Blog · How AI models recommend brands
How ChatGPT, Claude and Perplexity Recommend Brands
Dharini Shah · September 4, 2026
AI assistants recommend brands by combining two things: what the model learned during training and what it retrieves from the web when answering. The balance between those two sources, the sites each system prefers to retrieve from and how visibly each system cites its sources differ from one product to the next. That is why the same buyer question can produce a different shortlist in ChatGPT, Claude, Perplexity and Google AI Mode.
For marketers, the implication is simple: visibility in one model does not guarantee visibility in another. This article explains how the major assistants build recommendations, what public research says about the sources each one favors, and how to adapt your strategy without creating a separate playbook for every model.
How do AI models decide which brands to recommend?
Every modern AI assistant builds a brand recommendation from some mix of three inputs.
1. Parametric knowledge. This is what the model absorbed during training from a large body of text. If your brand was widely and consistently described across the web before the model's training cutoff, the model may "know" you without searching. If it was not, the model may omit you or describe you inaccurately.
2. Retrieved content. When an assistant searches the web, it pulls live pages, reads relevant passages and grounds its answer in them. Retrieval is how newer brands, recent launches and updated pricing can enter an answer. It is also where crawler access, rankings and page clarity matter.
3. Conversation context. The user's wording, stated constraints, location, earlier messages and, in some products, memory of past chats all shape the answer. A prompt that says "for a 10-person startup with no budget" will surface different brands than "for an enterprise with SOC 2 requirements."
Because generation involves sampling, the output also varies from run to run. SparkToro and Gumshoe's research collected 2,961 prompt runs across ChatGPT, Claude and Google's AI tools and found less than a 1 in 100 chance of getting the same brand list twice from ChatGPT or Google's AI. Claude was somewhat more consistent, but identical lists were still rare. Recommendations are best thought of as probabilities, not fixed positions.
How ChatGPT recommends brands
ChatGPT can answer from trained knowledge alone or search the web, depending on the question and the user's settings. When it searches, it relies on OpenAI's crawler infrastructure. OpenAI states that "sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." A separate crawler, GPTBot, collects content for model training, and ChatGPT-User fetches pages when a user asks ChatGPT to visit them.
What public research says about ChatGPT's sourcing:
- Encyclopedic sources rank high. Profound's analysis of 680 million citations from August 2024 to June 2025 found Wikipedia was ChatGPT's most cited domain, at 7.8% of all citations.
- Fresher content is favored. Ahrefs analyzed about 17 million citations and found ChatGPT showed the strongest preference for newer content among the assistants studied, citing pages that were, on average, hundreds of days newer than those in organic Google results.
What this means for brands: ChatGPT rewards a clear, widely corroborated entity description. Make sure your company is described consistently across your site, reputable directories and, where you legitimately qualify, reference sources. Keep key commercial pages current, and confirm OAI-SearchBot is not blocked by robots.txt or your CDN.
How Claude recommends brands
Claude answers from trained knowledge by default and can search the web when web search is enabled. Anthropic operates three crawlers with separate purposes, as summarized by Search Engine Land: ClaudeBot collects content that may be used for training, Claude-SearchBot indexes content to improve search results, and Claude-User fetches pages when a user's question requires it. Each respects robots.txt, and each needs its own rule.
In the SparkToro study, Claude produced somewhat more consistent brand lists than ChatGPT or Google's AI, though repeated identical lists remained uncommon. Public, large-scale citation research on Claude is thinner than for ChatGPT and Perplexity, so be cautious with anyone claiming precise knowledge of Claude's source preferences.
What this means for brands: Blocking ClaudeBot to opt out of training is a legitimate choice, but blocking Claude-SearchBot or Claude-User reduces your chance of appearing when Claude searches. Review those rules separately. Clear, factual pages with specific claims give Claude something concrete to cite.
How Perplexity recommends brands
Perplexity is built around retrieval. It searches for almost every answer and displays numbered citations prominently. Perplexity's documentation states that PerplexityBot is "designed to surface and link websites in search results on Perplexity" and is not used to crawl content for training foundation models. Perplexity-User fetches pages for user-initiated requests.
What public research says:
- Community sources dominate. Profound found Reddit was Perplexity's top cited domain, at 6.6% of citations, and accounted for nearly half of citations among Perplexity's top 10 sources.
- Freshness matters. Ahrefs observed that Perplexity, like ChatGPT, tends to order references from newest to oldest.
What this means for brands: Perplexity is the most sensitive of the major assistants to what is currently published about you. Genuine community discussion, recent reviews and up-to-date comparison pages carry weight. Because citations are visible, users can also click through, which makes Perplexity a meaningful referral source relative to its size.
How Google AI Overviews and AI Mode recommend brands
Google's AI features draw on Google's search index. Google says AI Overviews and AI Mode use a "query fan-out" technique that issues multiple related searches across subtopics, and that no special optimizations are required beyond standard SEO eligibility.
What public research says:
- Community and video sources are prominent. Profound found Reddit, YouTube and Quora were the top three cited domains in AI Overviews. The Pew Research Center similarly found Wikipedia, YouTube and Reddit were the most frequently linked sources in Google's AI summaries.
- Brand mentions correlate with visibility. Ahrefs' study of 75,000 brands found branded web mentions had the strongest correlation with AI Overview visibility of the factors tested. In a 2026 follow-up, Ahrefs reported that YouTube mentions were the strongest signal.
- Freshness preference is weak. Ahrefs found AI Overviews cited content of roughly the same age as organic results.
What this means for brands: Strong SEO across a topic cluster is the entry ticket, because fan-out sub-queries need pages to retrieve. Beyond that, presence on YouTube and in community discussions appears to matter.
Side-by-side comparison
| Model | Main recommendation inputs | Sources public research highlights | Visible citations | Practical priority |
|---|---|---|---|---|
| ChatGPT | Trained knowledge plus web search via OAI-SearchBot | Wikipedia, news and reference sites; favors fresher content | Yes, when searching | Consistent entity facts, fresh commercial pages, crawler access |
| Claude | Trained knowledge plus optional web search via Claude-SearchBot | Limited public large-scale data | Yes, when searching | Crawler rules per bot, clear factual pages |
| Perplexity | Retrieval-first on nearly every answer | Reddit and community sources, recent content | Yes, prominent | Current reviews, community presence, up-to-date comparisons |
| Google AI Overviews and AI Mode | Google index with query fan-out | Reddit, YouTube, Quora, Wikipedia | Yes, as links | Topic-cluster SEO, video and community presence, brand mentions |
Citation studies are snapshots. Source mixes shift as products change, so treat these patterns as tendencies to test against your own category rather than fixed rules.
Why the same brand gets different answers in different models
Brands are usually surprised by three kinds of cross-model gaps.
Knowledge gaps. A company founded recently may be absent from a model's trained knowledge and appear only when the model searches. If the user's question does not trigger search, the brand disappears.
Source gaps. If Perplexity leans on community threads and your brand is never discussed there, you may appear in ChatGPT but not Perplexity, even though your website is identical for both.
Description gaps. Models can repeat outdated or wrong information. Ahrefs' 2026 benchmark reported that most AI models it tested repeated fabricated claims as fact even when official sources contradicted them. If an old review describes your pricing wrongly, one model may repeat it while another does not.
How to improve brand recommendations across AI models
Building a separate strategy for every model is rarely worth it. The shared levers explain most of the difference. A practical order of operations:
- Check access. Review robots.txt and CDN bot rules for OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot and Googlebot. Decide on training crawlers such as GPTBot, ClaudeBot and Google-Extended separately.
- Define your brand facts once. Document category, audience, core use cases, pricing approach and differentiators. Align your website, docs and third-party profiles to that source.
- Answer buyer questions directly. Create pages that state plainly who you are for, what you cost or how pricing works, how you compare with alternatives and what proof exists.
- Close source gaps model by model. Look at which domains each model cites for your category prompts. If Perplexity cites community threads and Google cites YouTube, those are the channels to earn legitimate presence in.
- Correct misinformation at the source. Update outdated review profiles, directory listings and partner pages. Models retrieving those pages will pick up the change over time.
- Measure per model, act across models. Track visibility rate, citation rate and description accuracy separately for each assistant, then prioritize fixes that improve several at once.
A hypothetical example
Imagine a hypothetical HR software startup that appears in 40% of sampled ChatGPT answers for "best HR software for small businesses" but almost never in Perplexity. Reviewing Perplexity's citations for those prompts shows it relies on two recent Reddit threads and a comparison article published last month, none of which mention the startup. ChatGPT, meanwhile, cites the startup's own detailed pricing page. The fix is not more on-site content. It is legitimate participation in community discussion and outreach to the comparison publisher.
How Bob Builds AI helps
Bob Builds AI's Visibility Monitoring tracks how brands appear across AI models including ChatGPT, Gemini, Claude, Perplexity and Copilot, with visibility rate, citation rate, competitor positioning and the sources behind each answer. That makes cross-model gaps visible so teams can see which fixes will help several models at once. Brand Memory keeps the underlying brand facts consistent across the content and workflows that feed those models.
FAQ
Why does ChatGPT recommend my competitor but not my brand?
Usually because the competitor is more widely and consistently described in sources ChatGPT learned from or retrieves. Check whether OAI-SearchBot can crawl your site, whether your pages clearly state who you serve, and whether reference sites, reviews and comparison articles mention you. A competitor with more third-party coverage often wins even with a weaker website.
Do AI models use the same sources to recommend brands?
No. Public citation studies show different tendencies. Profound's analysis found Wikipedia was ChatGPT's top cited domain, while Reddit led for Perplexity and Google AI Overviews. These patterns change over time and vary by category, so check the sources each model actually cites for your own prompts.
Is Perplexity easier to get recommended in than ChatGPT?
Perplexity retrieves live content for almost every answer, so recent changes can show up faster than in a model answering from trained knowledge. That can make Perplexity more responsive to new reviews, community discussion and updated pages. It does not make it easier in every category, because competitors benefit from the same responsiveness.
Can I block AI training but still appear in AI search?
Yes, for several providers. OpenAI separates GPTBot for training from OAI-SearchBot for search. Anthropic separates ClaudeBot for training from Claude-SearchBot and Claude-User. Google offers Google-Extended for Gemini training controls, while AI Overviews follow normal Googlebot rules. Configure each user agent deliberately.
Why do AI recommendations change every time I ask?
AI models generate answers probabilistically, and searching models may retrieve different pages on each run. SparkToro and Gumshoe found that the same brand list rarely repeats across runs. Reliable measurement requires running the same prompts many times and tracking the percentage of answers that include your brand.
Does Claude recommend brands differently from ChatGPT?
Claude's recommendations draw on its own training data and, when enabled, its own search index built by Claude-SearchBot. In SparkToro's research, Claude's lists were somewhat more consistent than ChatGPT's. Public large-scale data on Claude's citation sources is limited, so test your own category prompts rather than assuming ChatGPT results transfer.
How often should I check my visibility across AI models?
For most brands, a weekly or biweekly sample of priority prompts is enough to spot trends, with more frequent checks around launches, pricing changes or competitive moves. Because individual answers vary, trends across many runs matter more than any single response.
Conclusion
ChatGPT, Claude, Perplexity and Google's AI features all recommend brands by blending trained knowledge with retrieved content, but they weigh those inputs differently and favor different sources. ChatGPT rewards a clear, corroborated entity and fresh pages. Perplexity leans heavily on current, community-driven sources. Google's AI features build on Google's index and appear to favor brands with broad mentions, including video.
The next step is to measure, not guess. Pick 20 to 30 prompts your buyers actually ask, run them repeatedly across the assistants your audience uses, and note which sources each model cites. The gaps will show you where to act. If you want that tracking done continuously across models, Bob Builds AI can set it up and turn the findings into prioritized fixes.