Ranking on Google Is Only Half the Job: How Enterprise Brands Get Cited by ChatGPT, Perplexity and Claude

To appear in ChatGPT & Perplexity answers, an enterprise brand has to be present in every index those assistants query, Google’s but also their own, let their crawlers in, and give them text they can reuse as it stands. That text is often limited to a title and the first lines of a page. ChatGPT, Perplexity and Claude each combine these sources differently.

• ChatGPT relies mostly on OpenAI’s own index in its default mode and on Google results in its paid reasoning mode.
• OpenAI states that sites blocking its search crawler, OAI-SearchBot, are not shown in ChatGPT search answers.
• Claude uses Brave Search, an index independent from Google, among its web search providers.
• What reaches the model is often a title and a short snippet, not the full page.
• Citation rate is measurable on non-brand prompts, and it is the only baseline that shows where to act.

Google Rankings Feed Some AI Answers, Not All of Them

Google rankings matter for AI answers, but only for part of them. Google’s AI Overviews and AI Mode are built on Google’s own index: a page must be indexed and eligible for a snippet to be cited, and Google documents that both features split a question into several related searches, a technique called query fan-out.

ChatGPT is less straightforward. OpenAI says it relies on third-party search providers alongside its own systems. Independent research published in August 2026, based on 1,200 ChatGPT answers, found that the paid reasoning mode drew about 75% of its results from Google, while the free modes relied mostly on OpenAI’s own index. The default experience weighs more: OpenAI reported 900 million weekly users and about 50 million paying subscribers in February 2026, so most conversations run on the free tiers.

Whatever the source, each assistant runs the same sequence, and each step can drop a brand that ranks well on its main keyword:

• Search decision: the model chooses whether to answer from what it already knows or to search the web.
• Query fan-out: the prompt is rewritten into several sub-queries, often phrased differently from the buyer’s words.
• Retrieval: candidates come from one or several indexes, depending on the engine and the mode.
• Selection: a handful of results are kept, synthesized and cited.

This is why LLM SEO is not a new label on old work. The fundamentals still apply, but the target moves from the main keyword to the sub-queries, and from one index to several.

To Appear in ChatGPT & Perplexity Answers, Unblock the Right Crawlers First

The first reason an enterprise site is missing from AI answers is often technical: it blocks the crawler that feeds the answer. OpenAI’s training crawler, GPTBot, was blocked by 5% of the top 1,000 websites when it launched in August 2023 and by more than a quarter of them a year later. Blanket rules written against AI training can also catch the crawlers that feed search answers.

Each vendor now separates training from search, which means a brand can refuse training and remain citable:

• OpenAI documents 4 agents. GPTBot collects training data. OAI-SearchBot surfaces sites in ChatGPT search, and OpenAI states that sites opted out of it are not shown in ChatGPT search answers, though they can still appear as navigational links. ChatGPT-User opens pages when a user request calls for it.
• Anthropic documents 3. ClaudeBot collects training data, Claude-SearchBot indexes content for Claude’s search, and Claude-User fetches pages when users ask questions.
• Perplexity documents 2. PerplexityBot must be allowed for a site to appear in its search results, and Perplexity-User visits pages to answer user questions.

The second filter is rendering. Independent tests show that the fetcher ChatGPT uses to open pages does not execute JavaScript, so content loaded client-side, such as pricing in tabs or FAQ blocks loaded on click, is invisible to it.

In large organizations, neither filter usually sits with the SEO team. Robots.txt rules, CDN settings and bot protection are often owned by security or infrastructure teams, across several domains and country sites. Reviewing them is the first item Primelis checks, because no content work matters until access is fixed.

ChatGPT, Perplexity and Claude Don’t Combine Their Sources the Same Way

There is no single AI index to optimize for: each assistant assembles its sources in its own way.

ChatGPT mixes OpenAI’s own index with results from third-party providers, and the balance depends on the mode. Ranking on Google weighs heavily in paid reasoning mode; in the free default mode, presence in OpenAI’s own index is the main gate.

Perplexity is built as a search product: it answers from web results, drawing in particular on its own continuously refreshed index, and shows its sources under each answer. Perplexity states that changes to robots.txt can take up to 24 hours to be reflected.

Claude decides when a question needs a search, formulates its own queries, then cites its sources. Anthropic lists Brave Search among the web search providers behind Claude. Brave runs its own independent index, so a page can sit at the top of Google and be thin or absent on Brave.

Asking how to rank in ChatGPT therefore frames the problem wrongly. There is no stable ranking inside ChatGPT to climb: answers vary with the mode, the model and the sub-queries generated. What a brand controls is its presence in each index these engines draw on, its eligibility for each crawler, and the quality of the text they retrieve.

What Reaches the Model Is Often a Snippet, Not Your Page

In most ChatGPT conversations, the model never reads the page it cites. The August 2026 research found that no page was opened in 93% of answers in the fast mode: the model worked from the page title and a snippet of about 200 characters taken around the H1. The meta description was ignored. When pages were opened, in reasoning mode, an opened page ended up cited 74% of the time, against 7% for a page only retrieved.

Two consequences follow for enterprise sites:

• The first lines after the H1 decide what the model knows. Category labels, dates, bylines or a table of contents placed between the H1 and the first sentence take up those 200 characters. On large sites, those elements come from templates owned by product or development teams, not by SEO.
• Opened pages still have to be quotable. A block that answers on its own, names the entity (“Primelis” rather than “we”) and carries a dated fact is easier to reuse than a narrative paragraph.

Consistency is the other enterprise trap. When country sites, product pages and press releases give different facts about the same offer, engines either hedge or cite a third party instead. This is where AI answer engine optimization departs from content production: it is governance of facts and templates across sites and teams, a task that nobody owns by default.

For Recommendation Prompts, Third-Party Sources Carry More Weight Than Your Own Site

When a buyer asks an assistant which vendor or product to choose, the answer draws heavily on sources the brand does not own: comparison articles, review platforms, B2B directories, forum threads such as Reddit discussions, and marketplace listings.

A brand absent from those sources is absent from the shortlist, whatever its own site says. The practical consequence is a distribution plan built from the sources engines actually cite on the category’s prompts:

• Profiles on directories and review platforms, complete and consistent with the brand’s own facts.
• Inclusion in the comparison content that assistants retrieve for the category.
• Expert contributions in the communities where the category is discussed.
• Digital PR aimed at the publications that appear in citations, not at reach alone.

At this stage, AI visibility stops being an SEO topic alone and brings in PR, brand and partnership teams.

Measuring Whether You Appear in ChatGPT & Perplexity Answers Starts With Non-Brand Prompts

A citation rate only means something when it is measured on the prompts buyers type before they know the brand’s name. Branded prompts flatter every company: ask an assistant about a brand by name and it will describe it. The baseline that matters is non-brand.

It also has to be measured engine by engine and mode by mode. In the August 2026 research, two ChatGPT modes answering the same prompts shared only about a third of the brands they mentioned. Before any recommendation, the measurement rests on four elements:

• A set of non-brand prompts mirroring the category’s buying journey, per market and per language.
• Repeated runs on each engine, since answers vary from one run and one mode to the next.
• Mentions and citations tracked separately: being named in the answer is not the same as being the linked source.
• Cited domains and fan-out queries recorded, to see which sources and which sub-questions shape each answer.

Primelis tracks this share of voice across the main AI engines, including ChatGPT, Gemini and Perplexity, with Outrank, its proprietary platform, which turns a one-off check into a trend a CMO can report on. The output is a gap list: prompts where competitors are cited and the brand is not, grouped by cause (crawler access, index coverage, snippet quality, third-party presence).

Where Enterprise Brands Should Start to Appear in ChatGPT & Perplexity Answers

Google rankings feed part of AI answers, not all of them. A brand has to be present in each index these engines draw on, readable by their crawlers, clear in the few lines they actually retrieve, and visible in the third-party sources they rely on for recommendations.

The order matters. Crawler access comes first, because it is fast to fix and blocks everything else. The text engines retrieve comes next, starting with the titles and first lines of the pages that answer the category’s questions. Third-party presence follows, because it takes the longest to build.

Primelis’s organic growth and GEO team starts every engagement with a non-brand citation baseline across the main AI engines, then ranks fixes by cause and expected impact. The first useful step is that baseline, run on your own category prompts.