Reading time: 13 minutes
Gemini, Perplexity and Claude choose sources in different ways, but all of them depend on access to your site. To show up in Gemini, a page must be in Google Search’s index, and the site can’t block Google-Extended. In Perplexity, the bot to allow is PerplexityBot, while in Claude they’re Claude-SearchBot and Claude-User. After that, citable content and mentions on other sites carry weight.
In short
- Gemini finds sources in Google’s index. Blocking Google-Extended removes the site from grounding for answers in the app, but not from Search.
- Perplexity has its own index, fed by PerplexityBot. Perplexity-User, which opens pages at a user’s request, generally ignores robots.txt.
- Claude searches with providers such as Brave Search, which doesn’t crawl what Googlebot can’t crawl.
- In Brazil, Gemini generated 17.95% of the visits that chatbots sent to websites in September 2026, while Perplexity and Claude were close to 2% and 1%.
How much weight do Gemini, Perplexity and Claude carry in Brazil?
Gemini is the AI that sends the second most visits to websites from Brazil, far behind ChatGPT, but rising. However, Perplexity and Claude carry little weight on this metric. In September 2026, according to StatCounter, Gemini generated 17.95% of these visits, Perplexity 2.14% and Claude 1.1%.
StatCounter counts the visits that chatbots send, from the country, to websites that carry the company’s code. So it doesn’t measure how many people use each tool. In the monthly series, Gemini had about 3% in September 2025. Meanwhile, Perplexity reached 4.84% in December 2025, and Claude reached 3.01% in June 2026.
Search interest tells another part of the story. On Google Trends, the term “claude” went from 3 to about 30 points in Brazil between October 2025 and October 2026. In the same period, “gemini” averaged 78 and “perplexity” averaged 2. Note that the index is relative, and “claude” also picks up searches for people with that name.
| Engine | Visits to websites (Sep 2026) | Google Trends (12-month average) |
|---|---|---|
| Gemini | 17.95% | 78 |
| Perplexity | 2.14% | 2 |
| Claude | 1.1% | 17 |
In practice, Gemini deserves priority right after ChatGPT and Google. Perplexity and Claude, on the other hand, call for the technical basics and monitoring. They need more effort only when the client’s audience uses these tools.
How does Gemini choose its sources?
Gemini takes its sources from the Google Search index, in a process Google calls grounding. In other words, the model receives content from the index while it answers, to improve accuracy and relevance. For this reason, the foundation is SEO: a crawlable, indexed and useful page.
The definition is in Google’s crawler documentation, in the section on Google-Extended. It says grounding provides the model with content from the Google Search index at prompt time. This applies to Gemini Apps as well as to Grounding with Google Search on Vertex AI.
The Gemini API documentation describes the step-by-step process in the developer version. First, the model analyzes the prompt and decides whether a Google search would improve the answer. Then, if needed, it generates one or more queries, runs the searches and writes the answer with citations.
AI Mode and AI Overviews also use Google’s index. They live inside Search and have their own rules, explained in the guide to Google AI Mode and AI Overviews.
Does blocking Google-Extended remove my site from Gemini?
It removes the site from grounding. With Google-Extended blocked, the content Google crawls no longer grounds answers in Gemini Apps. It also stays out of training for future models. But the site stays in Search, and Google says the token isn’t a ranking signal. The brand can still show up through mentions on other sites.
In fact, Google-Extended isn’t a bot. It’s a control token in robots.txt, with no user agent of its own, because Google’s regular bots do the crawling. That’s why it doesn’t show up in server logs.
The token also doesn’t remove the site from AI Overviews or AI Mode. For those, the control is in Search Console, under the Search generative AI control. According to Search Console Help, this control doesn’t affect AI training. Therefore, to limit training, use Google-Extended.
| Control | What it removes | What it doesn’t remove |
|---|---|---|
| Google-Extended | Grounding in Gemini Apps, and training | Search, AI Overviews and AI Mode |
| Search Console control | AI Overviews, AI Mode and AI in Discover | Training and the rest of Search |
In short, blocking Google-Extended is a business decision. If you want to be among the sources in the Gemini app, leave it allowed.
Is showing up in Gemini the same as showing up in Google?
No: the index is the same, but Gemini picks its sources differently. Of course, a page needs to be in the index, and ranking well puts it in the running. But the choice goes through the model, and Gemini often cites pages outside Google’s top results for the same question.
Ahrefs ran the same 15,000 queries in Google and in four assistants in early July 2025. In Gemini, ChatGPT and Copilot, only about 8% of cited links were in Google’s top 10. In addition, more than 80% came from pages that didn’t even rank for the query. Perplexity was the exception, with 28.6% of its citations in the top 10. Ahrefs sells an AI monitoring tool and warns that the results may have changed.
In practice, treat Gemini as a channel of its own, since its sources rarely appear in Google’s top 10. Monitor prompts in it, and don’t infer the brand’s presence from its Google ranking.
How does Perplexity choose its sources?
Perplexity searches the web for every question, in its own index fed by PerplexityBot, and answers with numbered citations. The index ranks not only whole pages but also passages. So Perplexity can cite a section that answers a sub-question well, even inside a long page.
According to Perplexity’s engineering blog, from September 2025, the index tracks more than 200 billion unique URLs. The company says it treats sections and passages of documents as units in their own right. The reason, according to the post, is that the model needs precise, compact context.
Two bots matter. PerplexityBot indexes the site for results, so Perplexity’s documentation recommends allowing it in robots.txt and accepting its published IP ranges. Perplexity-User, on the other hand, opens pages when a user asks a question. Because the user requests the visit, it generally ignores robots.txt. Rule changes take up to 24 hours to take effect.
Is it worth optimizing for Perplexity in Brazil?
It’s worth allowing access and monitoring the brand, because that costs little. However, producing content just for Perplexity rarely pays off in Brazil, where it carries little weight in visits and searches. The exception is brands whose audience uses the tool, such as technical niches.
After all, Perplexity’s audience in Brazil is small. The data comes from a study by the consultancy Cadastra with Similarweb, reported by the Brazilian trade publication Meio & Mensagem. In August 2025, Perplexity had 2.01 million visits in Brazil, while ChatGPT had 310.67 million. Perplexity grew 131.03% in a year, but only 1.04% of ChatGPT users also used it.
According to the report, Perplexity use is more technical and concentrated among men, who make up 56.8% of its audience. The reports on the study don’t include a figure for Gemini.
How does Claude search and cite sources?
Claude searches the web whenever a question calls for current information, and answers that use search include citations. To find pages, Anthropic uses its own bots as well as search providers, such as Brave Search. Claude-SearchBot indexes content, while Claude-User opens pages at a user’s request.
According to the Claude Help Center, Claude invokes its search tool for topics that call for current information. Every answer that uses search includes citations. In the new app experience, there’s no toggle: Claude searches when it helps. Image search uses Bing.
In the Anthropic Trust Center, Brave Search appears as a web search subprocessor for all products. TurboPuffer appears in the same role, but the list doesn’t detail what each one does.
Brave creates a practical dependency. Brave’s crawler doesn’t identify itself with its own user agent, according to Brave Search Help. It also doesn’t crawl what Googlebot can’t crawl. In other words, a folder blocked for Googlebot stays out of Brave’s crawl.
Anthropic’s bots follow robots.txt, according to Anthropic’s documentation. That includes Claude-User, unlike Perplexity’s equivalent bot. Blocking Claude-SearchBot can reduce the site’s visibility in Claude’s searches. In contrast, blocking ClaudeBot only signals that new content should stay out of training.
What changes in Gemini, Perplexity and Claude when you block each bot?
In Claude, as in ChatGPT, one bot decides whether the site can become a source, and another serves training. Blocking the first tends to remove the site from answers that use search. Blocking the second only keeps the content out of training. In Gemini, however, the same token handles both. According to the company, PerplexityBot plays no part in training.
| Bot or token | Engine | Effect of blocking |
|---|---|---|
| Google-Extended | Gemini | Drops out of grounding in the app and of training, but Search stays |
| Googlebot | Google, Gemini and Claude | Google and Brave stop crawling the page |
| PerplexityBot | Perplexity | The site tends to drop out of results |
| Perplexity-User | Perplexity | Little effect: it generally ignores robots.txt |
| Claude-SearchBot | Claude | Can reduce visibility in searches |
| Claude-User | Claude | Claude stops opening the page at a user’s request |
| ClaudeBot | Claude | New content stays out of training |
OpenAI’s bots and a summary by engine are in the table of search bots by AI engine. A sample file that allows search and refuses training is in the guide to robots.txt for AI. Blocking by IP or firewall calls for care. For example, Anthropic warns that blocking IPs can prevent it from reading robots.txt. Similarly, Perplexity recommends allowing its published IP ranges in the web application firewall (WAF). On sites behind Cloudflare, also check Cloudflare’s AI policies, which can block these bots.
However, allowing a bot isn’t enough if the page arrives empty. The comparison of SSR and CSR shows what each bot sees.
What type of source does each engine tend to cite?
Gemini, Perplexity and Claude cite different kinds of sources, according to studies by companies that sell AI monitoring. For example, in Gemini and Perplexity, Reddit led the citations. Claude cites sources in fewer answers, and scientific databases sit at the top of its list. For a brand, this defines where to seek mentions beyond its own site.
Muck Rack analyzed more than 25 million links cited by ChatGPT, Claude and Gemini, in a study from May 2026. The questions simulated consumers in 17 industries, but the published examples are in English. Gemini cited sources in 82% of answers, with an average of 8, and its most cited domain was Reddit. Claude cited sources in 55% of answers, but with an average of 13 when it did.
In Claude, PubMed Central, a database of scientific articles, topped the list of cited domains. It cited neither Reddit nor YouTube. Muck Rack sells public relations software and a tool that monitors brands in AI. Greg Galant, the company’s cofounder and CEO, offered this reading in the study’s press release:
“If your brand is not showing up in the media coverage AI is reading, you are not showing up in the answers AI is giving.”
Greg Galant, Muck Rack (2026)
The study didn’t include Perplexity, though. Profound analyzed 680 million citations from August 2024 to June 2025. In Perplexity, Reddit led with 6.6% of citations. YouTube came next, with 2%. Profound sells an AI monitoring tool.
List the domains each AI cites for the client’s category. Treat each one as a content or public relations target. In Gemini and Perplexity, the community sources AI cites, such as forums, deserve separate work.
How to rank in Gemini, Perplexity and Claude: step by step
The path combines technical access, citable content and presence off your site. In general, the steps apply to Gemini, Perplexity and Claude, with adjustments for each bot. None of them guarantees a citation, but each one removes a reason for the answer to leave the brand out.
- Allow the search bots. Leave PerplexityBot, Claude-SearchBot and Claude-User free in robots.txt, and check your CDN and firewall. Then test with curl whether PerplexityBot and Claude-SearchBot receive the page text, as the JavaScript SEO guide shows.
- Decide on Google-Extended with the client. If you allow it, the content can ground Gemini. If you block it, the content drops out of grounding and training.
- Don’t block Googlebot on important pages. That’s because, without it, Google and Brave stop crawling the page.
- Take care of indexing in Google. Gemini takes its sources from the Search index, so the same technical SEO work applies.
- Write sections that stand on their own. One question per H2 and the answer right below it, because Perplexity ranks passages.
- Earn mentions where each engine looks. For example, forums such as Reddit for Gemini and Perplexity; reference and research sources for Claude.
- Check the logs. Look for PerplexityBot, Perplexity-User, Claude-SearchBot and Claude-User. Google-Extended doesn’t appear in them.
- Measure each engine separately, with the same prompts, repeated and collected from Brazil, in Portuguese.
The collection protocol is in the guide on how to measure brand presence in each engine. For ChatGPT, the path changes, and the step-by-step is in how to show up in ChatGPT. For traffic, separate Gemini, Perplexity and Claude visits in GA4.
What we found when we analyzed 8 Portuguese-language guides on Gemini and Perplexity
We analyzed 8 Portuguese-language pages that show up in searches about how to show up in Gemini and Perplexity. Three cover Gemini, and none mentions Google-Extended or grounding in Search. Five cover Perplexity, but only 2 mention PerplexityBot by name.
In addition, the only one that mentions Perplexity-User says it respects robots.txt, the opposite of Perplexity’s documentation. None of the 8 mentions Claude-SearchBot, and no Portuguese-language guide on Claude showed up in our searches. Finally, none includes usage data for Brazil, and four are from 2025.
Want to know how your brand shows up in AI answers today?
NotHuman is opening spots for agencies and brands to track how they appear in AI answers. Join the waitlist and we’ll let you know when a spot opens up for you.
Main sources
- Google Search Central, Google’s common crawlers and the Google-Extended token (2026).
- Perplexity, Perplexity Crawlers (2026).
- Anthropic, Anthropic’s bots and how to block them (2026).
- Claude Help Center, web search in Claude (2026).
- StatCounter, AI chatbot market share in Brazil (2026).
- Muck Rack, What Is AI Reading?, May edition (2026).
