How ratings work
Every entry gets a stage from 1 to 4: how far a product has moved from serving its users toward extracting value from them.
An AI model applies the rules on this page to public evidence. If a rating doesn’t follow from them, the rating is wrong.
What a rating measures
The deal between a product and the people who use it—including the sellers, creators, and workers on a marketplace. The idea is Cory Doctorow’s enshittification: platforms are good to users while they grow, lock them in, then shift the value to business customers and shareholders.
It is not a score for quality, popularity, the company’s wider ethics, or its financial prospects. Each entry rates one product: Instagram and Facebook are rated separately.
The four stages
They compete to offer a good deal—features, trust, generosity—so people pile in.
Could you leave tomorrow without losing much?
Rule Alignment, no lock-in, at most one extraction criterion. 21 entries →
Your network, data, purchases, and habits make leaving costly or awkward, even as the bargain shifts.
Do people stay because it’s good, or because leaving costs them?
Rule Anything that isn’t 1, 3, or 4. 17 entries →
Ads, fees, dark patterns, and worse defaults as incentives tilt from pleasing users toward harvesting value.
Is it being made worse in ways that make money?
Rule Two extraction criteria, or lock-in plus one. 52 entries →
The experience rots—worse for users and often hollow for the business—as people drift away.
Would someone joining today find what made it worth joining?
Rule Stage 3, plus a decline criterion. 4 entries →
The criteria
Every entry is checked against the same 17 criteria. Alignment and lock-in describe how a product is built. Extraction and decline describe what it has done, and count only with a dated source from the last 3 years. Each entry page shows its checklist and sources.
Alignment What protects users from the incentive to extract?
- A1 Paid for without surveillance
- Funded by its users, or by ads that don't track them.
- A2 Nonprofit or community-run
- No shareholders to pay out.
- A3 Open and portable
- Open source or an open protocol; you can take your data and account elsewhere.
Lock-in What does it cost you to leave?
- L1 Network or reputation
- Your contacts, audience, or reputation can't come with you.
- L2 Data, library, or purchases
- Your content, history, purchases, or points can't move elsewhere.
- L3 Hardware or workflow
- Devices, integrations, or proprietary formats make switching costly.
- L4 Gatekeeper or default
- It controls how rivals reach people, or is the pre-installed or paid-for default.
Extraction Is it being made worse in ways that make money?
- E1 Pay more for the same
- Price rises, smaller plans, or features moved behind a paywall. New optional extras don't count.
- E2 More ads
- Ads where there were none, or more and longer ads.
- E3 Squeezing the other side
- Higher fees, lower payouts, or worse terms for sellers, creators, workers, or developers.
- E4 Hidden costs and dark patterns
- Junk fees, hard-to-cancel subscriptions, or deceptive design, shown by a regulator, court, or settlement.
- E5 Data grab
- Your data or content used for ads or AI training without a clear choice, or access to it closed off.
- E6 Self-dealing
- Favouring its own products, blocking rivals, or buying up competitors.
- E7 Worse defaults
- What you chose replaced by algorithmic or promoted content, or unwanted features forced on.
Decline Is it failing the people on both sides?
- D1 Users leaving
- A documented, sustained fall in users, usage, or paying customers.
- D2 Suppliers leaving
- Sellers, creators, contributors, or advertisers documented leaving.
- D3 Core function degraded
- Documented evidence that the main thing it does has got worse.
How the stage is decided
- Stage 3 Two extraction criteria, or any lock-in plus one.
- Stage 4 Stage 3, plus at least one decline criterion.
- Stage 1 Alignment, no lock-in, and at most one extraction criterion.
- Stage 2 Everything else.
Once people can’t easily leave, one documented move against them is enough to count as cashing in. The site won’t publish a stage that disagrees with its checklist, so judgment goes into one place: whether evidence meets a criterion.
- Evidence has to clearly show the criterion—for good claims as much as bad ones.
- Only the last 3 years count. Reversed changes don’t.
- Ownership alone never moves a stage; it has to show up in how people are treated.
- Squeezing sellers, creators, or workers counts the same as squeezing users.
- Changes forced by regulators, courts, or payment processors don’t count.
- When it’s unclear, it doesn’t count.
Evidence
Counts
- Prices, plans, and paywalls
- Ads and where they appear
- Terms, privacy, and data use
- Features removed or made harder
- Fees for sellers and creators
- Rulings, fines, and settlements
- Official user and customer figures
Doesn’t count
- Rumours and unsourced claims
- Individual complaints or outrage
- Stock prices
- Taste in design
Each piece of evidence is listed on its entry with a date and a source link, plus up to two tags for the business strategy behind it. Tags explain motive; they don’t decide the stage.
All 22 strategy tags
- Drip pricing
- Fees revealed late or in pieces — junk fees, service charges, store taxes, confusing promo structures that inflate the real price.
- ARPU growth
- Systematic price increases or feature cuts designed to raise average revenue per user, often announced as "value alignment."
- Take rate
- The cut the platform takes on each transaction — seller fees, creator payouts, OTA commissions, payment rails — incrementally expanded over time.
- Ad expansion
- Growing ad load, launching ad tiers on formerly ad-free products, or substituting organic results with paid placements.
- Tier fragmentation
- Splitting one product into multiple paid tiers to upsell existing users, gating previously free features, or adding paywalled subscription layers.
- Subscription bundling
- Packaging unrelated products together to inflate perceived value, prevent churn, or cross-sell captive audiences into adjacent services.
- Ecosystem lock-in
- Raising switching costs through data silos, proprietary formats, network dependencies, or deep workflow integration that makes leaving painful.
- Enterprise upsell
- Land-and-expand playbook — get free or cheap seats into an org, then convert to expensive enterprise contracts through seat growth or AI add-ons.
- Self-preferencing
- Using platform control to advantage own products — default settings, gatekeeper rules, distribution deals, anti-steering policies.
- Market consolidation
- Acquiring competitors, rolling up adjacent markets, or leveraging dominance to define market structure and crowd out alternatives.
- Two-sided squeeze
- Extracting value from both sides of a marketplace simultaneously — raising fees on merchants/publishers while degrading value for end users.
- Data enclosure
- Locking in or monetising data that users or third parties created — API paywalls, training data grabs, privacy-as-marketing while selling data.
- Algorithmic feed
- Optimising content ranking for engagement or ad revenue rather than user satisfaction — surfaces outrage, suppresses chronological feeds.
- Dark patterns
- UX designed to mislead — hidden cancellations, confusing defaults, manufactured urgency, consent dialogs engineered for the wrong click.
- Liquidity event
- IPO, direct listing, take-private, or PE acquisition that shifts incentives toward near-term extraction to satisfy new capital structure.
- Investor pressure
- Public-market or VC pressure to hit growth/margin targets — cost cuts, layoffs, product resets, and guidance cadence that shapes product decisions.
- Regulatory pass-through
- Compliance costs, tariffs, or regulatory settlements passed directly to users or merchants rather than absorbed by the platform.
- Regulatory risk
- Antitrust scrutiny, compliance liability, legal exposure, or jurisdictional risk that signals the platform has accumulated significant market power.
- AI upsell
- AI features bundled into premium tiers, "copilot" add-ons, or enterprise AI contracts used to justify price increases or tier upgrades.
- Creator squeeze
- Cutting creator payouts, raising platform fees on creators, or increasing dependency while reducing the economics of being a creator.
- Governance shift
- Leadership change, conversion from nonprofit to for-profit, donor dependency, policy opacity, or erosion of community governance norms.
- Wage arbitrage
- Classifying workers as contractors, suppressing pay via algorithmic dispatch, or using regulatory gaps to avoid labour obligations.
Who rates
An AI model—currently Claude, by Anthropic—chooses the evidence, writes the reviews, and applies the checklist. No human reviews individual ratings; people run the site and choose what to cover.
Claude is itself an entry. It is rated under the same rules, but weigh its evidence with that conflict in mind. AI models make mistakes, which is why every rating links its sources.
Changes
Each review either changes a stage or affirms it. Every change is recorded in the entry’s rating history with the reason; ratings never change silently. The method itself is versioned:
- 2.1 Lock-in plus one extraction criterion counts as stage 3. Every entry was searched for recent evidence and rescored.
- 2.0 Stages computed from a published checklist and a fixed rule.
- 1.0 First published.
A rating is editorial opinion based on public information—not a statement about any company’s intentions, and not legal, financial, or investment advice.