← All articles

How Each of the 10 AI Platforms Decides Who to Cite

ChatGPT, Google AI Overviews, Claude, Perplexity, Bing Copilot, Gemini, Meta AI, Apple Intelligence, Alexa+, and SearchGPT each have different citation mechanics. Here's the platform-by-platform breakdown — including which popular tactics do nothing.

SW
By Spencer Wilkinson
Co-Founder & CTO, GET SEEN

"Optimize for AI search" is useless advice, because the ten platforms that matter don't share one indexing mechanism. Some read Google's index. Some run their own crawlers. Some weight freshness heavily; some barely care. And several of the most-recommended tactics — llms.txt files, schema markup as an AI-citation lever — do nothing on the biggest platforms.

This is the platform-by-platform breakdown we use inside our own audit engine.

The universal levers (all 10 platforms)

Before the differences, the fundamentals that move every platform at once:

  • Content quality: originality, first-hand experience, specific data, named authors. Commodity round-ups don't get cited anywhere.
  • Indexability: if a crawler can't reach and read your pages, nothing else matters. Client-rendered JavaScript shells are invisible to most AI crawlers — they do not execute JavaScript.
  • E-E-A-T: visible experience, expertise, authority, and trust — named people with verifiable credentials, not brand-voice-only sites.
  • Technical health: semantic HTML and a clean accessibility tree, which increasingly matters for browser agents too.

Google AI Overviews and Gemini

Both use Google's regular search index. If you're indexable and snippet-eligible in Google, you're eligible here; if not, no AI-specific trick helps. Google's own guidance is explicit that structured data is not required for AI feature inclusion, and llms.txt is ignored. The levers are content originality, E-E-A-T, and snippet quality. Schema still matters — for rich results, not AI Overviews inclusion.

ChatGPT and SearchGPT

OpenAI runs its own crawlers: GPTBot (training and background index), OAI-SearchBot (search citations), and ChatGPT-User (live browsing on a user's behalf). All three honor robots.txt — one Disallow line can make you invisible to the fastest-growing search surface on earth, and we regularly find sites blocking these bots by accident. Beyond access: answer-ready structure (question-form headings with direct, standalone answers) is what gets extracted and cited.

Claude

Anthropic crawls with ClaudeBot (and respects the older anthropic-ai token). Claude's citation behavior leans on entity clarity — an unambiguous picture of who you are, what you do, and why you're credible — which makes Organization and Person schema, a real About page, and consistent naming across the web disproportionately valuable.

Perplexity

Runs PerplexityBot and is the most citation-forward platform — every answer shows sources. It rewards content that reads like a source: specific, verifiable claims, data points, and freshness. Perplexity noticeably prefers recently updated pages, which is why visible publish and update dates matter more here than almost anywhere else.

Bing Copilot

Built on Bing's index (Bingbot). The underrated lever: IndexNow, which pushes new and updated URLs to Bing instantly instead of waiting for a crawl. Everything else follows the same fundamentals as Google.

Meta AI

Crawls with meta-externalagent. Entity signals and social presence carry more weight here than elsewhere — Meta AI answers lean on its knowledge of brands as entities across its platforms.

Apple Intelligence

Crawls with Applebot (and Applebot-Extended controls training use). Apple favors clean, structured, semantically-marked-up content. If you explicitly allow AI bots in robots.txt, include Applebot — many sites list the OpenAI and Anthropic crawlers and forget Apple's.

Amazon Alexa+

The voice-first outlier. Spoken answers need concise, speakable content: FAQ structure where the first sentence fully answers the question, and LocalBusiness schema (complete with telephone) for local intent — Alexa can't build a clean spoken answer from a rambling paragraph.

What doesn't work (despite the hype)

  • llms.txt is an emerging opt-in signal at best. Google's AI features ignore it; ChatGPT and Perplexity haven't formally adopted it. Publish one for future-proofing if you like — never expect it to move citations today.
  • Schema as an AI-citation guarantee. Schema is a rich-results and machine-readability lever. It is not what gets you into Google AI Overviews.
  • "Rewriting for AI" and keyword-permutation pages. Every platform's quality system is tuned against exactly this.

The order of operations

  1. Verify crawler access for all ten platforms' bots (one robots.txt line can zero out a platform).
  2. Fix indexability and rendering (server-rendered content, not JS shells).
  3. Add named authorship with real credentials, dates, and entity schema.
  4. Publish first-hand content: case studies, original data, direct answers to the questions your buyers actually ask.
  5. Then — and only then — layer platform-specific extras: IndexNow for Bing, speakable FAQ content for Alexa, Applebot access for Apple.

We test all of this live in every audit — including running your buyers' real questions through an AI with web search and checking who actually gets cited. The gap between "technically fine" and "actually cited" is where almost every business we audit is losing.