AI search is still search, and every approach that has actually increased brand visibility in AI responses for me and my clients builds on that fact.
Each assistant reads a different index, and it retrieves against sub-questions it writes itself, not only the prompt the user typed. That framing explains almost everything below: rank gets you retrieved, structure gets you quoted, and structured data gets you mentioned. It also means citations and mentions are different goals. A citation is a linked source in the answer. A mention is your brand named without a link. They respond to different tactics, and this post covers both.
AI Search Indexes by Assistant
The first question that matters is which index each assistant actually reads, because that decides where your optimization effort goes.
| Assistant | Index | Practical Implication |
|---|---|---|
| ChatGPT | OpenAI’s own index plus external providers; paid Thinking mode leans on Google rankings | Snippet-level optimization for the free tier, Google rank for paid modes |
| ChatGPT Deep Research | Bing | Bing Webmaster Tools matters |
| Claude | Brave Search | Brave indexing matters |
| Perplexity | Own index, tracks Google closely | Google top 10 positions carry over |
| Google AI Overviews and AI Mode | Google’s index | Classic SEO is the entry fee |
ChatGPT is a hybrid. OpenAI runs its own index, known internally as Labrador, alongside external providers (Search Engine Land, Peec AI). On the free tier and in instant mode, almost everything comes from OpenAI’s own index, while the paid Thinking mode gets about 75% of its results from scraped Google rankings (Search Engine Land). More than 90% of ChatGPT users are on the free tier, so OpenAI’s own index decides most of your visibility there (Search Engine Land). Deep Research is the exception and still runs on Bing (Peec AI). One hard requirement: sites blocking OAI-SearchBot will not appear in ChatGPT search results at all (OpenAI docs).
Claude uses Brave Search, which is listed on Anthropic’s subprocessor page (TechCrunch, Ryan Doser). One vendor study found Claude citations match Brave’s top results 86.7% of the time (RivalHound).
Perplexity runs its own index but tracks Google most closely of the independents; nearly 1 in 3 of its citations rank in Google’s top 10 (Ahrefs).
Google AI Overviews and AI Mode read Google’s index. AI Mode has been measured at 93% of citations coming from inside Google’s top 10, with AI Overviews near 40% (Elmo).
The Snippet Problem
ChatGPT usually does not read your page. It queries an index, receives a title, a URL, and roughly 200 characters of text, and works from that. The actual page gets opened in about 1 of 80 cases, and only in Thinking mode (RESONEO).
The implication is blunt: your title, your meta description, and the first lines of each section are what the model actually sees. If the answer to the question is buried in paragraph four, the model never encounters it. This is why the answer-first tactic below is not a style preference, it is the retrieval budget.
Search Rank and Citation Overlap
How much does Google rank predict AI citations? The honest answer is that it depends on the surface, and the published numbers vary widely.
- Ahrefs studied 15,000 long-tail queries and found only about 12% of URLs cited by ChatGPT, Gemini, and Copilot ranked in Google’s top 10 for the same prompt, with ChatGPT alone around 8% (Ahrefs).
- A March 2026 Ahrefs analysis of 863,000 SERPs found 37.9% of AI Overview citations in the top 10, 31.2% in positions 11 to 100, and 31.0% from outside the top 100 entirely (Elmo).
The reconciliation is query fan-out. The assistant splits a question into sub-queries and retrieves results for each one separately. A page ranking #40 for the head term can still get cited because it ranks #2 for one of the sub-questions the model wrote itself. You are not optimizing for one query, you are optimizing for the cloud of smaller questions inside it.
One caveat worth stating plainly: published overlap estimates range from 12% to 93% depending on the surface and the measurement method, and most of this data comes from AEO vendors with something to sell (Elmo). Treat the direction as reliable and the precision as soft.
Citation Tactics
Problem/Solution Structure
I have written somewhere between 500 and 1,000 articles over roughly 20 years, and many of them follow the same shape: a specific problem, then a direct solution, written as a definitive source for one specific need. Some are short, some are long, but each exists to answer one thing. Without any AI-specific optimization, many of these began appearing in AI Overviews frequently.
The research supports the pattern. Extractable structure correlates with AI Overview citations: the answer in the first sentence of a section, the subject named explicitly rather than referred to by pronoun, one subject per chunk, tables for comparisons, and numbered lists for steps (dameSpeak).
A formal FAQ block is not required and may not even help. In a vendor study by SE Ranking, pages without FAQ schema averaged more ChatGPT citations (4.2) than pages with it (3.6) (NOVASTACKS summary of SE Ranking). The structure of the writing matters; the schema wrapper around it does not.
Answer-First Content
Put the most valuable information at the top of the page and at the top of every section. This follows directly from the snippet problem: roughly 200 characters is the budget the model works with, so the first lines of a section either contain the answer or the section does not exist as far as retrieval is concerned. Write the conclusion first, then support it. Save the context and the caveats for after the answer, not before it.
Content Consolidation
One definitive page per topic beats several partial ones. Referring domains were the single strongest predictor of ChatGPT citations in SE Ranking’s vendor study of 129,000 domains and 216,524 pages (SE Ranking, Search Engine Journal). Three thin pages on the same topic split their links three ways; one consolidated page concentrates them.
The condition that makes consolidation work with fan-out: cover each sub-question in its own self-contained section, with the answer stated at the top of that section. A long page is fine. A long page where each section can stand alone as a retrieved chunk is what you are actually building.
Freshness
SE Ranking found 95% of ChatGPT citations came from content published in the last 10 months (NOVASTACKS). The practical move is not a publishing treadmill. Updating your consolidated pages, with real revisions and a current date they can verify, matters more than shipping new thin ones. This is the same conclusion I keep arriving at from the classic SEO side: improve what you have.
Bing and Brave Indexing
Two indexes most teams never think about now matter. Submit your site to Bing Webmaster Tools, which feeds ChatGPT Deep Research and Copilot, and to Brave, which feeds Claude. Brave has a Submit URL page you can use to trigger a re-fetch (Convertos).
Then make sure you are not blocking the crawlers themselves. Allow OAI-SearchBot, ClaudeBot, and PerplexityBot in robots.txt and in your Cloudflare or WAF bot rules, which silently block AI crawlers on plenty of sites whose owners never chose that. Note from OpenAI’s own documentation: OAI-SearchBot (search) and GPTBot (training) are independent settings, so you can stay in ChatGPT search results while opting out of training (OpenAI docs).
Server-Rendered HTML
This one is about rendering, not speed. Vercel’s crawler analysis found that none of the major AI crawlers, including OAI-SearchBot, ChatGPT-User, GPTBot, ClaudeBot, and PerplexityBot, executed JavaScript; they fetch the JS files but cannot read content that only exists after client-side rendering (Vercel). One July 2026 log analysis saw signs OAI-SearchBot may partially execute JavaScript (Crawl Lab), but treat that as unreliable rather than something to build on.
The rule: keep your content in the initial HTML response with a low markup-to-content ratio. If your important copy renders client-side, it is invisible to most of these systems. This pairs with serving clean Markdown to AI crawlers, which takes the same idea further.
JSON API for Brand Mentions
This tactic targets mentions, not citations. The goal is getting a product named in answers to specific, edge-case questions: does it support a particular integration, what is included in a pricing tier, does it handle a niche use case. Those are exactly the long-tail questions where assistants have thin sources, and ChatGPT mentions brands roughly 3.2 times more often than it cites them with links (5WPR), so unlinked mentions are where most brand exposure actually happens.
What I build is a live JSON endpoint that publishes structured, always-current data about a product or service: features, specifications, integrations, pricing structure, use cases. This is not an llms.txt file, which is a separate idea with real evidence against it; this is machine-readable data at a real URL that crawlers demonstrably fetch.
I have seen results from this across my own site and client sites, and Cloudflare logs show AI crawlers hitting these endpoints frequently. To be precise about what the logs prove: they are evidence of consumption, not causation, and none of this affects model training. What they show is that the machines are reading the data you publish for them.
Third-Party Mentions
For branded queries, most citations are not yours to own. Omniscient Digital’s analysis of 23,387 citations found 57% went to reviews, listicles, forums, and case studies, 17% to directories, and owned content accounted for only 23% of total citations (Omniscient Digital, Omniscient Digital).
Reddit deserves specific attention. In SE Ranking’s data, domains with a heavy Reddit presence averaged 7 ChatGPT citations against 1.8 for domains with minimal presence (Contently summary of SE Ranking). Reviews, comparison articles by others, directory listings, and genuine community presence are inputs to AI visibility, not a separate PR channel.
AI Referral Tracking in GA4
Google Analytics now separates this traffic for you. GA4’s default channel group includes an AI Assistant channel covering arrivals from sources like ChatGPT, Gemini, Deepseek, Copilot, and Grok; when the referrer indicates an AI assistant source, GA4 automatically sets the medium to ai-assistant and the campaign to (ai-assistant) (Google Analytics documentation). You will find it in the Traffic acquisition report under the session channel group dimension.
Two caveats. First, Google’s documentation is explicit that the AI Assistant channel excludes AI Overviews and AI Mode; those clicks are classified as Organic Search, so your AI Overview visibility never shows up as AI traffic. Second, clicks from native mobile apps often arrive with no referrer at all and land in Direct. Both mean the channel undercounts, which is why I treat it as one signal inside a broader measurement approach rather than the scoreboard.
Summary
Rank gets you retrieved: each assistant reads a specific index, and you need to exist in it, whether that is OpenAI’s, Brave’s, Bing’s, or Google’s. Structure gets you quoted: answer-first sections, one subject per chunk, and titles and opening lines that work inside a 200-character budget. Structured data gets you mentioned: live machine-readable endpoints put your product’s specifics in front of the systems answering edge-case questions. Third-party mentions do the rest, because most branded citations go to sources you do not own.
None of this is separate from search. It is search, read through different indexes and assembled by a model, and the sites doing the fundamentals well are the ones the machines keep recommending.