agentspeed.
← all tools
Tool

Agent UA tester: see your site through five AI crawlers

Many sites cloak: serve a static shell to bots, real content to humans. Agents see the shell. Result: hallucinated answers about your product. We fetch your URL six times in parallel, once each as AgentSpeedBot, GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot, and a vanilla desktop Chrome, then diff the responses.

Tool
›
https://openai.com2 signals
Cloaking suspected

Browser sees substantially different content vs the baseline (similarity below 0.80). Agents that get the diverged variant will not represent the site accurately.

Agent blocked

AgentSpeedBot (403), GPTBot (403), ChatGPT-User (403), PerplexityBot (403), ClaudeBot (403) cannot fetch the page. Check your WAF, bot manager, or robots.txt for explicit blocks.

agentstatusbyteslatencysimilaritycontent-typebody sha256 (16)
AgentSpeedBot4039,93641ms1.00text/htmlfc7a04df66e5b358
GPTBot4039,97852ms1.00text/html5b4b7a5b3f87c93d
ChatGPT-User40310,08542ms1.00text/html6f99c0315f7828a6
PerplexityBot40310,00044ms1.00text/htmlbba600e8ac0e7d5f
ClaudeBot4039,97841ms1.00text/html629015e96cc3a483
Browser2001,950,925175ms0.00text/htmlb082333175d87f6e

Parsed head hints

agent<title>canonicalmeta robots
AgentSpeedBot———
GPTBot———
ChatGPT-User———
PerplexityBot———
ClaudeBot———
BrowserOpenAI | Research & Deploymenthttps://openai.com/index, follow
Result summary

Copyable, paste-ready text summary. Useful for sharing with infra owners or filing an issue with your CDN provider.

AgentSpeed UA tester: https://openai.com
Run 2026-07-21T06:25:53.376Z

Diff matrix:
  - AgentSpeedBot    status=403 bytes=9936 similarity=1.00 sha256=fc7a04df66e5b358
  - GPTBot           status=403 bytes=9978 similarity=1.00 sha256=5b4b7a5b3f87c93d
  - ChatGPT-User     status=403 bytes=10085 similarity=1.00 sha256=6f99c0315f7828a6
  - PerplexityBot    status=403 bytes=10000 similarity=1.00 sha256=bba600e8ac0e7d5f
  - ClaudeBot        status=403 bytes=9978 similarity=1.00 sha256=629015e96cc3a483
  - Browser          status=200 bytes=1950925 similarity=0.00 sha256=b082333175d87f6e

Parsed head hints:
  - AgentSpeedBot: title=— canonical=— meta-robots=—
  - GPTBot: title=— canonical=— meta-robots=—
  - ChatGPT-User: title=— canonical=— meta-robots=—
  - PerplexityBot: title=— canonical=— meta-robots=—
  - ClaudeBot: title=— canonical=— meta-robots=—
  - Browser: title=OpenAI | Research & Deployment canonical=https://openai.com/ meta-robots=index, follow

Signals:
  [FAIL] Cloaking suspected: Browser sees substantially different content vs the baseline (similarity below 0.80). Agents that get the diverged variant will not represent the site accurately.
  [FAIL] Agent blocked: AgentSpeedBot (403), GPTBot (403), ChatGPT-User (403), PerplexityBot (403), ClaudeBot (403) cannot fetch the page. Check your WAF, bot manager, or robots.txt for explicit blocks.

Run again at: https://agentspeed.com/tools/agent-ua-tester

Agents we send

AgentSpeedBot
Our scanner (baseline)
AgentSpeedBot/1.0 (+https://agentspeed.com/bot)
GPTBot
OpenAI training crawler
Mozilla/5.0 (compatible; GPTBot/1.2; +https://openai.com/gptbot)
ChatGPT-User
ChatGPT browsing on behalf of a user
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; ChatGPT-User/1.0; +https://openai.com/bot
PerplexityBot
Perplexity citation crawler
Mozilla/5.0 (compatible; PerplexityBot/1.0; +https://perplexity.ai/perplexitybot)
ClaudeBot
Anthropic crawler
Mozilla/5.0 (compatible; ClaudeBot/1.0; +claudebot@anthropic.com)
Browser
Vanilla desktop Chrome (control)
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/132.0.0.0 Safari/537.36
Watch this URL — get alerted when this changes

Turn this one-off check into a scheduled watch. We re-run it, diff the result, and alert you on a regression through the same channels as your agent alerts. Watches are a paid feature.

Manage watches at /watches. You’ll be asked to sign in if you haven’t already.

Examples

Click an example to run the tool with pre-filled inputs.

Why this matters for AI agents

Agents that get a different response than your real users misrepresent you. A 403 means the agent has nothing to cite. A redirect to /login means agents see only your auth wall. A static shell means agents read your nav menu and footer instead of your product copy.

AgentSpeed runs an end-to-end version of this diff as part of the journey.primary_probe check on every public scan.

Common fixes
AI bots redirected to /login
Render server-side preview content for unauthenticated requests on key landing pages: pricing, docs, marketing. Agents need a cite-able shell, not the auth wall.
Cloudflare challenges block GPTBot
In Cloudflare bot management, allow GPTBot, ClaudeBot, and PerplexityBot explicitly. Their UA strings are documented and the IPs are fixed.
Static shell with no copy
Server-side render the hero copy, value props, and pricing. JS-only sites force agents to read a near-empty body and infer everything from the URL.
Different titles per UA
If <title> differs across agents, your CMS or edge layer is rewriting metadata for bots. Audit any "agent-aware" rendering rules, since agents often get the worst variant.
Canonical URL drift
Canonicals should not differ between UAs. If your site adds tracking parameters per-bot, agents will see splintered URLs and dedupe poorly.
Linked rubric checks
FAQ
Is cloaking against Google’s policy?
Serving different content to Googlebot than to human users violates Google’s Webmaster Guidelines. Serving different content to AI agents is not formally policed by Google but tends to produce worse outcomes, because agents that get a static shell hallucinate answers based on partial information.
Why does my site differ for GPTBot?
Common causes: bot management products (Cloudflare, Akamai) issuing challenges; CMS rules that strip personalisation for bots; CDN routes that serve a marketing variant. Test in this tool to see exactly which agents get which variant.
What is a normal similarity score?
Above 0.95 is essentially identical. Below 0.80 means meaningful drift. A score under 0.50 between bots and the browser usually means cloaking or a content firewall.
Will fixing cloaking improve agent answers?
Yes. Agents that get richer body content cite more accurately. Sites that ship the same SSR shell to bots and users tend to outperform in agent-cited answers.
Do you store the responses?
No. Each run is fetched live. Nothing is persisted. The result page is rendered server-side and discarded.
Run a full agent-readiness scan
Full scan

Score the entire site against 30 agent-readiness checks across discoverability, readability, structured data, actionability, and performance.

Get the full report for openai.com →