/ 7 min read

AI Visibility Checker: Are You Showing Up in ChatGPT & Perplexity?

If a buyer asks ChatGPT or Perplexity for "the best [your category]" and your site never comes up, you are invisible in the channel that increasingly decides who gets the click. The good news: you can test this today. There are two reliable ways to check AI visibility — prompt the assistants directly and read who they cite, and scan your own site for the technical signals that let AI crawlers reach, parse, and trust your content. This page walks through both, shows you exactly what blocks visibility, and gives you a do-it-now checklist.

Two ways to check: prompt-testing vs. readiness scanning

"AI visibility" splits into two distinct questions, and you need both answers. The first is an outcome question: when people ask AI assistants about your topic, do you get mentioned or cited? The second is a cause question: can AI systems actually crawl, extract, and trust your pages in the first place?

Prompt-testing answers the outcome question. You ask ChatGPT, Perplexity, Google AI Overviews, and Gemini real buyer questions and record whether your domain appears in the answer or its source links. It is direct evidence, but it is noisy — results vary by phrasing, account, region, and date, and the platforms are opaque about why they cite what they cite.

Readiness scanning answers the cause question. It inspects the mechanics: are AI crawlers allowed in your robots.txt, is your main content in the HTML or hidden behind JavaScript, do you have clean structure and schema. This is what you actually control. If you fail the cause checks, no amount of prompting will fix the outcome — fix the plumbing first, then re-test prompts over the following weeks.

Manual prompt-testing: how to do it right

Run the same set of prompts across all four major surfaces and log the results in a spreadsheet. The goal is a repeatable baseline you can re-check monthly, not a one-off impression.

  • Use buyer-intent prompts, not your brand name — Ask the questions your customers actually ask — "best CRM for small law firms," "how do I clean a saltwater tank" — not "tell me about [your brand]." Branded prompts almost always surface you and tell you nothing.
  • Test all four surfaces — ChatGPT (with web search on), Perplexity, Google AI Overviews (run the query in Google), and Gemini. Each uses different crawlers and citation logic, so visibility on one does not imply the others.
  • Turn on sources and read the citations — Perplexity lists numbered sources; ChatGPT and Gemini show link cards when search is active; AI Overviews show source chips. Note whether your domain appears, at what rank, and which specific page is cited.
  • Vary the phrasing — Run three to five wordings per topic. A single prompt is anecdote; a cluster of prompts is a signal. Watch for competitors who appear consistently — that is who is winning the citation.
  • Log the date and re-run monthly — Answers drift as indexes refresh. A dated log lets you tie visibility changes to fixes you shipped.

What actually blocks AI visibility

When prompt-testing shows you are absent, the cause is almost always one of these. Most are invisible in a browser and only show up when you inspect the raw response an AI crawler receives.

  • robots.txt blocks the AI crawlers — The biggest silent killer. AI systems use named user-agents — GPTBot and OAI-SearchBot (OpenAI), PerplexityBot (Perplexity), Google-Extended (Gemini training/grounding), and ClaudeBot (Anthropic). A Disallow rule for any of these — sometimes added by a plugin or a privacy-minded teammate — removes you from that platform entirely. Allowing Googlebot does not allow these.
  • Your content is rendered by JavaScript — Many AI fetchers grab the raw HTML and do not execute JS the way a browser does. If your main text only appears after client-side rendering, the crawler may see an empty shell. Server-side rendering or static HTML is the safe bet. View source (not the inspector) and confirm your actual content is in the markup.
  • Weak structure and no extractable answers — AI systems lift self-contained passages — a clear H1, descriptive H2s, short answer-first paragraphs, and real Q&A. Walls of text, vague headings, and content that needs the whole page for context are hard to quote, so they get skipped in favor of cleaner sources.
  • Missing structured data — Schema.org JSON-LD (Article, FAQPage, Product, Organization) gives machines unambiguous facts about what a page is and who published it. It is not a magic ranking lever, but it reduces the guesswork that makes a crawler pass you over.
  • Thin authority and entity signals — AI tends to cite sources it can corroborate. Consistent brand naming, an author/Organization identity, and being referenced elsewhere all raise the odds you are treated as a citable source rather than noise.

Do-it-now checklist

Work top to bottom. The first three items are pass/fail gates — if you fail them, nothing downstream matters.

  • 1. Confirm AI crawlers are allowed — Open yourdomain.com/robots.txt and search for GPTBot, OAI-SearchBot, PerplexityBot, Google-Extended, and ClaudeBot. Any Disallow: / under those agents (or a blanket block) is a problem to fix now.
  • 2. Verify content is in the raw HTML — Right-click your top page, View Source, and Ctrl+F for a sentence of your main body text. If it is not there, your content is JS-dependent and likely invisible to fetchers — move to server-side or static rendering.
  • 3. Check sitemap and HTTPS — Confirm /sitemap.xml exists and lists your real pages, and that the site loads over valid HTTPS. These are baseline discoverability requirements.
  • 4. Make answers extractable — On key pages, lead with a one- to two-sentence direct answer under a clear heading, use descriptive H2s, and add a real FAQ with FAQPage schema.
  • 5. Add structured data — Ship Article/Product/Organization JSON-LD with accurate author, name, and description fields. Validate it.
  • 6. Consider an llms.txt — A root-level /llms.txt is an emerging convention that points AI to your most important pages in Markdown. No major platform has officially committed to it and it is low-cost insurance, not a silver bullet — add it, but do not expect it alone to move citations.
  • 7. Re-run your prompt tests in a few weeks — Crawlers need to re-fetch and indexes need to refresh. Treat AI visibility as a slow loop: fix causes, wait, re-measure outcomes.

An llms.txt starter you can copy

If you want to ship the llms.txt from the checklist, here is a minimal valid file. Only the H1 title is strictly required; the blockquote summary and section links make it more useful. Place it at the root of your domain (yourdomain.com/llms.txt).

# Your Brand Name

> One-sentence summary of what your site is and who it helps.

## Main
- [Homepage](https://yourdomain.com): What visitors can do here
- [Pricing](https://yourdomain.com/pricing): Plans and costs

## Guides
- [Getting Started](https://yourdomain.com/start): Step-by-step setup
- [FAQ](https://yourdomain.com/faq): Common questions answered

## Optional
- [Changelog](https://yourdomain.com/changelog): Release history

Be honest about what is knowable

AI search is new and the platforms are deliberately opaque. No one outside these companies knows the exact weighting that decides citations, and the rules change without notice. Anyone selling you a guaranteed-citation formula is guessing.

What you can know with confidence is the controllable layer: whether AI crawlers can reach your pages, whether your content is in the HTML they fetch, whether your structure and schema make passages easy to extract, and whether your authority signals are coherent. Get those right and you are in the consideration set. The rest is iteration — measure outcomes by prompt-testing, fix causes, and re-measure.

See your AI search readiness score

Manually testing robots.txt, raw HTML, sitemaps, schema, and prompt results across four AI surfaces is doable but tedious — the free Am I Citable scanner does the cause-side checks in one pass. Enter your URL and it inspects whether the major AI crawlers (GPTBot, OAI-SearchBot, PerplexityBot, Google-Extended, ClaudeBot) are allowed, whether your content and structure are extractable, whether you have structured data and an llms.txt, and how strong your citation signals are. It returns a 0-100 AI readiness score with prioritized fixes and can auto-generate an llms.txt file for you to drop in. Run it before you spend another month wondering why you are invisible in AI answers — fix the plumbing the scan surfaces, then re-run your prompt tests to watch visibility improve.

Run the Free Scan

FAQ

Check two things. First, open yourdomain.com/robots.txt and confirm you are not blocking the AI user-agents — GPTBot and OAI-SearchBot for ChatGPT, PerplexityBot for Perplexity. Second, View Source on a key page and confirm your main text is in the raw HTML, since many AI fetchers do not run JavaScript. If a crawler is blocked or the content is JS-only, the assistant cannot see or cite you regardless of how good the content is. An automated scan checks both at once and flags exactly which bots are blocked.

Google ranking and AI citation use different crawlers and different logic. Googlebot may be fully allowed while Google-Extended, GPTBot, OAI-SearchBot, PerplexityBot, or ClaudeBot are blocked — so the assistants never ingest you. AI systems also favor self-contained, answer-first passages with clear structure and schema, which is a different bar than traditional rankings. Verify AI-crawler access first, then improve extractability and structured data.

On its own, no. llms.txt is an emerging convention with no official commitment from the major AI platforms, so treat it as low-cost insurance rather than a citation lever. The fixes that reliably matter are unblocking AI crawlers, serving content in the raw HTML, structuring answer-first passages, and adding accurate schema. Add an llms.txt after those are handled.

Related Guides