How to Keep Specific Content Out of AI Answers

Every documented way to keep content out of AI answers is subtractive. You remove a piece of text, a page or a path from what an engine may show. Choose the narrowest control that does the job, because the wider ones also cost you in classic search.

The section-level option is the one most people need. Bing's October 2025 announcement says content marked data-nosnippet "is still indexed normally, but it will be excluded from snippets and AI summaries," and that it remains available for ranking. Our digest of Bing's data-nosnippet support covers the details, and the noindex vs nosnippet vs data-nosnippet comparison walks through edge cases.

Google's AI features page sets the bar for AI Overviews and AI Mode: a page must be indexed and eligible to be shown with a snippet. Remove either condition and the page is no longer eligible as a supporting link.

Be plain with your stakeholders about the limits before you ship anything.

• Training data already collected stays. A model answering without live retrieval uses what it learned before its cutoff. Nothing you change today reaches those answers. Blocking training crawlers such as GPTBot or ClaudeBot only affects future training.
• Google AI features have no robots.txt token. Google states that robots.txt rules for Googlebot are the control for AI in Search. Blocking Googlebot removes you from Google Search as well, so it is almost never the right move.
• Google-Extended is a different lever. It controls training and grounding for Gemini Apps and Vertex AI, and Google says it does not affect inclusion in Google Search. See what Google-Extended controls.
• User-triggered fetchers may ignore robots.txt. OpenAI says robots.txt rules "may not apply" to ChatGPT-User, and Perplexity says Perplexity-User "generally ignores robots.txt rules."
• Third parties can still quote you. If another site republishes your text, your controls do not apply to their page.
• Opting out of ChatGPT search is not total. OpenAI says sites that block OAI-SearchBot can still appear as a navigational link.

If your goal is a full exit rather than targeted suppression, read whether you can completely stop AI from using your content first.

Work from the narrowest control outward. Stop at the first step that covers what you need.

1. Write down exactly what must stay out. Name the sections, pages or paths, and the engines you care about. "Our internal rate card on /partners/" is a usable brief. "Sensitive stuff" is not.
2. For a section, add data-nosnippet. Google supports it on span, div and section elements.

<div data-nosnippet>Partner discount table</div>

3. For a whole page, add a snippet directive. Use nosnippet to drop all text, or max-snippet to cap the length.

<meta name="robots" content="nosnippet">

4. Use noindex only if the page should not be in search at all. It removes the page from classic results too.
5. For a specific AI search engine, disallow its search bot by path. Name the retrieval bot, not the training bot.

User-agent: OAI-SearchBot
Disallow: /partners/

PerplexityBot and Claude-SearchBot are the equivalents for Perplexity and Claude search. Do not block Googlebot or bingbot for this purpose. 6. For Google AI features across a whole property, use the Search Console control. As of October 2026 it is an opt-out: inclusion is the default, and child properties inherit the parent's setting. Our guide to Google's AI features control explains where to find it. 7. Deploy, then verify the raw HTML and the live robots.txt. Fetch the page without JavaScript and confirm the attribute or meta tag is in the served HTML. Then confirm robots.txt resolves the way you intended for each bot, for example with the robots.txt Checker for AI Crawlers.

Run this list before any suppression change goes live.

• The scope is written down: section, page, path or property.
• The narrowest control that covers it has been chosen.
• data-nosnippet wraps only the intended block, not a shared template component.
• Snippet directives sit in the server-rendered HTML, not injected later by JavaScript.
• noindex is used only on pages that should leave search entirely.
• robots.txt names retrieval bots (OAI-SearchBot, PerplexityBot, Claude-SearchBot) separately from training bots (GPTBot, ClaudeBot, Google-Extended).
• Googlebot and bingbot are untouched unless you mean to leave classic search.
• Canonical tags on suppressed pages point where you expect, since canonical mistakes can affect AI Overview eligibility.
• The Search Console AI control has been checked on the parent and every child property.
• Someone owns the change and knows when it will be reviewed.

You cannot fully confirm absence from AI answers, because retrieval runs per query and engines decide when to search. Judge the change by the signals you can check.

Good results

• The suppressed section is missing from the classic snippet, but the page still ranks for its main queries.
• The page with nosnippet is still indexed in Google Search Console and Bing Webmaster Tools.
• ChatGPT search shows the path only as a navigational link, if at all, after you block OAI-SearchBot on it.
• Every other page on the site serves the same robots meta tags it did before the change.

Bad results

• A data-nosnippet added to a shared component now hides pricing or product facts on every page. That removes the sentences engines could quote about you.
• A blanket "block AI" robots.txt rule removed OAI-SearchBot when you only meant to stop training. The block AI training but stay in AI search playbook covers that split.
• noindex was used where nosnippet would have done, and organic traffic to the page disappeared.
• A child property inherited an AI opt-out from its parent that nobody chose. See auditing inherited AI controls.

If you are worried about rankings, Bing states that data-nosnippet content remains available for ranking. We answer the wider question in whether hiding text from AI summaries hurts rankings.

Suppression drifts. Templates get reused, CMS plugins add meta tags, and properties get reorganized. Treat these controls as configuration you audit, not a one-time fix.

• After every template, theme or CMS change: re-check that data-nosnippet and robots meta tags appear only where intended.
• After any robots.txt edit: re-resolve the rules for each named bot.
• After CDN or hosting changes: confirm robots.txt is still reachable and still the file you think it is.
• When a new Search Console property or child property is added: confirm what AI setting it inherited.
• On a fixed calendar, such as quarterly: run the full checklist above, and re-read the vendor crawler pages. OpenAI, Anthropic, Perplexity and Google update their bot lists, and behavior described here is as of October 2026.

Keep a short log of each suppressed item, the control used, and the date. When someone asks why a page no longer shows a snippet, the answer should take one minute to find.