Munich ruling (unlawful unilateral modification; migration email misleading) is under appeal; the Bavarian appellate court rejected a consumer-group damages claim. A US class action was dismissed with prejudice in July 2025.
- RegulatoryLuxembourg Administrative Court — CNPD appeal judgment, 12 March 2026The €746 million GDPR fine annulled and the matter remitted to the regulator; the underlying violations confirmed. The confirmed violations are used; the annulled fine is not displayed as a fine anywhere on this page.
- Contract changeAmazon — US Conditions of Use and Prime Terms, effective 14 August 2026Individual arbitration and class-action waiver reinstated, five years after the May 2021 removal. Verified across five concordant secondary sources on date, class waiver, absence of a reported opt-out, 60-day notice, JAMS batching protocol, small-claims carve-out and exclusion of pending litigation. Amazon's own page was not retrievable at build time; batch-size figures conflict between sources and are deliberately not published.
- Independent studyProfitero — Price Wars 2025, 18 November 2025Lowest-priced major US online retailer for the ninth consecutive year at approximately a 14% average price advantage, measured across more than 10,000 products and 23 retailers. The most load-bearing single fact in the CVI case.
- Satisfaction indexAmerican Customer Satisfaction Index — internet retail, 2023–2026Amazon at 84/83/83/82, tied first among online retailers in 2026 against a 79 industry average and Walmart's 77. Independent, longitudinal and customer-reported, measured during the years this page documents.
- Peer-reviewedYu (2024); Dash et al. (2024) — sponsored-result quality measurementSponsored results at approximately 16% higher price and 83% fewer reviews than the organic results they displace. Used at reduced weight, and separated from the unadjudicated intent allegations in the FTC complaint.
- Independent measurementThe Markup (2021); Marketplace Pulse (2019, 2022)Sponsored share of the first screen, Buy Box retention and organic-result density. Establishes the measured share; the FTC's characterisation of Amazon's internal “defects” discussion is excluded as unadjudicated.
- Trade reportingAdweek — Prime Video ad load, June 2025Approximately 2–3.5 minutes per hour rising to approximately 4–6 minutes per hour. Not denied by Amazon. Reported rather than measured, and labelled as such at every appearance.
- Company filingAmazon.com, Inc. — Annual Report on Form 10-K, FY2025The one advertising figure this page permits itself: $68.6 billion in advertising revenue for 2025. Every other Amazon revenue and profit total was excluded on the principle that scale is not hostility.
- Analyst estimateJPMorgan — Prime bundle valuation, 2025 estimateApproximately $1,430 a year against a $139 fee. Displayed as a 2025 estimate, which is the correct year; the 2024 estimate was approximately $1,345.
Entries here are cited to a specific dated document rather than to a URL, following this site's existing convention where no stable public canonical link was verified at build time; every one is sourced in full in the underlying CHI research package. That package runs to a 34-section investigative dossier, a gap-closing and pre-build freeze that re-verified the load-bearing items against primary sources, and a final scoring memo — and is deliberately not reproduced here.
This page is a synthesis of a frozen evidence package, built from a scored and sealed record rather than from a fresh investigation.
The research pipeline was: a full investigative dossier, then a repair pass that reconciled labels, re-verified the load-bearing evidence and froze what could move a score, then an independent scoring memo. Where the dossier and the freeze conflicted, the freeze controlled. Two corrections from that pass are reflected throughout and are load-bearing: the Luxembourg fine was annulled rather than upheld, and the JPMorgan Prime valuation is a 2025 estimate rather than a 2024 one. One development postdated the dossier entirely — the 14 August 2026 arbitration reinstatement — and it removed the 2021 arbitration removal from the customer-positive set rather than merely adding an item, because a repair a company undoes is not a repair.
Evidentiary status is preserved and never upgraded for readability. Pending allegations are labelled as allegations at every appearance and Amazon's denials are carried alongside them. Amazon-reported operational figures — delivery volumes, review-blocking volumes — are marked as company-reported rather than independently verified. Where two phrasings of the same fact differ, the page states the baseline explicitly: Ring Basic is +66% from 2015 and +25% from July 2022, and both numbers appear.
Omitted rather than softened: the annulled Luxembourg fine as a current fine; arbitration batch-size thresholds, where sources conflict; the Conditions of Use “Other Businesses” clause as a quotation, which could not be captured verbatim and is described in prose instead; the unadjudicated FTC search-“defects” and Project Nessie allegations as fact; any personalized-pricing insinuation; and the US Subscribe & Save default-enrollment claim. CHI 67/100, CVI 88/100 and CFS +21 are settled analytical values under Methodology v2.0. They are not a range.
Final verdict
CONCERNING · EXCEPTIONAL VALUE — “FAIR”
Amazon is one of the most valuable consumer companies in this index, and one of the broadest practitioners of post-adoption monetization in it. Neither finding cancels the other, and the whole point of publishing two numbers is that neither has to. The core retail relationship — buy something, have it arrive fast, send it back if it is wrong, at an independently verified price advantage — is not merely non-hostile. It is the strongest value case in the corpus, and it has improved for twenty-one years while the fee tracked inflation.
One layer down, a different company is operating. Between 2022 and 2026, Amazon inserted advertising into a benefit two hundred million people had already paid for and then raised the price of the way out by 67%. It expanded a music catalogue fiftyfold and removed the ability to play a song on demand. It moved the modes on a doorbell customers had wired into their own walls behind a monthly fee — while exempting the customers already embedded, which is the most revealing document in the file. It withdrew, by email, a privacy setting some people had bought the device to use. It removed the last sanctioned way to keep a copy of a book sold under a button that says “Buy now.” And after promising in 2018 that it had no plans to advertise inside Alexa, it shipped two advertising products inside Alexa.
None of that is the worst conduct this index has recorded, and this page does not pretend otherwise. There is no safety harm, no coercive scarcity, no personalized price gouging, no adjudicated antitrust liability — and the twelve loudest claims about Amazon failed testing and are published as failures. The single most severe item, an enrollment and cancellation architecture a federal court found unlawful, has been repaired under order, and CHI credits that repair in full. What holds the number at 67 rather than the 70s is severity; what pushes it there rather than the 50s is that seven distinct surfaces show the same mechanism, all of them post-adoption, all of them still in force, and not one of the 2022–2026 media or device re-tierings has ever been reversed.
That is the honest shape of the record, and it has an asymmetry worth naming. Amazon repairs access and process — cancellation flows, ad intrusiveness, bricked hardware, disclosure — reliably and often generously, sometimes before anyone compels it. It has not repaired a single monetization decision. It is generous on remediation and immovable on the ladder.
The membership costs what inflation says it should. The cheapest, fastest version of almost everything is still here.
So why does keeping what you already had cost $4.99 a month more than it did in 2024?