research withRon

Technical AEO

GPTBot is not a search crawler. Here is what to check instead.

Search access, user-requested retrieval, and model training are separate policy choices. Your robots.txt should treat them that way.

Harsh Rana·May 21, 2026·8 min read

The short answer

Check OAI-SearchBot, Claude-SearchBot, PerplexityBot, ChatGPT-User, Claude-User, and Perplexity-User for search or user retrieval; govern GPTBot and ClaudeBot separately as potential model-training access.

3 jobs

Search discovery, user-requested retrieval, and model training need separate access decisions

A broad robots.txt rule can block a provider's search or user-retrieval agent while ordinary Google indexing still looks healthy. The error stays invisible unless you inspect the policy, delivery logs, or answer-level visibility.

The important correction is that AI user agents do not all do the same job. A training opt-out is not a search opt-out. The current provider documentation names separate agents, and a responsible audit keeps those choices separate.

The documented agents in this check

Use current provider documentation as the source of truth. This table covers the agents Ron actively checks as of August 2026.

User-AgentVendorPurposeType
GPTBotOpenAIPotential use to improve generative modelsTraining
OAI-SearchBotOpenAIChatGPT search discovery and answer linksSearch
ChatGPT-UserOpenAISome actions taken at a user's requestUser retrieval
ClaudeBotAnthropicModel developmentTraining
Claude-SearchBotAnthropicImproves search-result qualitySearch
Claude-UserAnthropicAccess directed by a userUser retrieval
PerplexityBotPerplexityBuilds and refreshes its search indexSearch
Perplexity-UserPerplexitySupports user-requested page accessUser retrieval

The one distinction that actually matters: training vs. search-and-answer

Most confusion starts with one blended AI crawler score. There are at least three policy jobs here: search discovery, user-directed retrieval, and potential model training.

Training crawlers

GPTBot and ClaudeBot are documented for potential model improvement or development. Blocking them is a data-use decision. It should not lower a search-access score, and allowing them should not be sold as a direct citation or ranking tactic.

Search and answer crawlers

OAI-SearchBot, Claude-SearchBot, and PerplexityBot are documented for search discovery. ChatGPT-User, Claude-User, and Perplexity-User support forms of user-requested access. Blocking them can limit page-content access in those workflows, but permission still does not guarantee crawling, selection, citation, or recommendation.

Blocking GPTBot is a training choice. Blocking OAI-SearchBot is a search-access choice. One score should never hide that difference.
Ron

How to check your robots.txt right now

Your robots.txt lives at yourdomain.com/robots.txt. Open it in a browser. You are looking for two things.

  1. A wildcard block that catches everything: a User-agent: * line followed by Disallow: / or a broad path. If this exists without explicit Allow rules for specific bots above it, every crawler including AI bots is blocked.
  2. Named agent blocks: search for all eight documented names above and classify each block by purpose before changing it.
  3. Group specificity: a matching named user-agent group is evaluated instead of a wildcard group for that agent. Test the effective rule rather than relying on visual order.
  4. Empty or missing file: a missing robots.txt means no restrictions at all, which is generally fine. The problems come from files with rules.

Common ways sites end up accidentally blocked

The corrected robots.txt snippet

This starter expresses one common policy: allow documented search and user-retrieval agents while blocking documented training agents. Review current provider docs, legal requirements, and your own publishing policy before use.

robots.txt: separate access by purpose

# Search and user-requested retrieval
User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Perplexity-User
Allow: /

# Potential model training
User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

# Real private paths still require authentication
User-agent: *
Disallow: /account/
Disallow: /checkout/

An Allow rule removes one access barrier. It does not guarantee a fetch, index entry, citation, recommendation, or referral visit. robots.txt is also public and is not access control for private information.

The trade-off worth thinking through

The training decision belongs with the organization that owns the content and risk. The search decision belongs with the distribution strategy. Private, licensed, user-generated, and expensive application routes may also need path-specific handling beyond a simple sitewide choice.

For public pages intended to appear in AI search or user-requested answers, unintended access blocks are worth fixing. Preserve the reason, owner, provider source, test URLs, and next review date so the policy stays accountable.

8

current OpenAI, Anthropic, and Perplexity agents included in Ron's documented policy check

One silent failure worth knowing about

A robots.txt block does not produce a universal alert. The most reliable process is to test the effective rules, inspect CDN and origin behavior, and then measure answer citations or recommendation context separately.

That is the core reason to run an audit now rather than assuming your current configuration is fine. The defaults from 2022 or 2023 almost certainly did not account for OAI-SearchBot or PerplexityBot. They did not exist yet.

What to do next

Start by pulling up your robots.txt. Scan it for the user-agents in the table above. If you find a Disallow on any of the search-and-answer bots, fix it using the snippet above. If the file looks clean at the robots.txt level, check your server-side WAF rules and any security plugins for bot filtering logic that might be operating separately.

If you want a faster read without digging through files manually, the free AI Crawler Checker tool linked below fetches your robots.txt and highlights any rules affecting known AI crawlers. It takes about ten seconds and flags training-only blocks separately from search-and-answer blocks so you know exactly what you are dealing with.

Questions

Does blocking GPTBot block ChatGPT search?

OpenAI documents GPTBot and OAI-SearchBot separately. GPTBot governs potential model-training access, while OAI-SearchBot is the agent to review for ChatGPT search discovery and summaries.

Why is Claude-User in the list?

Anthropic documents Claude-User for page access directed by a user, separately from Claude-SearchBot and ClaudeBot. Older crawler lists often omit it.

How often should I check my robots.txt for new AI crawlers?

Every six months is a reasonable cadence right now given how fast the space is moving. New products from established AI labs often ship a new crawler user-agent with little fanfare. Checking once a quarter is not excessive if you are serious about AI search visibility.

What if my robots.txt allows these bots but a firewall blocks them at the IP level?

The robots.txt rule is irrelevant if the request never reaches your server. Firewall or WAF blocks that drop traffic from known AI crawler IP ranges override robots.txt entirely. You would need to allowlist those IP ranges in your firewall settings, or check whether your WAF has a named rule for AI bots that can be toggled off.

R

Harsh Rana

I build Ron at 617 Software Studio, a small Boston shop. I run real AI visibility audits by hand and pour what I learn into how Ron works. These notes come from the actual reports, not a content brief. More about Ron.

Keep going

Sources

Find out what AI actually says about you.

~5 min scan · $39 · refunds if useless

Run my audit →