Substack block AI training: what the setting really does
Where Substack's Block AI training setting is, what it signals to AI crawlers, what it cannot stop, and whether you can edit your robots.txt yourself.
By Crawlito · 8 min read
In short
- Substack's help article describes the AI training setting as a signal to tools such as ChatGPT, Claude and Gemini, and says it only affects tools that respect it.
- Substack's own article warns that blocking training may limit discoverability in tools and search engines that return AI-generated results.
- In a Substack publication's robots.txt that we fetched on 11 November 2026, the AI block named training crawlers such as GPTBot and ClaudeBot but not OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot or Bingbot.
- We found no Substack documentation of a way to edit robots.txt yourself, so the setting is the control you have.
- A setting that blocks training does not make paid posts private: what a crawler can read depends on what Substack serves to a logged-out visitor.
On this page
- What is the Block AI training setting on Substack?
- Where do you turn it on?
- What does it ask AI tools to do?
- Can you edit robots.txt on Substack?
- Does blocking training affect visibility in AI search?
- How do you check what Substack serves?
- What protects paid posts?
- Frequently asked questions
- Check what your publication serves
The Block AI training setting in Substack is a publication-level switch that tells AI tools not to train their models on your posts. Substack's help article says it works only for tools that respect the setting, so it is a request, not a lock. It sits in your publication's Settings page, and it is the only AI-crawler control we found: we found no documented way to edit your own robots.txt.
This post explains where to find the setting, what it appears to change on the public side, what it does not cover, and how to check your own publication. Facts about Substack come from its help article on blocking AI training and from robots.txt files we fetched on 11 November 2026. Substack changes its interface, so treat labels and locations as "at the time of writing".
What is the Block AI training setting on Substack?
Substack describes it as a way to signal to AI tools such as ChatGPT, Claude and Gemini that their models should not train on the content you publish. The article is explicit about two limits:
- It only applies to AI tools that respect the setting. A tool that ignores it is not stopped.
- Blocking training may limit your publication's discoverability in tools and search engines that return AI-generated results.
Read that second point carefully. It is Substack's own warning, and it is a reason to know exactly what you are switching on. We cover the difference between training and search below.
The control has gone by more than one name. At the time of writing, Substack's help article labels it "Tell AI tools not to train their models on your content", while older posts and third-party guides call it "Block AI training" and some place it under Publication details. This post uses "Block AI training" because that is the name people search for.
Where do you turn it on?
Substack's help article says to open your publication's Settings page and look in the Privacy section for the toggle about AI training. The exact label may differ, as noted above. Some third-party guides place it under Publication details instead, so if you do not see it where this post says, scan the whole Settings page for the words "AI" or "training".
- Open your publication's dashboard and go to Settings.
- Find the toggle that tells AI tools not to train on your content.
- Switch it on and save.
- Wait, then check your robots.txt (see the section on checking below).
Do not assume its current position. We could not verify what the default is, so look at the toggle instead of trusting a forum comment, including this one.
What does it ask AI tools to do?
Substack's article calls it a signal. A signal only works on crawlers that read and honor it, and robots.txt is the standard place for a crawler to find one. On 11 November 2026 we fetched robots.txt from a Substack publication (boodsy.substack.com), which served a file with a block of named AI groups and a Content-Signal line. The substack.com and on.substack.com files had no such block. We cannot see the publication owner's settings, so we cannot prove which toggle position it had. The difference is consistent with the setting, and the excerpt below is what the publication served:
User-agent: GPTBot
Disallow: /
User-agent: ClaudeBot
Disallow: /
User-agent: anthropic-ai
Disallow: /
User-agent: Claude-Web
Disallow: /
User-agent: CCBot
Disallow: /
User-agent: Bytespider
Disallow: /
User-agent: Applebot-Extended
Disallow: /
User-agent: Google-Extended
Disallow: /
User-agent: *
Disallow: /action/
Disallow: /subscribe
Content-Signal: search=yes, ai-input=yes, ai-train=noThat is an excerpt, not the full file. The same file also named other crawlers such as Amazonbot, meta-externalagent, cohere-ai and AI2Bot, and kept the platform's normal * rules for paths like /publish and /sign-in. The substack.com and on.substack.com files, which had no AI block, held only the platform's normal rules for *, plus BLEXBot, Twitterbot and facebookexternalhit.
Three things follow from the excerpt.
It names training crawlers. GPTBot is OpenAI's training crawler, ClaudeBot is Anthropic's, and CCBot is Common Crawl's. Google-Extended and Applebot-Extended are opt-out tokens, not separate crawlers. OpenAI documents that disallowing GPTBot signals its content should not be used for training, and Anthropic documents the same for ClaudeBot. See GPTBot, OAI-SearchBot and ChatGPT-User and ClaudeBot, Claude-SearchBot and Claude-User.
It includes older tokens. anthropic-ai and Claude-Web are not in Anthropic's current crawler documentation. They are harmless to keep, but only ClaudeBot is the documented training token.
The Content-Signal line is a stated preference. It says ai-train=no while saying search=yes and ai-input=yes. Cloudflare's documentation says robots.txt compliance is voluntary and the file does not technically prevent access. Anthropic says it has begun recognizing a machine-readable "no AI training" signal based on an IETF draft, but its page does not name this Content-Signal line, and OpenAI's crawler page does not mention content signals. Our guide to Content Signals in robots.txt explains the three values.
Who may ignore it?
Any crawler that does not follow robots.txt. Anthropic says its crawlers respect robots.txt, and OpenAI documents robots.txt as the control for GPTBot and OAI-SearchBot, but robots.txt is voluntary and nothing in the file enforces it. OpenAI also says robots.txt rules may not apply to ChatGPT-User, which acts on a user's request. See do AI crawlers respect robots.txt for what the evidence shows.
Can you edit robots.txt on Substack?
We found no Substack documentation that lets you edit your publication's robots.txt, and the files we fetched shared the same platform rules for *, with the AI block as the visible difference. We did not test a publication on a custom domain, so we cannot say whether it serves a different file.
In practice this means:
- As far as we found, you cannot add a
Disallowfor a crawler Substack does not name. - You cannot remove a rule Substack adds.
- You cannot add your own
Content-Signalline. - We found no setting for
X-Robots-Tagresponse headers either.
If you need that level of control, you need a host where you do, and that is a bigger decision than this setting. For the mechanics of editing the file where you can, read robots.txt for AI crawlers.
Does blocking training affect visibility in AI search?
Substack warns that it may. Here is what we can say about it with evidence.
AI companies document training and search as separate jobs with separate controls. OpenAI says its GPTBot and OAI-SearchBot settings are independent, and that a site blocked from OAI-SearchBot will not appear in ChatGPT search answers, though it may still show up as a navigational link. Anthropic says blocking Claude-SearchBot or Claude-User may reduce your visibility in Claude's search results. Our guide to training, search and user crawlers lays out the split.
In the Substack publication file we fetched, none of OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot or Bingbot had its own group. Under RFC 9309 a crawler without a group of its own follows the * group, and that group only blocks platform paths such as /publish. So on the evidence we have, the setting does not tell the main search crawlers to stay away from your posts.
Why, then, the warning? We do not know, and we found no explanation of it from Substack. Do not read it as a statement that search crawlers are blocked, and do not read the robots.txt evidence as a promise that nothing changes. For the general policy, allow AI search, block AI training shows the split in a file you would write yourself.
Being readable by a search crawler is not the same as being quoted in an answer. Which sources an assistant uses depends on many things outside your publication.
How do you check what Substack serves?
Do not rely on the toggle's position. Fetch the file crawlers actually receive.
curl -s https://yourname.substack.com/robots.txt | grep -A1 -E "^User-agent: (GPTBot|ClaudeBot|CCBot)|Content-Signal"If you use a custom domain, fetch it from that domain, since crawlers will:
curl -s https://example.com/robots.txt | grep -A1 -E "^User-agent: (GPTBot|ClaudeBot|CCBot)|Content-Signal"What to look for:
| Result | What it means |
|---|---|
User-agent: GPTBot followed by Disallow: / | The training opt-out is published for that crawler |
A Content-Signal line with ai-train=no | A stated preference about training |
| No matches | The AI block is not in the file the crawler gets |
| An error status | Check the address and your domain setup |
Crawlers cache robots.txt, so a change can take time to be noticed. OpenAI says its search systems can take about 24 hours to adjust after a change, and our post on how long robots.txt changes take covers the rest.
What protects paid posts?
The setting is about training, not about access control. A paid post is protected by what Substack sends to someone who is not logged in, and we found no Substack documentation saying the AI setting changes that.
We found no official Substack page that states what a crawler receives for a paid post, and the accounts we found from third parties disagreed with one another. So test it rather than guess:
curl -s https://yourname.substack.com/p/your-paid-post | grep -c "a sentence from the paid part"A count of 0 means that sentence was not in the response an anonymous request received. A count above 0 means it was served publicly. Repeat the test with a sentence from the free preview to confirm the check works.
Anything an anonymous request can read, a crawler that ignores your preferences can read too. Our guide to paywalls and AI crawlers explains how different paywall designs look to a crawler. If a post must not be readable, do not rely on a training opt-out: do not publish it publicly.
Frequently asked questions
Does Substack's Block AI training setting stop ChatGPT from reading my newsletter?
Not on its own. Substack's article says it only affects AI tools that respect the setting. GPTBot, OpenAI's training crawler, is disallowed in the publication file we fetched. ChatGPT-User acts on a user's request, and OpenAI says robots.txt rules may not apply to it.
Will turning it on remove me from AI search?
Substack warns it may limit discoverability in tools and search engines that return AI-generated results. In the publication robots.txt we fetched, the main search crawlers had no rule of their own. We cannot tell you how each AI product treats you, so check the file and keep your posts publicly readable if you want them found.
Can I add my own robots.txt rules on Substack?
We found no documentation that says you can. The setting is the control Substack gives you. If you need per-crawler rules, headers or logs, you need a host where you control the server.
Is my old content already in training data?
The setting is a signal about future use. Anthropic's help article says blocking ClaudeBot signals that your future materials should be excluded from training datasets. We found no operator documentation that promises removal of content already collected.
Check what your publication serves
Run the free check on one of your post URLs: it reads the robots.txt rules for each named AI crawler and shows what the page returns, with no account needed. It runs from Crawlito's servers, so it shows what is published, not what any crawler did. If you run a site where you control the server, an audit checks up to 500 pages.
Filed under Platform guides. Topics: substack, ai-crawlers, robots-txt, ai-training