Platform Content Liability Checklist: Section 230 Elements, Carve-Outs, Moderation Records, and Escalation

By Casey Scott McKay · 2023-11-02

This checklist runs the platform-liability posture end to end, in the order the work should actually be done - before a complaint arrives, and again the week one does. It starts with the language audit that generates most live counts, then works the three statutory elements against each count, the four carve-outs, and the separate copyright safe harbor that Section 230 never touches. It covers the records a platform must be able to produce - dated terms versions, assent evidence, moderation logs, support tickets, design decision memos - and the escalation triggers that route a product decision to counsel before it ships. Gates mark the items that should clear before any public safety claim is published. It closes with the exposure model, the litigation-day sequence, and the annual review.

IP and Technology > Internet | Checklist | Published 2 November 2023 - Updated 15 November 2024 | Casey Scott McKay - marksy.us

Summary. This checklist runs the platform-liability posture end to end, in the order the work should actually be done — before a complaint arrives, and again the week one does. It starts with the language audit that generates most live counts, then works the three statutory elements against each count, the four carve-outs, and the separate copyright safe harbor that Section 230 never touches. It covers the records a platform must be able to produce — dated terms versions, assent evidence, moderation logs, support tickets, design decision memos — and the escalation triggers that route a product decision to counsel before it ships. Gates mark the items that should clear before any public safety claim is published. It closes with the exposure model, the litigation-day sequence, and the annual review.

Keywords: section 230 elements · publisher or speaker · information content provider · material contribution · independent duty · promissory estoppel · outcome promises · support macros · moderation logging · terms versioning · DMCA agent · repeat infringer policy · FOSTA carve-out · intellectual property carve-out · design decision records · escalation path · assent records · judicial notice · litigation hold · annual review

How to use this checklist

| Phase | What it covers | |---|---| | 1 | The language inventory | | 2 | Support macros and individualized promises | | 3 | Badges, verification, and guarantees | | 4 | Terms versioning and assent records | | 5 | Moderation logging | | 6 | The design decision record | | 7 | The copyright program, kept separate | | 8 | Trafficking and known-harm escalation | | 9 | The intellectual property posture | | 10 | Product escalation triggers | | 11 | The count-by-count chart | | 12 | Element one and element three | | 13 | Element two: the duty test | | 14 | The carve-outs, tested | | 15 | The evidence package | | 16 | Litigation day one | | 17 | Arbitration and insurance | | 18 | Modeling the exposure | | 19 | Remediating after a ruling | | 20 | Ownership and annual review |

Boxes marked [Gate] should clear before any public claim about safety, verification, or removal is published.

The matter. A marketplace with four million listings was sued on nine counts after a counterfeit sale caused an injury. Four counts were clean dismissals. Three of the five survivors existed only because of sentences the company had written about itself.

Phase 1. The language inventory

[ ] [Gate] Collect every published sentence describing what the company will do about content or accounts.
- Why. Barnes v. Yahoo! allows a claim built on a promise, because the duty comes from the promise rather than from publisher status under 47 U.S.C. § 230(c)(1).
- Where to look. Terms of service, community guidelines, help center, seller and buyer protection pages, onboarding email, in-app tooltips, badge explainers, marketing pages, investor materials, and the careers page.
- Trap. Auditing only the terms. The terms are usually the safest document the company publishes.
[ ] Sort every sentence into discretion, process, or outcome.
- Discretion. "We may remove content that violates these guidelines." Safe.
- Process. "Reports are reviewed by our team." Safe until it acquires a number or a deadline.
- Outcome. "We remove all counterfeits." "Every seller is verified." "Your purchase is guaranteed authentic." Live counts.
[ ] For every outcome sentence, choose one of three dispositions and record it.
- Rewrite as discretion or process; qualify it accurately; or accept it as a priced commitment with an owner.
- Why. Somebody should make this choice deliberately. Right now a copywriter is making it.
[ ] Flag every number. Response times, review percentages, accuracy rates, and removal volumes are all discoverable representations.
[ ] Check the company's own assertions. Category labels, automated summaries, computed scores, and editorial content are the company's speech and outside § 230(c)(1) entirely.

Phase 2. Support macros and individualized promises

[ ] [Gate] Export every customer-support template in use and read them all.
- Why. Macros are individualized statements to named people sent thousands of times, which is the ideal shape for promissory estoppel.
[ ] Delete or rewrite any macro containing "we will," "we have removed," "this will be resolved," or a date.
- Safe replacement. State what was received, what the policy is, and that a decision will be communicated.
[ ] Audit outsourced support separately. Vendor-written macros are the company's promises.
[ ] Confirm agents cannot free-text commitments without an approved template, or train them on the distinction.
[ ] Retain support tickets in a form retrievable by user and by content ID.
- Why. The first thing defense counsel needs is every ticket for the plaintiff. The first thing plaintiff's counsel gets is the same set.

Phase 3. Badges, verification, and guarantees

[ ] [Gate] Document what each badge actually means, in a page a court can read.
- Trap. A "Verified" label that means an email was confirmed, next to a tooltip implying identity was checked.
[ ] Confirm the verification actually performed matches the representation.
[ ] Read every guarantee against what operations really does, including the exceptions, the claim window, and the payout cap.
[ ] Confirm the guarantee is contractually located — in accepted terms, with a stated scope — rather than floating on a marketing page.
[ ] Check who the promisee is. A guarantee to buyers is not a promise to a competitor or a brand owner, and standing is often the cleaner defense.

Phase 4. Terms versioning and assent records

[ ] [Gate] Confirm the company can produce, for any user and date, the exact terms version in force and proof of acceptance.
- Why. Without it, the motion to compel arbitration is unavailable and the contract defense is unprovable.
[ ] Archive every version with effective dates, retrievable years later.
- Trap. Assuming the current page is enough. The plaintiff will produce an archived capture of the old one.
[ ] Capture the assent flow itself — a screenshot or rendering of what the user saw, not just the text.
[ ] Log acceptance events with user, version identifier, timestamp, and surface.
[ ] Confirm the modification clause and the notice practice match. See the Online Terms Enforceability Checklist.
[ ] Confirm the arbitration clause and class waiver are present in the versions that matter, and model the mass-arbitration fee exposure before relying on them.

Phase 5. Moderation logging

[ ] [Gate] Log every report and every decision with reporter, content ID, timestamp, policy applied, decision, and actor.
- Why. This is the record that shows a publishing decision made under a published standard, which is the heart of the Zeran analysis.
[ ] Log inaction as a decision. "Reviewed, no action, policy X" is far better than silence.
[ ] Retain logs long enough to cover the limitations periods for defamation, consumer protection, and the relevant IP claims.
[ ] Make logs queryable by account and by pattern, not only by ticket.
- Why. Repeated reports about one account are the fact pattern behind contributory infringement and the § 230(e)(5) trafficking carve-out.
[ ] Enforce policies consistently. Inconsistent enforcement is evidence in every one of these theories.
[ ] Train moderators that logs are business records, not privileged, and will be read aloud.

Phase 6. The design decision record

[ ] [Gate] Record safety-relevant design decisions — the feature, the risk identified, the alternatives considered, the decision, and the reason.
- Why. Lemmon v. Snap made feature design a live theory that Section 230 does not reach.
[ ] Write them accurately and without speculation about liability.
- Trap. An engineer's note reading "legal will hate this but ship it."
[ ] Distinguish content-screening features from product features. Daniel v. Armslist holds that a claim about missing screening is still a publishing claim; Lemmon holds that a claim about the feature itself is not.
[ ] Route post-incident reviews through counsel where they are genuinely conducted for legal advice, and structure them that way from the start.
[ ] Record subsequent changes and their reasons, aware of how the record reads in the relevant jurisdiction.

Phase 7. The copyright program, kept separate

[ ] [Gate] Confirm the designated agent registration under 17 U.S.C. § 512(c) is current.
- Why. Section 230 does not touch copyright — § 230(e)(2) carves out intellectual property — so § 512 is the only shield, and it is conditional.
- Trap. A lapsed registration discovered during briefing.
[ ] Publish and actually enforce a repeat-infringer policy under 17 U.S.C. § 512(i).
[ ] Measure takedown turnaround and keep the metric internally rather than publishing it.
[ ] Handle counter-notices under § 512(g) with the statutory timing, and preserve the misrepresentation exposure analysis under § 512(f).
[ ] Confirm no financial benefit directly attributable to infringement the company has the right and ability to control — promoted listings and commissions are the pressure points.
[ ] See the DMCA Takedown Notice Checklist and The DMCA Safe Harbor.

Phase 8. Trafficking and known-harm escalation

[ ] [Gate] Build an escalation path for reports alleging trafficking, exploitation, or imminent physical harm that reaches a named human within hours.
- Why. § 230(e)(5) preserves civil claims under 18 U.S.C. § 1595 predicated on 18 U.S.C. § 1591, and Doe v. Twitter shows reports plus continued hosting surviving a motion.
[ ] Record what was reported, when, and what was done, with unusual care in this category.
[ ] Consider a warning capability independent of removal.
- Why. Jane Doe No. 14 v. Internet Brands allows a failure-to-warn claim precisely because warning does not require any publishing act.
[ ] Confirm law enforcement referral procedures and who may make them.
[ ] Do not let the immunity become a reason to build no process. Knowledge acquired and ignored is the input to every theory that survives Section 230.

Phase 9. The intellectual property posture

[ ] [Gate] Accept that federal IP claims are outside the statute — 15 U.S.C. § 1125, 17 U.S.C. § 501, 35 U.S.C. § 271 — and build the program accordingly.
[ ] Determine the circuit's rule on state IP claims. Perfect 10 v. CCBill reads § 230(e)(2) as federal-only; Universal Communication Systems v. Lycos points the other way.
- Consequence. Right-of-publicity exposure against an intermediary is circuit-dependent. See Your Face Is Not Public Domain.
[ ] Run the contributory analysis on its own terms. See Who Else Is Liable.
[ ] Maintain a brand-owner reporting channel and measure its response time internally.
[ ] Track repeat sellers across accounts, since willful blindness is the standard that matters here.

Phase 10. Product escalation triggers

[ ] [Gate] Publish four triggers that route a product decision to counsel before launch.
- A new badge, verification concept, or guarantee.
- A new public claim about safety, authenticity, or review speed.
- A change to how content is ranked, recommended, or generated.
- A feature whose failure mode injures a user directly.
[ ] Make the trigger list short enough to be remembered — four items, on one slide, in the product onboarding.
[ ] Give it a service level. A legal review that takes three weeks will be routed around.
[ ] Log the triggers that fired and the outcome, which becomes the compliance record.

Phase 11. The count-by-count chart

[ ] [Gate] Build the chart before any research. One row per count; columns for theory, duty, whose content, carve-out, and call.
[ ] Write the duty sentence in the defendant's voice. "The plaintiff says we had a duty to ___."
- Publishing verbs — remove, screen, edit, withhold, take down — mean the count is covered.
- Promise, warn, design, process, and "not say" mean it is not.
[ ] Mark counts to concede. Conceding two counts in the first paragraph buys credibility for the other seven.
[ ] Send the chart to the client on day two.

Phase 12. Element one and element three

[ ] Confirm interactive computer service status under § 230(f)(2). One paragraph. Batzel v. Smith covers individual users too.
[ ] For each count, identify what makes the content unlawful — not what it is about.
[ ] Identify who supplied that specific defect.
- Compelled dropdowns and filtering: Fair Housing Council v. Roommates.com — content provider.
- Soliciting and paying for unlawful material: FTC v. Accusearch — content provider.
- Neutral forms: Carafano v. Metrosplash — protected.
- User-chosen categories: Chicago Lawyers' Committee v. Craigslist — protected.
- Aggregated ratings: Kimzey v. Yelp — protected.
- Format conversion: Marshall's Locksmith Service v. Google — protected.
- Encouragement without contribution: Jones v. Dirty World — protected.
[ ] Test any generated or summarized content separately. Log input, output, model, and version.

Phase 13. Element two: the duty test

[ ] Run each count against the four escape hatches.
- Contract and promissory estoppel — Barnes.
- Failure to warn — Internet Brands.
- Product design — Lemmon against Armslist.
- Conduct regulation — HomeAway.com v. Santa Monica.
[ ] Pull the plaintiff's support tickets before drafting. Every time.
[ ] Assess recommendation counts by circuit. Force v. Facebook protects matching; Anderson v. TikTok, reading Moody v. NetChoice, does not.
[ ] Do not lead with § 230(c)(2). Its good-faith element invites discovery, and Enigma Software v. Malwarebytes narrowed "otherwise objectionable."

Phase 14. The carve-outs, tested

[ ] [Gate] Do not concede FOSTA on the label. Test whether the § 1591 predicate is pleaded with the particularity Ashcroft v. Iqbal requires.
[ ] Concede federal IP counts and fight them on the merits.
[ ] Brief the state IP question against your circuit, including the adverse authority.
[ ] Check § 230(e)(4) if any count sounds in wiretap or stored communications.
[ ] Confirm no count sounds in copyright before briefing 230 on it.

Phase 15. The evidence package

[ ] [Gate] Build the motion out of the complaint's own allegations wherever possible.
[ ] Identify documents incorporated by reference — anything the complaint quotes or relies on — and attach them complete.
[ ] Draft a narrow judicial notice request: the fact of publication and the contents as published, not the truth of the matter.
[ ] Do not attach explanatory declarations. Conversion under Rule 12(d) forfeits the statute's economic value.
[ ] Move any count needing outside facts to summary judgment deliberately.

Phase 16. Litigation day one

[ ] [Gate] Issue the litigation hold covering moderation logs, support tickets, terms and policy versions, design records, and account data.
[ ] Pull the plaintiff's complete ticket history.
[ ] Retrieve the terms version and assent record for this plaintiff.
[ ] Capture the help center and marketing pages as they exist today, before anyone edits them.
- Trap. A well-meaning marketing edit after suit that reads as spoliation.
[ ] Tender to carriers.
[ ] Assess arbitration before Section 230, and reserve all defenses in the motion to compel.
[ ] Move to stay discovery with the dispositive motion, expecting targeted discovery on live counts.

Phase 17. Arbitration and insurance

[ ] Confirm the applicable terms version contains the clause and that assent is provable.
[ ] Sequence the motions so the 230 filing is not characterized as waiver.
[ ] Model mass-arbitration fee exposure at realistic claimant volumes.
[ ] Identify the responsive policies — media liability, technology errors and omissions, and the personal-and-advertising-injury grant. See Who Pays for the IP Lawsuit.
[ ] Read the IP and contractual-liability exclusions before telling anyone it is covered.
[ ] Resolve panel counsel and consent-to-settle early. See Tendering and Winning an IP Coverage Claim.

Phase 18. Modeling the exposure

[ ] Count the outcome promises found in Phase 1 and multiply by the population they were shown to. That is the contract class size.
[ ] Count the macros containing commitments and the volume sent. That is the estoppel population.
[ ] List features attacked or attackable under Lemmon and whether a safety record exists for each.
[ ] Estimate the cost delta between a pleadings-stage resolution and one reached at summary judgment. That number funds everything in Phases 1 through 10.
[ ] Identify the counts money cannot fix — regulatory and criminal exposure — and treat them separately.

Phase 19. Remediating after a ruling

[ ] Rewrite every outcome promise identified in the audit.
[ ] Rebuild the macros.
[ ] Version and date every policy page, with a retrievable archive.
[ ] Close the logging gaps the litigation exposed.
[ ] Fix the § 512 program — agent, policy, enforcement, turnaround.
[ ] Never settle into a promise. A settlement committing to future moderation outcomes manufactures the next Barnes claim.

Phase 20. Ownership and annual review

[ ] [Gate] Name one owner for the platform-liability posture, with authority over public safety language.
[ ] Re-run the Phase 1 audit annually and after any marketing refresh.
[ ] Re-check the § 512 agent registration annually.
[ ] Re-test the assent record by asking for proof of acceptance for a randomly chosen user and date.
[ ] Review the escalation triggers against what product actually shipped.
[ ] Report to the board on outcome promises outstanding, live commitments, and the delta between promised and actual response times.

Phase 21. Generated and summarized content

[ ] [Gate] Inventory every place the product creates text, images, or scores that a user did not write.
- Listing summaries, review digests, computed safety or quality scores, autocomplete suggestions, translated content, category assignments, and chat responses.
- Why. § 230(c)(1) protects information "provided by another." Whether machine output qualifies is unresolved, and § 230(f)(3) defines a content provider as anyone responsible "in whole or in part" for creation or development.
[ ] Log input, output, model identity, and version for every generation, retained for the limitations period.
- Why. The defense depends on being able to show what came from a third-party source and what the system added.
[ ] Classify each output surface as reproductive or generative.
- Reproductive — restates, translates, reformats, or aggregates supplied content. Closer to Marshall's Locksmith Service v. Google and Kimzey v. Yelp.
- Generative — asserts something no source asserted. Treat as the company's own speech.
[ ] Tie retrieval-based output to identified sources, since a visible citation is evidence of whose information it is.
[ ] Disclose generated content as generated.
[ ] Do not rely on Section 230 in a risk memo about model output. Rely on the underlying defenses, the terms, and the review workflow. See Buying a Model and the AI Procurement and Governance Toolkit.

Phase 22. Account actions and the removed user

[ ] Publish the grounds for suspension and termination in the terms, with discretion reserved.
[ ] Give notice and a stated appeal path where feasible, and log both.
- Why. Most removed-user claims are contract claims, not § 230(c)(2) claims, and the terms decide them.
[ ] Resolve removal disputes on § 230(c)(1) where possible — deciding not to publish is a publishing function, and (c)(1) has no good-faith element.
[ ] Watch competitive context. Enigma Software v. Malwarebytes holds that "otherwise objectionable" does not cover blocking for anticompetitive reasons.
[ ] Preserve the removed content and the decision record, since the terminated user's claim will turn on both.
[ ] Confirm data export and wind-down obligations in the terms. See Managing Platform Account Risk.

Phase 23. The privacy layer, which Section 230 does not touch

[ ] [Gate] Separate content liability from data liability in the risk register.
- Why. How the platform collects, shares, and retains user data has nothing to do with publishing another's content, and no 230 argument reaches it.
[ ] Confirm the privacy notice matches actual practice, including adtech, analytics, and any sale or sharing.
[ ] Confirm the rights workflow works for access, deletion, correction, and opt-out. See the State Privacy Law Applicability and Readiness Checklist.
[ ] Screen for minors. A general-audience service with actual knowledge of child users acquires obligations that no immunity offsets. See the Children's Privacy Compliance Checklist.
[ ] Confirm the incident response plan covers user-content systems, including moderation logs and support tickets.
[ ] Check § 230(e)(4) before disclosing user data outside legal process.

Phase 24. Working with brand owners and rights holders

[ ] Publish one reporting channel per right — copyright, trademark, counterfeit, publicity — with the elements each requires.
[ ] Measure internal response times by channel, and do not publish the numbers.
[ ] Track reporters as well as reports. A rights holder submitting hundreds of notices is a relationship, and a repeat bad-faith reporter is a § 512(f) problem for them and a resource problem for you.
[ ] Escalate patterns, not just items. One seller with forty notices across nine accounts is a different problem from forty sellers with one each.
[ ] Keep the counterfeit program's evidence usable. Test buys, seller records, and payment trails become the plaintiff's exhibits in a mass action. See Running an E-Commerce Counterfeit Enforcement Program.
[ ] Do not promise a rights holder an outcome in a program agreement. The same Barnes problem applies with a more sophisticated promisee.

Phase 25. Communications discipline

[ ] [Gate] Train product, marketing, and support that everything they write about safety is evidence.
[ ] Ban absolute words in public copy — all, every, always, never, guaranteed, ensures — unless legal has approved the specific sentence.
[ ] Route press statements about incidents through counsel. A statement that the company "immediately removed" content it removed after six days is a consumer-protection claim.
[ ] Keep internal escalation channels usable. Employees who fear writing anything down produce no record at all, which is worse than an imperfect one.
[ ] Distinguish privileged advice from business discussion in writing, and keep legal analysis out of product tickets. See the IP Privilege Protection Checklist.

Phase 26. Multi-defendant and mass-filing postures

[ ] Recognize a campaign early. Similar complaints from one firm, filed in one district, with identical count structures.
[ ] Build one chart template and one motion template, but do not file identical motions where counts differ — count-by-count discipline is what makes these motions work.
[ ] Coordinate the record. Every motion in the campaign will be read against the others, and inconsistent characterizations of the same feature are fatal.
[ ] Consider consolidation or coordination where the district permits it.
[ ] Assess whether the campaign targets a specific product surface, and whether changing that surface ends it.

Phase 27. The board and budget conversation

[ ] Present the delta. Cost of a pleadings-stage resolution against cost of the same case at summary judgment. That number funds Phases 1 through 10.
[ ] Present the promise inventory as a liability schedule, with population counts.
[ ] Be precise about what the statute does. It bars a theory of liability, not a lawsuit, and in most circuits a denial cannot be appealed immediately.
[ ] Identify the counts money cannot fix — regulatory and criminal — and route them separately.
[ ] Ask for the one thing that needs funding, usually the assent-record and terms-versioning capability, since it is the cheapest capability with the largest effect.

Phase 28. The ninety-day plan

[ ] Days 1-15. Phase 1 language audit and Phase 2 macro export. Produce the inventory with dispositions.
[ ] Days 16-30. Rewrite approved language. Publish the four escalation triggers. Confirm the § 512 agent registration.
[ ] Days 31-50. Stand up terms versioning with dated archives and an assent event log.
[ ] Days 51-70. Close moderation logging gaps; make logs queryable by account and pattern; start recording design safety decisions.
[ ] Days 71-85. Rebuild macros; train support; run one tabletop on a trafficking-category report and one on a rights-holder pattern report.
[ ] Days 86-90. Name the owner, set the annual review date, and report the promise inventory to the board.

Phase 29. Rewrite reference: before and after

Use these as the pattern for Phase 1 dispositions. The left column is what companies write; the right is what survives being quoted back.

| Written | Rewritten | |---|---| | "We remove all counterfeit listings." | "Listings reported as counterfeit are reviewed against our authenticity policy, and we may remove them." | | "Every seller is verified before listing." | "Sellers must complete our onboarding checks, which are described here." | | "Reports are reviewed within 24 hours." | "Reports are reviewed by our team. Volume affects timing." | | "Your purchase is guaranteed authentic." | "Eligible purchases are covered by the Buyer Protection terms, which set out the conditions, the claim window, and the limits." | | "We will remove that listing today." (macro) | "We've received your report about listing #12345. It has been routed for review under our authenticity policy, and we'll let you know the outcome." | | "We have removed the content." (macro, sent before removal) | "Your report has been escalated. You'll receive a notification when a decision is made." | | "Verified" badge, no explanation | "Verified" badge, tooltip: "This seller confirmed a business email address and completed identity checks." | | "Our platform is safe for children." | Remove entirely, or state the specific measures and their limits. | | "We use AI to ensure listing accuracy." | "Listing descriptions may be summarized automatically. Summaries are generated from seller-supplied text." | | "We never share your data." | State the actual disclosures, in the privacy notice, matching the data map. |

[ ] Run every flagged sentence through this pattern, and record who approved the replacement.
[ ] Check that the replacement is still true. A process promise the company does not keep is a consumer-protection claim with better facts than the outcome promise it replaced.
[ ] Update the archived version set the same day.

Phase 30. The evidence request, written in advance

Draft this as a standing internal request so that on litigation day one it can be sent unchanged. Each item names the system and the owner.

[ ] Complete support ticket history for the named user, including internal notes, from the ticketing system.
[ ] Every macro version sent to that user, with send timestamps.
[ ] The terms version in force on each relevant date, plus the rendered assent screen and the acceptance event record.
[ ] Help center and marketing page captures for each relevant date, from the archive.
[ ] Complete moderation history for the content and the account — reports received, decisions, policies applied, actors, timestamps.
[ ] All prior reports concerning the same account or the same pattern.
[ ] Revenue attributable to the content — commissions, promoted placement fees, ad revenue.
[ ] Design and safety records for each feature named in the complaint, plus any change record after the incident.
[ ] Generation logs for any automated output at issue, with model and version.
[ ] § 512 records — agent registration history, notices received for this content, counter-notices, repeat-infringer actions.
[ ] Insurance policies in force on the relevant dates.
[ ] The escalation trigger log for the feature at issue.

Why this belongs in the checklist rather than in a litigation memo. Half of these items do not exist in retrievable form at most companies, and finding that out in week one of a lawsuit is a much worse way to learn it than finding out during an annual review.

Phase 31. Deployment-pattern boxes

Different services fail differently. Find the pattern that matches and run its extra boxes.

Marketplace with fulfillment.

[ ] Determine whether state products-liability law treats the platform as a seller, which is not a Section 230 question at all and varies by state and fulfillment model.
[ ] Separate listing content from transaction conduct in the risk register; payment processing and fulfillment are the platform's own acts, per HomeAway.com v. Santa Monica.
[ ] Confirm the authenticity guarantee's terms, cap, and claim window are in accepted terms rather than on a marketing page.

Review or ratings site.

[ ] Confirm computed scores are attributed to their inputs, following Kimzey v. Yelp.
[ ] Confirm sales staff never suggest that review placement is influenced by advertising, which converts a protected arrangement into the site's own actionable representation.
[ ] Check for an applicable anti-SLAPP statute alongside the § 230(c)(1) motion.

Classifieds or listings without fulfillment.

[ ] Keep categories user-selected rather than compelled, per Chicago Lawyers' Committee v. Craigslist and against Fair Housing Council v. Roommates.com.
[ ] Confirm no required field forces a legally significant characteristic.

Social or community product with recommendation.

[ ] Determine the circuit's position on recommendation claims — Force v. Facebook against Anderson v. TikTok.
[ ] Record ranking-system design decisions with the same discipline as safety features.
[ ] Build the trafficking escalation path in Phase 8 with unusual care.

Retailer, publisher, or employer hosting user content incidentally.

[ ] Confirm a designated agent exists under 17 U.S.C. § 512(c); these companies almost never have one.
[ ] Note the upside: with no trust-and-safety marketing, the Phase 1 audit usually finds nothing, which makes the motion clean.

Infrastructure or service provider to other platforms.

[ ] Confirm interactive computer service status in its own right under § 230(f)(2).
[ ] Review customer contracts for indemnities and for any commitment to police customer content.

Phase 32. Quarterly metrics worth keeping

Six numbers, reported internally, never published.

[ ] Outstanding outcome promises, by surface. Target: zero unowned.
[ ] Median and ninety-fifth percentile report-to-decision time, by category. Compare against any published claim.
[ ] Reports closed with no logged decision. Target: zero.
[ ] Accounts with three or more upheld reports still active. This is the willful-blindness metric.
[ ] Escalation triggers fired, and how many launched anyway.
[ ] Assent-record retrieval success rate, tested on ten random users and dates per quarter.

Outcome. The audit found thirty-one outcome sentences across nine surfaces, of which nineteen were rewritten, eight were qualified, and four were kept as priced commitments with a named owner. Eleven support macros were rebuilt. The DMCA agent registration had lapsed fourteen months earlier and was restored before the copyright count was briefed. Of the nine counts, four were dismissed with prejudice, two were dismissed without prejudice on standing and promisee grounds, and three proceeded — the Lanham Act count, a state consumer-protection count based on the company's own advertising, and a design count. The company could not produce an assent record for the plaintiff, so the arbitration motion was never filed; building that capability was the first item on the remediation list.

Key Authorities at a Glance

| Authority | Proposition | |---|---| | 47 U.S.C. § 230(c)(1) | No treatment as publisher of another's content | | 47 U.S.C. § 230(c)(2) | Good-faith restriction; scienter element | | 47 U.S.C. § 230(e)(2) | Intellectual property carve-out | | 47 U.S.C. § 230(e)(4) | Communications privacy carve-out | | 47 U.S.C. § 230(e)(5) | FOSTA carve-out | | 47 U.S.C. § 230(f)(2)-(3) | Definitions | | Zeran v. America Online | Publishing functions | | Barnes v. Yahoo! | Promises create independent duties | | Fair Housing Council v. Roommates.com | Material contribution | | Carafano v. Metrosplash | Neutral tools | | Chicago Lawyers' Committee v. Craigslist | User-chosen categories | | Kimzey v. Yelp | Aggregated ratings | | Marshall's Locksmith Service v. Google | Automated conversion | | Jones v. Dirty World | Encouragement is not development | | FTC v. Accusearch | Soliciting unlawful content | | Jane Doe No. 14 v. Internet Brands | Failure to warn | | Lemmon v. Snap | Design defect independent of content | | Daniel v. Armslist | Screening claims in design costume | | HomeAway.com v. Santa Monica | Platform's own conduct | | Force v. Facebook | Algorithmic matching protected | | Anderson v. TikTok | Recommendation as first-party speech | | Moody v. NetChoice | Curation as expression | | Perfect 10 v. CCBill | (e)(2) federal-only | | Universal Communication Systems v. Lycos | Contrary state-IP view | | Enigma Software v. Malwarebytes | "Otherwise objectionable" limits | | Doe v. Twitter | FOSTA carve-out applied | | Batzel v. Smith | Provider or user | | Ashcroft v. Iqbal | Pleading standard | | 17 U.S.C. § 512 | Copyright safe harbor; agent; repeat infringers | | 18 U.S.C. § 1591 | FOSTA predicate | | 18 U.S.C. § 1595 | Civil trafficking remedy | | 15 U.S.C. § 1125 | Lanham Act, outside 230 | | Fed. R. Civ. P. 12 | Motion practice and conversion |

The five things people get wrong

Auditing the terms and stopping. The terms are the safest document the company publishes. The live counts come from the help center, the badges, and the macros.

Treating Section 230 and § 512 as one shield. They are separate statutes with separate elements, and copyright is expressly carved out of Section 230. A platform can win every 230 count and lose the case on a lapsed agent registration.

Believing immunity is a reason to build no process. Knowledge acquired and ignored is the input to the trafficking carve-out, contributory infringement, and failure-to-warn. Moderation does not forfeit protection; ignoring reports creates other exposure.

Not being able to prove what the terms said. Without dated versions and assent records, the arbitration motion is unavailable and the contract defense is unprovable, and the plaintiff's archived capture becomes the only evidence.

Writing design safety analyses as if nobody will read them. After Lemmon they are litigation exhibits. Write them accurately, record the decision made and why, and leave the speculation out.