Reading time: 14 minutes
The most cited sources in ChatGPT in the US are Reddit and Wikipedia, according to Ahrefs and Similarweb. In Brazil, however, industry studies put social networks and communities at the top, with Wikipedia behind. As a result, much of the answer about a brand takes shape outside its own site. Here’s how to gain presence in the sources cited by AI without breaking rules.
In short
- In the US, Reddit has 16.8% of the combined citations of the 50 domains ChatGPT cites most, according to Ahrefs.
- In Brazil, however, Profound measured Reddit at 9.0% of ChatGPT citations and Wikipedia at just 1.8%.
- In a study by the agency Conversion covering 1,315 projects, the complaint site Reclame Aqui appeared among the sources of 18.8% of them.
- Paid posts without disclosure run into Reddit’s rules and Brazil’s Consumer Defense Code (CDC). Likewise, reviews in exchange for discounts violate Google Maps policy.
What are the most cited sources in ChatGPT?
According to rankings from Ahrefs and Similarweb, Reddit and Wikipedia top the list of ChatGPT sources in the US. Below them, the list spreads across review sites, retailers, video and the press. The percentages change from one study to another, because each ranking has its own method, period and sample.
In the September 2026 ranking from Ahrefs, Reddit has 16.8% of the combined citations of the 50 most cited domains, and English Wikipedia has 7.0%. In other words, that isn’t the share of all ChatGPT citations. The data also comes from broad US queries across all topics.
The same list also includes Consumer Reports (3.7%), YouTube (2.5%), the retailer Walmart (2.3%), Trustpilot (1.5%) and the BBB (1.2%), the US organization that takes consumer complaints. Ahrefs sells the tool that collects this data.
However, in Similarweb data from January and February 2026, compiled by 5W, the order flips. Wikipedia has 13.15% of all ChatGPT citations in the US, and Reddit has 11.97%. The dataset adds up to about 600,000 citations from ChatGPT and Google’s AI Mode. Still, 5W, a public relations agency, warns that it only compiled the data, without checking it.
One ranking measures the share among the top 50 ChatGPT sources, while the other measures the share of the total. So compare rankings only when the engine, country, period and method are the same. For Gemini, Perplexity and Claude, see the type of source each engine tends to cite.
Why does AI cite Reddit and not your brand’s site?
Two factors help explain it, although no study isolates the cause. To begin with, Reddit has what a brand’s site doesn’t: usage stories, comparisons and criticism from third parties. It also signed deals to deliver its content in real time to OpenAI and Google. A corporate page says what the brand promises, while a thread shows what customers say about it.
In May 2024, Reddit announced that OpenAI would access its Data API to bring this content to ChatGPT. Earlier, in February of the same year, Google announced a similar deal. Steve Huffman, Reddit’s cofounder and CEO and an interested party in the deal, summed up the argument:
“Reddit has become one of the internet’s largest open archives of authentic, relevant, and always up to date human conversations about anything and everything.”
Steve Huffman, Reddit (2024)
But the deal explains access, not choice. According to OpenAI, when ChatGPT searches the web, it typically rewrites the question into more specific queries, and the answer may include the cited sources. For questions like “is it worth it” or “X or Y”, a thread with usage stories answers what was asked. A product page, on the other hand, often doesn’t. The full path is in how ChatGPT chooses the sites it cites.
What are the most cited sources by AI in Brazil?
In Brazil, social networks and communities appear at the top of the sources cited by AI, while Wikipedia carries little weight. That’s what two 2026 studies show, while a third shows that the ranking changes from month to month. However, all three come from companies that sell monitoring or GEO services, and none uses a sample that represents the market.
| Study | Scope | Main findings |
|---|---|---|
| Profound (Aug 2026) | ChatGPT, Brazil, 3 months, sample not disclosed | Reddit 9.0%; Wikipedia 1.8% |
| Conversion (Oct 2026) | 1,315 projects, May to Aug, 3 engines | YouTube 16.3%; Instagram 11.0%; Reddit 7.2% |
| Ranqia (Sep 2026) | ChatGPT, 73.1 million citations, Apr to Aug | 4 different leaders among social networks |
Profound, which sells an AI monitoring platform, compares ChatGPT sources in Brazil and the US. In Brazil, Reddit has 9.0% of ChatGPT citations, versus 2% to 3% in the US. Meanwhile, Wikipedia drops from 7.8% to 1.8%. The US figure differs from the rankings above because of a difference in method. However, the public page doesn’t say how many prompts the count includes. According to the company, 99% of Reddit citations point to specific threads.
What does Conversion’s study show about AI sources?
The Brazilian tech news site TI Inside reported the Panorama de GEO 2026, from the agency Conversion, on October 13. The article calls the study the 2027 edition, but Conversion’s page uses 2026. The data comes from 1,315 projects monitored on Conversion’s own platform, and Conversion sells GEO services. Of the projects, 98.33% are linked to Brazil, but the study warns that it isn’t a probability sample of the market.
Between May 9 and August 28, 2026, in ChatGPT, Gemini and AI Overviews, social networks accounted for 31.8% of citations, corporate pages for 24.2% and the press for 9.8%. Among social networks and communities, Reddit has the highest citation rate, that is, it appears in the answer 73.4% of the times it is consulted.
Do ChatGPT sources change from month to month?
They do, and fast, according to Ranqia, which sells brand diagnostics for AI answers. Between April and August 2026, it recorded 73.1 million ChatGPT citations in Portuguese-language business questions. In five months, for example, four different social networks took the lead: LinkedIn, Reddit, YouTube and Instagram. Similarly, among news sites, the newspaper Folha went from outside the top 50 in April to 1st place in August.
Does Reclame Aqui influence what AI says about your brand?
Reclame Aqui, a Brazilian consumer complaint site, appears among the sources cited by AI in Brazil. No public study, however, measures how much it changes the tone of the answer. Even so, it shows up in a significant share of the analyses, and consumers use AI to summarize reviews. So the brand’s page there needs to be in order.
In the Conversion study, Reclame Aqui appeared among the sources of 18.8% of the projects. As part of its GEO service, Conversion sells work on the brand’s Reclame Aqui profile. Profound puts forward the hypothesis that local sources, such as Reclame Aqui, may be more citable than Portuguese Wikipedia.
The survey O Mapa da Busca no Brasil 2026 (“The Search Map in Brazil 2026”) is by the SEO consultancy Optimiza Marketing. In it, summarizing consumer reviews accounted for 25% of answers about using AI when buying. The figure comes from the Brazilian retail news site GiroNews, which doesn’t disclose the sample.
Do reviews and ratings carry weight in AI answers?
One study shows that domains on review sites appear more often among the sources cited by ChatGPT, but that’s correlation, not causation. One possible explanation is that brands with many reviews tend to be bigger and better known. Even so, complete profiles and recent reviews give AI third-party material to describe the brand, and the company partly controls that.
SE Ranking analyzed 129,000 domains across 20 niches. Domains present on several review platforms, such as Trustpilot, G2 and Capterra, averaged 4.6 to 6.3 citations in ChatGPT. By comparison, those absent averaged 1.8. However, the study doesn’t state the country, and SE Ranking sells a ChatGPT visibility tool.
Asking for reviews is allowed, but buying them isn’t. Google Maps policy prohibits offering payment, discounts or free goods in exchange for reviews. It also prohibits asking only selected customers for positive reviews. So the allowed path is to ask everyone to review their real experience, with no incentive.
For businesses with a physical address, reviews on Google Business Profile follow this policy. For online stores, the same care applies to product reviews and marketplaces.
How do the press and digital PR get into AI answers?
The press gets into the answer as one of the sources cited by AI and as proof that third parties talk about the brand. Paid content, however, barely shows up. So digital PR for AI search means earning coverage with data, research or experts, not buying sponsored content. The press also carries more weight in questions about industry trends.
Muck Rack, which sells public relations software, analyzed more than 25 million links cited by ChatGPT, Claude and Gemini. In the May 2026 study, journalism accounted for about 27% of citations. In contrast, paid and advertorial content came to 0.3%, and press releases to 1.1%. Questions about trends cited journalism in 46% of answers, more than twice the rate of how-to and comparison questions. However, the report doesn’t state the country of the questions.
Brand mentions, with or without a link, track presence in AI. In an Ahrefs study of 75,000 brands, for example, brand mentions on the web had a 0.664 correlation with presence in AI Overviews. The number of backlinks, by comparison, came in at 0.218. But Ahrefs itself warns that correlation isn’t causation.
Coverage also opens the door to Wikipedia. Under the notability criteria of the Portuguese version, a topic needs significant coverage in reputable, independent sources, and press releases don’t count. For agencies, this puts digital PR in the agency’s GEO scope, alongside content and measurement.
Is it worth creating profiles or posts on Reddit to show up in AI?
It’s worth participating, as long as you do it openly and helpfully. However, fake profiles, paid posts without disclosure and coordinated votes violate Reddit’s rules and, in Brazil, may violate advertising law. The allowed path is to reply on behalf of the brand, disclose the affiliation and help the conversation. That holds even when the answer doesn’t favor the product.
Reddit’s rules call for authentic participation. For example, Rule 2 prohibits spam and content manipulation. Rule 5 also prohibits intentionally misleading other users or impersonating a person or entity. For posts with a commercial tie, Reddit offers the Brand Affiliate tag, which marks the post or comment as incentivized or having commercial intent.
What do the CDC, CONAR and Wikipedia require?
In Brazil, Article 36 of the Consumer Defense Code requires advertising to be identifiable as such, easily and immediately. Likewise, the influencer guide from CONAR, Brazil’s advertising self-regulation council, updated in 2026, requires clear identification of advertising content. Under the guide, an employment or partnership tie with the company can make content commercial.
| Practice | Allowed? | Rule |
|---|---|---|
| Replying as an employee, with the affiliation disclosed | Yes | Brand Affiliate tag |
| Paying someone to praise without disclosure | No | Reddit, CDC and CONAR |
| Using several accounts to boost a post | No | Reddit Rule 2 |
| Asking all customers for reviews | Yes | Google Maps |
| Giving a discount in exchange for a review | No | Google Maps |
| Editing the brand’s article without disclosure | No | Wikimedia Terms of Use |
| Proposing a correction on the article’s talk page | Yes | Wikipedia |
On Wikipedia, anyone who edits for pay must disclose their employer and client, according to the Wikimedia Terms of Use. The conflict of interest policy of the Portuguese version says that public relations edits are particularly frowned upon. For controversial edits, people with a conflict of interest are strongly encouraged to propose the change on the article’s talk page. However, removing criticism backed by a reliable source isn’t allowed.
How to gain presence in the sources cited by AI: step by step
The work starts by listing the sources cited by AI for your category, in Portuguese, and ends with monthly measurement. In between, each type of source calls for its own action: correct data in reference sources, real reviews, disclosed profiles in communities and earned press.
- List the cited domains. Run the category prompts in Portuguese in ChatGPT and other AI tools, then note the sources cited in each answer. With many prompts, use GEO tools that export the sources.
- Classify the sources as community, review, reputation, press, reference, marketplace and own site.
- Align the brand’s data across the site, profiles and directories: name, offer, price and address.
- Ask all customers for reviews, with no incentive, and respond to criticism publicly.
- Take care of Reclame Aqui with public replies on time, resolved cases and data that matches the site.
- Join communities with a disclosed profile. Answer the question asked and disclose the affiliation.
- Produce data for the press, such as original research and industry analysis.
- Measure every month. After all, ChatGPT sources change fast, as the change of leaders in the Ranqia data shows.
The evidence-based GEO tactics still apply to the site, while this work covers what’s outside it.
How do you correct wrong information that AI repeats about the brand?
Correct it at the source, not in the AI. Identify the page the answer cited and ask whoever published it for a correction. Also update the brand’s site with the right data and the date. If the source is Wikipedia, propose the change on the talk page, with your affiliation disclosed and an independent source.
If one of the cited sources fixes the page but Google still shows old text, use the Remove Outdated Content tool. However, flagging the error in the assistant isn’t enough; see how to report a ChatGPT error.
Then repeat the same prompts for a few weeks and see whether the error goes away. The protocol for how to measure sentiment and accuracy in answers shows how to log each error and the cited source.
What we found when we analyzed 11 Portuguese-language pages on the sources cited by AI
We ran 7 searches about the sources cited by AI. Then we analyzed 11 Brazilian Portuguese pages on the topic. Eight include numbers on cited sources. Five use data from outside Brazil: three don’t say which country it comes from, and one says so only in part.
Three include data from Brazil, but only one explains the study’s sample, period and limits. Three recommend joining communities such as Reddit, but only one says to disclose the affiliation with the brand. None mentions the Brand Affiliate tag, Google’s policy against incentivized reviews, the CDC or the CONAR guide. Finally, one suggests that the brand update its own Wikipedia article, without mentioning the conflict of interest rule.
Frequently asked questions
Hardly at all. In the Muck Rack study of more than 25 million links cited by ChatGPT, Claude and Gemini, paid and advertorial content came to 0.3% of citations. Journalism, by comparison, accounted for about 27%. So the way forward is to earn coverage with data, research or experts.
Rankings of ChatGPT sources change because each study has its own method, period and sample. Ahrefs, for example, measures the share among the 50 most cited domains, while 5W measures the share of the total. Compare rankings only when the engine, country, period and method are the same.
A brand shouldn’t edit its article without disclosing the affiliation. According to the Wikimedia Terms of Use, anyone who edits for pay must disclose their employer and client. For controversial edits, the Portuguese Wikipedia policy encourages people with a conflict of interest to propose the change on the article’s talk page.
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
- Ahrefs, The 50 Most-Cited Websites in ChatGPT, September edition (2026).
- Profound, prompt and citation behavior in Brazil, in Portuguese (2026).
- TI Inside, report in Portuguese on Conversion’s Panorama de GEO study of the sources cited by AI (2026).
- Muck Rack, What Is AI Reading?, May edition (2026).
- Reddit, Brand Affiliate tag for content with a commercial tie (2026).
- Google, prohibited and restricted content on Google Maps (2026).
