AI visibility
You can vanish from ChatGPT without breaking anything
No alert, no page in error, no change in Google. Just one line in robots.txt, copied from a blog post, and three weeks later the traffic that came from answer engines no longer exists.
“I blocked GPTBot so as not to feed training. I didn't know a different bot was building the answers.”
Since 2024, OpenAI and Anthropic each run two bots.
- bots tracked
- 14
- answer engines
- 3
- language models used
- 0
- on the Ship Score
- −25
bots tracked
answer engines
language models used
on the Ship Score
What we look at
01Who is allowed in, bot by bot
Your robots.txt is read for each of the bots answer engines run today, with Allow rules and wildcards, the way a real crawler applies them.
02The Google trap everyone misses
Google-Extended looks like an answer-engine token.
03What the page contains before JavaScript
An answer engine reads the HTML your server sends.
04Something worth citing
Readable structured data, a unique title, a consistent canonical, and the llms.txt file if you have one.
What we're often asked
Is this yet another tool guessing what ChatGPT says?
No, and that's the difference that matters. Citation-tracking tools query the models with hundreds of questions and count how often your name comes up — a measure that shifts from one day to the next for reasons nobody controls. Here everything is deterministic: we read your robots.txt, your headers and your HTML. No language model is called, and two checks on an unchanged site give the same verdict.
Do you impersonate those bots?
The header we send carries the bot's token, so your rules apply as they would to it, and ours, so you can find us in your logs. And we say what that doesn't prove: firewalls verify real bots by reverse DNS, so a refusal obtained this way says nothing about the fate of the official bot. We flag it without making it a failure.
Will blocking training cost me points?
Never. It's your decision, it's the majority one, and the screen writes it as a choice — not in red. Only the bots that build the answers, and those a human triggers by pasting your link into a conversation, count in the score.
Does it work on the free plan?
The verdict and its score, yes, on every plan, and the check runs on every deploy. The bot-by-bot detail, the deciding robots.txt line and the alert when an engine closes require Pro.
Paste your site's URL. PostShip checks it in front of you, no account and no card.