Which AI assistant should I optimize for first?

Pick based on your audience. B2B and engineering audiences over-index on ChatGPT and Claude. Research-curious and journalism-adjacent audiences over-index on Perplexity. Consumer search queries that previously went to Google now often go to Google AI Overviews + Gemini. Most sites end up optimizing for the cross-platform fundamentals (which work for all of them) and only doing per-platform tuning for the one driving the most measurable traffic.

Do I need to allow all the AI crawlers?

For most sites that want traffic and citation, yes. The major AI crawlers (GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot, Brave's) are well-documented and respect robots.txt. Blocking them means opting out of citation by that platform. Some publishers block on purpose (paywall protection, training-data licensing); know the cost before doing it.

What is the single highest-leverage tactic across all of them?

Writing well, with concrete data and primary-source citations. This is what the Aggarwal et al. paper found across generative engines, and it holds in 2026 across platforms. Tactical SEO moves help at the margins; content quality is the constant. Pages that read like a useful, well-cited answer to a real question get cited; pages that read like marketing copy do not.

How to Get Cited by Claude, ChatGPT, and Perplexity

As of 2026-05-24

The four major AI assistants in 2026 — Claude with web search, ChatGPT (including SearchGPT-style answers), Perplexity, and Google Gemini — all build on top of the same general pattern: crawl, index, retrieve, generate, cite. The per-platform differences are real but smaller than they sound. The cross-platform fundamentals are most of what matters.

The cross-platform fundamentals

Before any per-platform tactics, get these right:

Crawlability. Your site has to be discoverable. Check that your robots.txt does not block:

GPTBot (OpenAI training crawler)
ChatGPT-User (OpenAI on-demand fetcher for user queries)
OAI-SearchBot (OpenAI's SearchGPT index)
ClaudeBot (Anthropic crawler)
anthropic-ai and Claude-Web (older Anthropic user agents)
PerplexityBot (Perplexity crawler)
Google-Extended (Google's "extended" crawler used for AI features)
Bingbot (still important — OpenAI has historically licensed Bing's index)

All of these respect robots.txt. The default for most sites should be to allow them.

Site health. Fast loads, HTTPS everywhere, mobile-friendly. The fundamentals search engines have demanded for years are unchanged.

Topical authority. Inbound links, mentions in reputable sources, depth of coverage on a topic. AI citation engines are downstream of search rank in most cases — their retrieval often uses similar signals.

Clear structure. Title that matches the page, descriptive headings, summary at the top, FAQ where it fits. AI assistants disproportionately favor content shaped like an answer to a question.

Schema markup. Article schema, FAQPage, HowTo, BreadcrumbList where appropriate. Helps both classic search and AI citation. JSON-LD is the standard format.

Get all of the above right and you have moved further than any per-platform tactic will move you.

Claude (Anthropic) with web search

Claude's web search returns inline source citations alongside answers. The crawler is ClaudeBot (and historically anthropic-ai and Claude-Web). What is documented:

Anthropic respects robots.txt for ClaudeBot.
Claude's web search citations tend to favor primary sources (papers, lab blogs, official documentation) over aggregator and SEO-driven pages.
Citation selection appears to weight authority and direct relevance heavily.

Tactics that align with what works:

Cite primary sources yourself. Pages that link to original research, official docs, and named-author commentary get cited more often.
Match the question shape. When the user asks "what is X," pages with H1 or early H2 of "What is X" get the easy match.
Be the depth, not the breadth. Claude tends to cite a small number of authoritative sources per answer rather than many shallow ones.

ChatGPT and SearchGPT

OpenAI's setup has more moving parts. GPTBot crawls for training; ChatGPT-User fetches on demand when a user query needs web data; OAI-SearchBot builds the SearchGPT index. Citations appear in answers when the model used web data.

What works:

Clean, scannable structure. SearchGPT-style answers seem to favor pages that read as direct answers. Lead with the answer; back it up with the explanation.
Schema markup for FAQ and HowTo. When the user query is a question, FAQ-schema pages have an easier path to citation.
Recent updates matter. Like Google, SearchGPT seems to weight freshness for time-sensitive queries.
Site authority still matters. OpenAI has historically leaned on the Bing index; pages that rank in Bing tend to be candidates for citation.

Perplexity

Perplexity is the most search-engine-like of the four — it crawls (PerplexityBot), indexes, retrieves, and shows inline numbered citations on every answer. Their help docs describe the source mix.

What works for Perplexity specifically:

Freshness. Perplexity weights recency heavily. Date-stamped content with visible publication dates gets surfaced more in time-sensitive queries.
Direct, factual writing. Perplexity's answers favor sources that read as factual reference material over opinion or marketing.
Citation density. Pages that cite their own sources extensively get cited in turn.
Structured snippets. Bullets, definition lists, and short, self-contained paragraphs are easier for Perplexity to lift into an answer.

Google Gemini and AI Overviews

Google's AI features (Gemini, AI Overviews in Search) share infrastructure with classic Google Search. The crawler signal is Google-Extended (separate from regular Googlebot, used for the AI features).

What works:

Classic Google SEO still does most of the work. Pages that rank well in Google Search are the candidate set for AI Overviews and Gemini citations.
Structured data. Google's existing schema preferences (Article, FAQPage, HowTo, Organization) carry over to AI Overview citation.
E-E-A-T signals. Experience, expertise, authoritativeness, trustworthiness. Google's editorial signals matter for AI Overview surface area.
Allow Google-Extended. Blocking it specifically (while allowing regular Googlebot) opts you out of AI Overview citation.

You.com, Brave Leo, and the long tail

The smaller AI assistants — You.com, Brave Leo, Arc Search, smaller agentic tools — generally follow the cross-platform fundamentals more than they impose platform-specific quirks. Optimizing well for the big four covers most of the rest.

Tactical patterns that work everywhere

A short list of editorial moves that improve citation rates across platforms:

Lead with the answer. Put the direct answer to the implied question in the first paragraph or in a clearly-marked TLDR.
Use citations and data. Specific numbers, named sources, and links to primary material. The Aggarwal et al. finding holds: this is the highest-leverage single move.
Structure for scannable answers. Headings phrased as questions. Bullets where the content is a list. Definition lists for terms.
Show your work. Methodology pages, dated edit logs, and explicit limits ("here is what we did not test"). This is what makes a page citable as authoritative.
Keep the URL stable. AI citations get baked into model behavior over time. URL changes break those bakes.

What does not work as well as it used to

A few things that worked for classic SEO but do less for GEO:

Keyword density. AI assistants do not match on keyword frequency the way old Google did. Write naturally.
Long-tail keyword stuffing. AI answers come from synthesizing a few sources; an article optimized for 200 long-tail variants reads as junk and gets cited less.
Pure SEO content with no real authority. AI assistants are good at sniffing out content that exists primarily for ranking. Pages with substantive depth and primary-source citations beat them.

The honest verdict

Per-platform GEO tactics are real but secondary. The single most impactful thing you can do is write substantive, well-cited, well-structured pages on topics you have actual depth in, and make sure the AI crawlers can read them. Everything else is rounding.