Latest source review
Last reviewed: September 19, 2026. This review re-checked the Anthropic model documentation, the watermarking explainer, the Claude memory announcement, the file-creation Help Center article, the Claude Design announcement, Anthropic Privacy Center pages, U.S. Copyright Office materials, Federal Register guidance, and FTC AI-claim enforcement materials.
Three product changes drove most of the edits: the model lineup moved twice since the last review, Claude began watermarking generated text in August 2026, and memory became a single store shared between chat and Claude Cowork in September 2026.
The signature experience on this site is The Manuscript at /manuscript/: a writer’s desk that walks a labeled specimen from draft to edited to ready-to-send, with an overlay for watermarked future-model text versus the pre-August 2, 2026 transition. It is editorial — disclosure, what the mark cannot prove about a person — not a detector and not the everyday how-to, which belongs on Claude Helps.
The next useful refresh should check whether the model comparison table has moved again, whether the watermark detection API has expanded beyond private preview, whether Claude Design has left research preview, and whether memory defaults have changed on Team or Enterprise plans.
September 19, 2026 changelog
This update brought every page back in line with the current product state and added one new guide.
Model guidance was rewritten around the current lineup. Claude Opus 5, released July 24, 2026, is the documented starting point for most workloads and therefore the default for daily drafting and editing. Claude Fable 5.1, released September 1, 2026, is the escalation for long-horizon work such as whole-manuscript audits. Claude Sonnet 5 is the volume model for variants and routine passes, and its $2 and $10 per-million-token rate became standard rather than rising on September 1. Guidance referring to Claude Opus 4.8 as current, or Claude Fable 5 as the flagship, was removed.
A new watermarking guide was added at /watermarking, covering how Claude text watermarking works, the EU marking obligation behind it, what it cannot prove, how editing changes detectability, C2PA content credentials on generated image files, and who can run a detection check.
The Manuscript was added at /manuscript/ as the site’s signature desk: a labeled synthetic specimen walked through draft, edited, and ready-to-send, with a toggle between watermarked future-model text and the pre-2026-08-02 transition. It teaches author implications — disclosure, what the mark is not — and links the everyday how-to to Claude Helps rather than duplicating it.
Memory guidance was added to the setup and fiction guides, reflecting that chat memory and Claude Cowork memory are now the same store with topic-level editing and sensitive topics off by default. Export guidance was added to the workflow and editing guides, covering file creation and Claude Design as export surfaces that still require a human read of the exported file. The rights and ethics guide gained an explicit note that watermarking does not make the disclosure decision.
- Added /watermarking as a new guide.
- Added /manuscript/ as the signature writer’s desk for watermarking author implications.
- Added current model-choice guidance for drafting, editing, and volume work.
- Added memory-across-surfaces guidance for long manuscripts.
- Added file creation and Claude Design export guidance with human review gates.
- Added the watermarking-is-not-disclosure boundary to rights and ethics.
- Removed stale Opus 4.8, Fable 5, and Sonnet 5 price-increase claims.
- Refreshed all page review dates and added eight primary sources, bringing the library to 26.
Launch changelog
Published the initial Claude Writes sitemap: workflow, fiction, editing, copywriting, prompts, fact-checking, rights and ethics, setup, resources, and freshness log.
Added T4 search and AI-crawler treatment: llms.txt, llms-full.txt, dense JSON-LD, source blocks, quotable stat callouts, cite-this-page blocks, and primary-source callouts.
Added the independent-publication disclaimer and Claude Network footer links required for the domain cohort.
Watchlist for future updates
The volatile topics are product surfaces and policy controls: the model comparison table, pricing, watermark detection access, memory defaults, projects, skills, styles, Claude Research, Claude for Word, consumer model-improvement settings, commercial data handling, and data retention.
The durable topics are process-oriented: separate drafting from critique, keep source packets, maintain human authorship records, and verify claims before publication.
- Model comparison table: new releases, retirement dates, and knowledge cutoffs.
- Pricing page: per-token rates and cache-read discounts.
- Watermark detection API: whether access expands beyond private preview.
- Watermarking coverage for models launched before August 2, 2026.
- Memory defaults, especially on Team and Enterprise plans.
- Claude Design and file creation: availability, plan gating, and export formats.
- Claude app feature availability and plan limits.
- Skills and styles migration status.
- Claude for Word beta limitations.
- Anthropic privacy and retention language.
- Copyright Office AI report updates.
- FTC enforcement examples involving AI-written claims or fake reviews.
FAQ
Why does this site need a freshness page?
Claude features and policies change. A freshness log gives readers and future maintainers a visible place to refresh source-sensitive guidance.
How often should this page be updated?
Weekly if the site is being actively grown; immediately when Anthropic changes relevant product, privacy, or policy docs.
What should a future update include?
Date checked, sources reviewed, pages affected, and a short explanation of what changed.
Primary sources
Official guidance on clarity, context, examples, XML tags, roles, and long-context prompt structure.
Anthropic DocsChoosing a modelReviewed 2026-09-19Official model-selection matrix and the guidance that most workloads start with Claude Opus 5.
Anthropic DocsModels overview and comparison tableReviewed 2026-09-19Official model IDs, context windows, output limits, and knowledge cutoffs for the current lineup.
AnthropicHow Claude's text watermark worksAugust 14, 2026 (updated September 1, 2026)Official explanation of Claude text watermarking: SynthID-Text method, EU AI Act compliance, no identifying information, and what a watermark does and does not prove.
Claude BlogClaude's memory works everywhere, and you decide what's in itSeptember 2026Official announcement that chat memory and Cowork memory are the same, with topic-level review, editing, deletion, and sensitive topics off by default.
AnthropicIntroducing Claude Design by Anthropic LabsApril 17, 2026Official description of Claude Design for slides, one-pagers, and decks, including export to PDF, PPTX, Canva, or HTML.
Claude Help CenterWhat are projects?March 16, 2026Official description of project workspaces, project knowledge, and project instructions.
Claude Help CenterWhat are skills?June 1, 2026Official definition of skills as folders of instructions, scripts, and resources for repeatable tasks.
Anthropic Privacy CenterIs my data used for model training? - ConsumersMarch 16, 2026Official consumer data-use conditions for model improvement, safety review, explicit opt-in, and incognito chats.
Anthropic Privacy CenterIs my data used for model training? - Commercial customersMarch 16, 2026Official commercial-product default that inputs and outputs are not used for training unless feedback or permission is given.
U.S. Copyright OfficeCopyright and Artificial IntelligenceReviewed 2026-09-19Official hub for the Copyright Office AI reports and public-comment process.
Federal Trade CommissionFTC Announces Crackdown on Deceptive AI Claims and SchemesSeptember 25, 2024FTC enforcement examples for deceptive AI claims, fake reviews, and unsupported performance promises.