MorseSign inStart freeOpen Morse

Does Substack train AI on your posts? What its AI block covers and what it leaves open

The Morse teamPublished 9 min read

No, Substack says it doesn't use what writers publish to train generative AI models, as of November 13, 2026, and none of its policies mention selling or licensing posts to AI companies. Substack's Block AI training setting asks outside AI crawlers not to train on a publication, but only when the writer turns it on, only on that publication's own web pages, and only for crawlers that choose to listen. Notes, which live on substack.com, sit outside it.

We read Substack's Terms of Use (updated October 6, 2026), Publisher Agreement (November 2, 2026), Privacy Policy (May 14, 2026), CCPA Policy, Content Guidelines, 14 help center articles, one post on Substack's own blog, and the robots.txt files of substack.com and several publications, all on November 13, 2026. Substack deserves credit for putting its no-training line in writing, and for giving every writer a switch that asks AI crawlers to stay away.

Substack and AI: the key facts

Read November 13, 2026.

Setting or featureDefaultWhat it coversSource
Block AI training ("Tell AI tools not to train their models on your content")Off until the writer turns it on; Substack's own blog has it offThe publication's robots.txt: 14 named AI crawlers and ai-train=no, with ai-input=yes left onSubstack help, robots.txt
Notes and your profileNo settingServed from substack.com, whose robots.txt names no AI crawlersubstack.com robots.txt
Paid postsBehind the paywallCrawlers without a subscription see the free preview onlySubstack help
Scan for AI text (Pangram)On for posts and notes published since July 21, 2026; writers turn it off post by postReaders can send your text to Pangram for an AI estimateSubstack help
AI images, transcripts, and clipsUsed when you click themRun by "third-party service providers" under Substack's privacy policyTerms of Use
Support chatbotUsed when you open itChats go to Decagon.aiTerms of Use
Substack MCP serverOff until a Bestseller admin connects itRead-only publication stats sent to the AI assistant you pickSubstack help
Direct messagesOn; anyone with a profile can send a requestNot end-to-end encrypted; staff may read them for enforcement and supportPrivacy Policy, Substack help

Does Substack train AI on your posts?

Substack says it doesn't. Its help article on AI detection says, "Neither Substack nor Pangram uses publisher content to train generative AI models" (Substack). Substack's July 24, 2026 blog post on the same tools says they "do not use publisher content, either through Substack or Pangram, to train generative AI models" (Substack).

The contracts are quieter. The Terms of Use, updated October 6, 2026, don't mention AI training either way (Substack). They say "you own what you create" and give Substack a "royalty-free, perpetual, irrevocable, and worldwide" license to your posts "to enable us to provide, improve, and notify you about new features within Substack." The Publisher Agreement, updated November 2, 2026, adds a license to run your writing through "AI-based detection systems" and says that license "does not permit us to use your content to train third-party AI models" (Substack). So the plainest no-training promise sits in a help article and a blog post, while the binding documents rule out third-party training only for the detection license.

Does Substack sell your data to AI companies?

Substack says it doesn't sell personal information. Its CCPA Policy says, "We do not sell your Personal Information" (Substack). None of the Substack policies we read mention licensing posts to AI companies, and we found no such deal announced on Substack's own pages.

The Publisher Agreement's license is "sublicensable" and "transferable," but its stated purpose is "to market your publications and to permit others to use, access, and download your publications through Substack" (Substack). The Privacy Policy reserves the right to share anonymized information "to trusted partners not specified here" (Substack). It also lists "generative AI services" among the service providers that receive personal information on Substack's behalf. Reddit, by contrast, writes it into the license: its User Agreement includes "the right to use Your Content to train AI and machine learning models" (Reddit). We cover Reddit's deals in does Reddit sell your data to AI.

How do you block AI training on Substack?

Open your publication's Settings, scroll to Privacy, and slide the toggle next to "Tell AI tools not to train their models on your content" to the right (Substack). Substack says the setting "will only apply to AI tools that respect this setting." For how robots.txt opt-outs work across WordPress.com, Squarespace, and your own site, see how to block AI training on your website.

The help article doesn't state a default. It describes the block as something you switch on, and the publications we checked without it, including Substack's own blog at on.substack.com, carry none of its lines (on.substack.com robots.txt). Treat it as off until you turn it on.

What does Substack's AI block write?

It adds lines to one file. Substack's help article doesn't list them, so we read the robots.txt of a publication with the block on and of Substack's own blog without it, on November 13, 2026. The page data behind a publication's posts names the switch block_ai_crawlers.

robots.txt lineBlock on (example)Block off (on.substack.com)
GPTBotDisallow: /Absent
ClaudeBot, anthropic-ai, Claude-WebDisallow: /Absent
Google-Extended, Applebot-ExtendedDisallow: /Absent
meta-externalagent, CCBot, BytespiderDisallow: /Absent
cohere-ai, cohere-training-data-crawler, AI2Bot, Ai2Bot-DolmaDisallow: /Absent
AmazonbotDisallow: /, except Allow: /feed/alexaAbsent
Content-Signalsearch=yes, ai-input=yes, ai-train=noAbsent
__usage_ai-training (the shared token drafted October 30, 2026)AbsentAbsent
PerplexityBot, OAI-SearchBotAbsentAbsent

The Content-Signal line uses the vocabulary of Cloudflare's Content Signals Policy, which defines ai-input as feeding content into AI models for "retrieval augmented generation, grounding, or other real-time taking of content for generative AI search answers" (Cloudflare). With the block on, Substack keeps that signal at yes. A publication that blocks training still tells AI search tools they may read its posts to answer questions, which fits Substack's warning that blocking may cost you discoverability in AI results.

The file has no group for __usage_ai-training, the shared training token in an Internet-Draft submitted on October 30, 2026 (IETF). That draft says that if no group names the token, "no preference has been expressed," and that a User-Agent: * group "MUST NOT be interpreted as matching" it. Our guide to blocking AI training on your website covers which AI companies have said they'll read it.

Does the block cover Notes, comments, and paid posts?

Not Notes. A note's address is substack.com/@handle/note/…, so the file that governs it is substack.com's robots.txt, and that file names no AI crawler and has no Content-Signal line (substack.com). We opened a note by the writer of the example publication above on November 13, 2026: it loaded from substack.com, with no robots meta tag, even though that writer's publication blocks AI training. A post opened through the Substack Reader at substack.com redirects to the publication's own address, so posts stay under the publication's file.

Comment threads on a post sit on the publication's domain, so its file applies to them. Links to single comments (/p/*/comment/*) are disallowed in the general User-agent: * group on every publication we read, with or without the block.

Paid posts are partly covered by the paywall itself. Substack's help center says free readers see a preview on the web, where the post displays "This post is for paid subscribers" (Substack). A crawler without a subscription sees the same preview, so the text above the paywall is as public as a free post. Emails are outside robots.txt entirely. Substack notes that subscribers may forward an emailed post "even if your Substack publication is private" (Substack).

What do Substack's own AI features do with your writing?

They send text to outside AI providers, under Substack's privacy policy. The Terms of Use say Substack's AI services "are facilitated by third-party service providers, together with those vendors' own subprocessors and service providers," and that the Privacy Policy covers that data (Substack). Substack's own pages describe each feature:

  • Scan for AI text. Readers can run any post or note published since July 21, 2026 through Pangram, which estimates how much was "human-written or AI-assisted" (Substack). Writers can turn detection off from a draft's Publish page with Disable AI detection, one post at a time. The Publisher Agreement says detection tools are "prohibited from retaining your content" (Substack). Writers can also add a "How I make this" statement under Settings, then Details, which readers see when they scan one of their posts, notes, or replies (Substack). Readers who turn on Label content see a badge on posts whose writers turned detection off (Substack).
  • AI images. The post editor's image button offers Generate, which makes an image from a prompt (Substack).
  • Transcripts and clips. Substack auto-generates transcripts for audio posts, which it calls "AI-generated" (Substack), and offers "AI-generated clips" of video posts (Substack).
  • Support chatbot. Support chats are "facilitated by and provided to a third-party service provider, Decagon.ai" (Substack).
  • The MCP server. Admins of Bestseller publications can connect Claude, ChatGPT, or Cursor to mcp.substack.com, which gives the assistant read-only access to subscriber counts, revenue, and traffic (Substack). It "cannot access profile data or Notes activity." Once those numbers reach the assistant, its own training settings apply.

What does Substack do with readers' data?

It runs your account, shares your details with the writers you subscribe to, and personalizes your feed. The Privacy Policy says that when you subscribe, Substack gives the writer "the information necessary (including your name and email address)" (Substack). Substack profiles you "to suggest content on our platform that is relevant to your interests," and you can object at privacy@substackinc.com. Your public profile "may be shared or accessed via automated means (e.g., APIs or indexing services)."

Direct messages get the plainest warning. The Privacy Policy says they "are not end-to-end encrypted, and are not a substitute for secure messaging services" (Substack). It says Substack personnel "may access the contents of direct messages" to enforce its terms, secure the platform, or give support. The Terms of Use add "automated scanning of direct messages for spam, malicious content, and child abuse material" (Substack). Neither document says DMs, comments, or subscriptions are used to train AI.

How Morse keeps unpublished drafts private

A published post lives on public pages where a robots.txt line is only a request, and a draft you paste into an AI tool sits on that tool's servers under that company's rules. No Substack setting reaches either one.

On Morse, the conversation about a draft lives on your device. Your chat history is stored in your browser, encrypted, not on Morse's servers. Morse doesn't train AI on your content, sell your data, or show ads, so there's no toggle to remember before the next post. Pick the mode per chat, starting with the one for the draft in front of you:

  • Private, for the essay you're still shaping. Every new chat starts here, on open-weight models like DeepSeek, Kimi, and GLM, or on Grok. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
  • E2EE, on Pro, for the piece no one should read before it runs. Your browser checks the enclave's hardware attestation, then seals the draft so only that enclave can open it. Morse's servers pass along ciphertext they can't read. The guarantee depends on the enclave hardware and attestation working as designed.
  • Anonymous, for a line edit from GPT, Claude, or Gemini on passages you don't mind a provider seeing. Morse's proxy removes your account, email, and IP address before the passage goes out under Morse's own account. The provider reads it without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy, so keep unpublished drafts that matter in Private or E2EE.

Sign out and your history locks. Because we don't store chats, we usually have very little to give authorities. Clearing your browser deletes those chats, and Morse can't restore them, so keep the manuscript in your own files. Private AI for writers covers that workflow in more depth.

What to change on Substack today

  • Your publication: turn on "Tell AI tools not to train their models on your content" under Settings, then Privacy, on every publication you run.
  • Your Notes: treat them as outside the block, and post longer pieces as posts on your publication.
  • AI detection: use Disable AI detection on each post you don't want sent to Pangram, and write a "How I make this" statement under Settings, then Details.
  • DMs: move anything sensitive to an end-to-end encrypted messenger, since Substack's policy says its DMs are "not a substitute for secure messaging services."
  • Drafts: work on them in a Private or E2EE chat on Morse, where your history stays in your browser.

Morse's Free plan gives you 15 messages a day at $0, Pro adds E2EE chats and unfiltered models for $19 a month, and Max is $49 a month. Start a free account before your next draft, no card needed.

Common questions

How can I tell whether a Substack publication has blocked AI training?

Add /robots.txt to the publication's address and open it. With the block on, the file has a group for GPTBot, ClaudeBot, Google-Extended, and other AI crawlers, plus a Content-Signal line ending in ai-train=no. Without it, none of those lines appear.

Does turning on Substack's AI block hurt my search ranking?

Substack's help article says blocking training "may limit your publication's discoverability in tools and search engines that return AI-generated results." The robots.txt lines the block writes, read on November 13, 2026, keep search=yes and add no rule for Googlebot or Bingbot.

Does deleting my Substack account take my posts out of AI training?

Not from models already trained. Substack's Terms of Use say deleting an account deletes its posts, but "any Post that you have made public may remain available," and it "may not be possible to completely delete your content from Substack's records or backups."

Does Substack use my subscriptions to target ads?

Substack's CCPA policy says, "we do not share your information with third parties for the purpose of cross-context behavioral advertising."

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next