How to Check If AI Engines Have Indexed Your Website (2026)

"Indexed" is the wrong question for AI engines — the question that actually predicts citations is retrievability: whether a given engine will surface or cite your specific URL when a relevant prompt is asked, and in 2026 that is something you verify per page and per engine, not something you look up in a single status panel.

For twenty years, "is my site indexed?" had a clean answer. You opened Google Search Console, you saw which URLs were indexed, and you could run site:yourdomain.com to confirm. AI answer engines broke that model. Most of them do not maintain a public, queryable index you can inspect. ChatGPT and Perplexity retrieve at answer time from a search layer (both lean heavily on Bing-class web data). Claude fetches and reasons over content in ways it does not expose. Google AI Overviews sit on top of Google's existing index but apply their own source-selection logic. There is no universal "index status" to check — so the honest version of the question becomes: will this URL get retrieved and cited when it matters?

That distinction is not pedantic. It changes what you measure and what you fix.

Why this is not Google indexing

Google's index is a stored, mostly-stable representation of the web that ranks pages for a query. An AI answer engine does something different: it retrieves a handful of passages at answer time, re-ranks them on its own relevance and structure criteria, and synthesizes a response that may cite some, all, or none of them. Being in the underlying index is a necessary condition, not a sufficient one.

This is why ranking #1 on Google no longer guarantees an AI citation. Semrush's enterprise study of 2,855 keywords found that AI Overviews appeared on 13.14% of analyzed Google queries by March 2025, up sharply from the prior year — and crucially, the URLs cited inside those Overviews frequently were not the #1 organic result. Google's own Search Central guidance on AI features states there is no separate markup or submission step to appear in AI experiences; eligibility flows from the same crawlable, useful content that powers regular Search. In other words, you cannot opt in — and you cannot assume your organic ranking carries over. The selection happens downstream, on criteria you do not control.

The Princeton "GEO: Generative Engine Optimization" paper made the same point empirically: content optimized for classical search ranking is not automatically the content that generative engines choose to cite, and source-level structure (statistics, quotations, clear citations) measurably shifts which passages get pulled. Different system, different rules.

The manual check: spot-checking one page

You can verify a single page by hand. The general technique is the same across engines:

1. Pick a target page and pull a unique string from it — a quoted phrase, a named statistic, a product name — that should only appear on that URL.
2. Ask the engine a question that page should answer, or query it with that unique quoted phrase, and watch whether the engine surfaces or cites your URL in the response.
3. Repeat with a second phrasing. Retrieval is probabilistic; one miss is not proof of absence, and one hit is not proof of reliable coverage.

If the engine cites your URL — or reproduces a passage that could only have come from it — you have direct evidence it is both retrievable and being selected. If it never does across several phrasings, that is a signal to investigate the structural prerequisites below.

Two cautions. First, the naive operator tricks have decayed. Some URL- and site-style operators that worked in 2024 have quietly stopped doing anything useful on certain engines in 2026; an engine that ignores the operator and answers from general knowledge will happily produce a confident answer that tells you nothing about retrieval. Second, single-phrase checks are noisy. Treat any one query as a data point, not a verdict — which is exactly why doing this by hand across a whole site does not scale.

State of play, engine by engine

The engines are not equally checkable. Being honest about where verification is reliable and where it is genuinely unsolved is the whole point — overclaiming here is how you end up trusting a number that means nothing.

Engine	Can you verify retrieval today?	Why
ChatGPT	Yes — most reliable	Web-browsing answers expose links/citations, and behavior is consistent enough to test a specific URL repeatedly.
Perplexity	Yes	Cites its sources inline by design, so whether your URL was retrieved is directly observable in the answer.
Claude	Hard / unreliable	Fetches and reasons over content without exposing retrieval in a way that makes single-URL verification dependable. The obvious methods do not reliably work.
Google AI Overviews / Gemini	Murkiest — largely unresolved	Google's search index is not the same as AI Overview retrieval. Being indexed in Search tells you little about whether an Overview will cite you, and the surface is inconsistent across queries, accounts, and regions.

ChatGPT and Perplexity are the two engines where a careful person can get a trustworthy answer about a specific page — each has its own per-engine playbook, from the step-by-step method for verifying ChatGPT retrieval to the equivalent process for confirming Perplexity has picked up your pages. Claude is a genuinely hard problem: the methods that work elsewhere do not transfer cleanly. Google AI Overviews and Gemini are the open frontier — Google's documentation is explicit that there is no AI-Overview-specific index to inspect, so "I'm indexed in Google" and "I get cited in AI Overviews" are two different facts, and only the first is easy to check.

Structural prerequisites: what makes a page retrievable at all

Before you worry about citations, confirm the page can even be seen and parsed. Treat this as a checklist — miss the first two items and the rest are moot.

• AI crawler access. Open your robots.txt and confirm you are not blocking the agents that feed these engines: GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot (Perplexity), and Google-Extended (which governs Gemini/Vertex use of your content). A single overzealous Disallow is the most common self-inflicted reason a page is invisible. Crawler access is a prerequisite, not a guarantee — but its absence is a guaranteed failure.
• Server-side rendering vs JS-only content. If your meaningful content only appears after client-side JavaScript executes, assume retrieval agents may never see it. Many AI crawlers do not render JS the way Googlebot does. Server-side render or pre-render the content you want cited.
• llms.txt. An emerging, optional convention: a plain-text llms.txt at your domain root that points engines at your most important, clean content. Adoption is uneven and no engine treats it as authoritative yet, but it is low-cost and signals intent.
• Structured data. Valid schema (Article, FAQPage, Organization, Product) gives engines unambiguous, machine-readable facts to lift. It does not force a citation, but it lowers the cost of selecting you.
• Clean sitemaps and internal links. A current XML sitemap and strong internal linking are still how crawlers find and prioritize pages. Orphaned pages with no inbound links are the ones that quietly never get retrieved.

If you only do one thing after reading this, open robots.txt and check the five crawler names above. It is the highest-leverage five-minute audit in AI visibility.

Where OpenLens fits

The manual check works for one page on one engine on one afternoon. It does not work for a 400-page site across seven engines, re-run as the engines change behavior month to month — which they do.

That gap is what OpenLens automates. Instead of spot-checking a single URL by hand, OpenLens runs the retrievability check across every page and every engine, and maintains the underlying method as engines shift how they retrieve and cite. We are most confident about the two engines where verification is genuinely reliable today — ChatGPT and Perplexity — and we are honest that Claude and Google AI Overviews remain an active frontier rather than a solved, fully-GA check. We would rather tell you "this one is hard" than hand you a number we do not trust.

OpenLens also includes a Site & Agent Readiness audit — a 0-100 score covering the structural prerequisites above: discoverability (sitemaps, internal links), content accessibility (server-side rendering vs JS-only), bot-access policy (the robots.txt crawler checks), and agent protocols (llms.txt, structured data). It turns the checklist in the previous section into a measured score you can track over time and hand to a client.

OpenLens tracks brand visibility across 7 platforms — ChatGPT, Google AI, Gemini, Perplexity, Grok, Claude, and DeepSeek — and the free tier requires no credit card, so you can run a first retrievability pass before deciding anything. If you are still assembling a stack, our roundup of the best free AI visibility tools for agencies is a useful starting point, and the side-by-side look at OpenLens against Profound covers the enterprise end of the market.

Last updated June 18, 2026.

Sources

• Semrush, "AI Overviews Market Research" — enterprise study of 2,855 keywords finding AI Overviews on 13.14% of analyzed queries by March 2025 (Semrush, 2025).
• Google Search Central, "AI features and your website" / Search guidance on AI experiences — confirms there is no separate index, markup, or submission step for AI features (Google, 2024-2025).
• Aggarwal et al., "GEO: Generative Engine Optimization," Princeton University (2024) — source-level structure (statistics, quotations, citations) measurably shifts which passages generative engines cite.
• OpenAI, Anthropic, Perplexity, and Google crawler documentation — GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, and Google-Extended user-agent and robots.txt behavior.