OpenAI's crawler documentation for GPTBot, OAI-SearchBot, and ChatGPT-User never mentions sitemaps. Anthropic's documentation, updated as recently as February 2026, doesn't either. Perplexity's crawler documentation is silent on the subject too. The one AI-adjacent search engine that discusses sitemaps and AI-generated answers in the same breath is Bing, and even Bing stops short of promising that submitting one gets a page cited. If someone has told you a special sitemap is the missing piece of your GEO strategy, the crawler vendors' own documentation says something more specific than that.
What OpenAI's own documentation says about GPTBot and sitemaps
OpenAI names several crawlers in its developer documentation: GPTBot, which gathers training data; OAI-SearchBot, which powers ChatGPT's search results; and ChatGPT-User, which fetches pages during a live conversation. The documentation explains what each one does and how to control them through robots.txt, the same rules we've broken down bot by bot. It does not describe sitemap parsing, a submission endpoint, or a priority value for any of the three.
That omission is worth noticing because OpenAI is specific about other technical details in the same document, including the published IP ranges site owners can use to verify a crawler isn't spoofed. If a sitemap changed how thoroughly GPTBot crawled a site, or which pages OAI-SearchBot pulled into a live answer, this is where OpenAI would say so. The closest the documentation gets is recommending that sites allow OAI-SearchBot in robots.txt to appear in ChatGPT's search results, which is guidance about access, not about sitemap files. For anyone used to Google Search Console's sitemap reports, that's a real gap: none of OpenAI's three named crawlers reads sitemap.xml as a discovery signal, at least not by the vendor's own account of how they work.
Anthropic and Perplexity: newer documentation, the same silence
Anthropic updated its crawler documentation in February 2026, clarifying what ClaudeBot, Claude-User, and Claude-SearchBot each do. Perplexity's documentation separates PerplexityBot, used for search-style indexing, from the retrieval that actually pulls pages into a live answer. Neither vendor's page mentions sitemap.xml, a submission endpoint, or a lastmod field anywhere.
Anthropic's refresh mainly gives site owners a more granular choice: block ClaudeBot's training crawl while still letting Claude-User fetch a page when someone asks Claude a direct question about a brand. Perplexity's documentation draws a similar functional line and states plainly that PerplexityBot itself isn't used to crawl content for AI foundation models at all, a distinction easy to miss if you're reading blog summaries instead of the source. In both cases, the guidance that does exist is entirely about robots.txt directives and crawl-delay settings. Discovery mechanics, meaning how either crawler decides which URLs to visit next, simply aren't addressed in either document.
Bing is the exception, and even Bing hedges
Bing is the only major engine behind a mainstream AI chatbot that documents a direct relationship between sitemaps and AI-generated answers. Microsoft Copilot draws its citations from the Bing index, the same infrastructure that ranks ordinary Bing results, so Bing's webmaster team writes about sitemaps with that overlap in mind. Even so, the guidance comes with an explicit disclaimer attached.
In a July 2025 post, Bing's webmaster blog stated that sitemaps “remain a foundational signal for ensuring comprehensive URL coverage” and singled out the lastmod field: an accurate, timestamped entry helps Bing decide which URLs to recrawl first, since “freshness signals directly influence how quickly updates are reflected in search results and AI generated answers.” Bing pairs that advice with IndexNow, the real-time submission protocol it co-created with Yandex in 2021, which Microsoft says now carries more than 3.5 billion URL submissions a day. The same Bing post is careful to add that “no tool can guarantee when or how your content will appear in AI-generated results.” Sitemaps and IndexNow speed up discovery and recrawling. Whether a freshly discovered page actually gets cited in a Copilot answer is a separate question Bing doesn't claim to answer.
Google set this precedent years before AI search existed
Google's position on sitemaps predates generative search by more than a decade, and it hasn't changed now that Gemini and AI Overviews sit on top of the same Googlebot index. The company's search help documentation still states the same limitation it always has.
“While a sitemap file can help us learn about your site, it does not guarantee indexing or increase your site's ranking,” Google's Search Central FAQ states.
Gemini and AI Overviews inherit that position because they run on Googlebot's crawl and index rather than a separate pipeline, the same overlap we found when testing whether AI crawlers render JavaScript. A sitemap can shorten how long an unlinked or newly published page waits to be crawled. It has never been, and still isn't, a signal Google uses to decide what ranks, and by extension what an AI Overview or a Gemini answer quotes. The distinction that matters for GEO is the same one that has mattered for SEO since sitemaps existed: a sitemap is a discovery aid, not a relevance or trust signal.
What a sitemap still does well, and where llms.txt fits
A sitemap earns its keep in three specific situations, and persuading an AI engine to cite you isn't one of them.
Large or thinly linked sites: a sitemap surfaces pages an internal link graph misses entirely.
Fast-moving or frequently updated content: an accurate lastmod timestamp shortens the wait before a crawler revisits.
Site migrations and URL changes: a sitemap hands crawlers a clean map when old internal links go stale.
That's a narrower job than llms.txt, the separate file some sites now publish specifically to summarize their most important pages for AI crawlers. Ahrefs' analysis of 137,000 sites found that 97% of llms.txt files received zero requests from the AI bots they were written for, a reminder that neither file substitutes for the thing that actually earns a citation: content a retrieval system judges worth quoting. Our own guide to llms.txt covers where that file does and doesn't help. A sitemap and an llms.txt file can sit on the same domain without conflict. Neither is a shortcut around writing something an AI engine's retrieval layer wants to extract.
The practical takeaway for a GEO checklist
Keep the sitemap. Keep it accurate, keep lastmod current, and submit it through Bing Webmaster Tools and Google Search Console the way you always have, alongside checking that a CDN like Cloudflare isn't quietly blocking the crawlers you want to reach. None of that is wasted effort. It's infrastructure that helps every engine, human search included, find and revisit your content faster.
What changes is the expectation attached to it. A sitemap is not a citation strategy, and no crawler vendor's own documentation claims otherwise. The GEO work that actually moves citation rates, meaning content structured to answer a specific query directly, named sources, and consistent entity signals, happens on the page itself. Reading crawler behavior in server logs will tell you whether GPTBot or ClaudeBot found a page at all. Whether they chose to cite it once they did is a content question, not a sitemap question.
Frequently Asked Questions
Do AI crawlers like GPTBot or ClaudeBot use my XML sitemap to find pages?
Not according to either vendor's own documentation. OpenAI's crawler docs for GPTBot, OAI-SearchBot, and ChatGPT-User don't mention sitemap parsing, and Anthropic's February 2026 update to its ClaudeBot documentation doesn't either. Both companies document robots.txt controls in detail but say nothing about sitemap files as a discovery mechanism.
Does submitting a sitemap help me get cited in ChatGPT or Perplexity answers?
There's no documented mechanism for that. A sitemap can help a crawler discover that a page exists, but neither OpenAI's nor Perplexity's documentation describes sitemaps as influencing which pages get pulled into a generated answer. Citation depends on what the retrieval system judges relevant and quotable once it has the page, not on how it found the URL.
Is Bing different, and does that mean sitemaps help Microsoft Copilot?
Bing is the exception that documents sitemaps in the context of AI search, recommending accurate lastmod timestamps because Copilot draws citations from the same Bing index that ranks regular search results. But Bing's own webmaster blog is explicit that this speeds up discovery and recrawling only. It states plainly that no tool can guarantee when or how content appears in an AI-generated result.
Should I still maintain an XML sitemap for a GEO strategy?
Yes. A sitemap remains useful for large or thinly linked sites, for content that updates often, and for migrations where URL structures change. It shortens how long a new or updated page waits to be crawled by any bot, human-search or AI. It just isn't the lever that determines whether an AI engine cites that page once found.
What's the difference between a sitemap and an llms.txt file?
A sitemap lists every URL on a site for general crawler discovery. An llms.txt file is a newer, separate proposal that summarizes a site's most important pages specifically for AI systems. Ahrefs found 97% of llms.txt files get zero AI bot requests, so neither file is a substitute for content built to be directly quotable.
Does Google Gemini or AI Overviews treat my sitemap differently than regular Google Search does?
No. Gemini and AI Overviews run on the same Googlebot crawl and index as standard Google Search rather than a separate pipeline. Google's long-standing position, that a sitemap helps discovery but doesn't guarantee indexing or ranking, applies to what surfaces in an AI Overview just as it applies to the traditional results beneath it.