ChatGPT Favours OpenAI: What Self-Citation Bias Means for Brands

ChatGPT recommends OpenAI's own models twice as often as rival AI engines recommend those same models, according to new research from 5W AI Communications. Gemini and Google AI Overviews show similar patterns, favouring their own parent companies, while Claude is the only major assistant that doesn't. For any brand relying on AI engines to be discovered, this means the "shelf" you're competing on isn't neutral, and the platform answering the question may already have a favourite.

What changed?

5W AI Communications published a synthesis of its 2026 research library, pulling together findings on how ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews describe the companies, categories and brands that people now research inside an AI engine before opening a browser tab, as detailed in The State of AI Citations 2026.

The headline finding is stark. A dedicated benchmark, the AI Companies AI Visibility Index 2026, found that OpenAI captures 24.6% of all AI citations about the AI industry, and every major assistant except one shows measurable self-citation bias. ChatGPT recommends OpenAI models 2.0x more often than other engines recommend the same OpenAI models, Gemini recommends Google DeepMind 1.7x more often, and Google AI Overviews recommends Google 1.6x more often, per PR Newswire.

The one exception stands out. Every major AI engine except Claude shows measurable self-citation lift, which 5W's researchers describe as a structural fact that communications teams need to account for, in the same way they already account for cable-news lean or platform algorithm changes, according to Yahoo Finance.

The underlying benchmark is substantial in scale. The AI Companies AI Visibility Index 2026 analysed 32,200 prompts across ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews, run in two independent waves to test for stability across retrieval drift, model updates and cross-wave volatility, according to the 5W benchmark.

There's a wrinkle worth flagging too. The source mix behind these answers is unusual: across two waves, GitHub and ArXiv accounted for 31% of citations about AI companies, second only to Wikipedia, according to 5W's State of AI Citations 2026. No other industry 5W has measured surfaces code and research papers as primary sources, a very different citation diet from most consumer categories, where editorial press and reviews dominate.

Alongside the AI-companies finding, the wider 5W research library points to a bigger shift in how people find brands generally. Companion studies find Wikipedia and Reddit out-cite the entire prestige business press, and 35% of consumers now begin product research inside an AI engine rather than Google, per the same 5W research.

Why does this matter for your business?

If you sell AI tools, cloud infrastructure or anything adjacent to the big model providers, you're not just competing against other vendors. You're competing against a platform with a built-in incentive to talk up its own family of products when someone asks it a question.

This isn't necessarily manipulation in the crude sense. As the researchers put it, self-citation lift is "not a scandal", it's closer to a structural bias that any team building a visibility strategy now has to plan around, in the same way PR teams already factor in outlet bias. But structural or not, it changes the maths for anyone trying to get recommended by a chatbot that has skin in the game.

It also reinforces a theme that keeps surfacing in AI visibility research this year: citation sources vary enormously by category, and assuming one playbook works everywhere is a mistake. The AI-industry category leans on code repositories and academic papers, while most consumer categories still lean on press, reviews and community discussion such as Reddit. Brands need to know which citation ecosystem their category actually sits in before deciding where to invest.

For founders and marketers outside the AI industry itself, the takeaway is smaller but still real: the assistant answering your prospective customer's question might have a thumb on the scale for certain adjacent products or affiliated services. Factor that in when you interpret any visibility gap you're seeing between engines.

What should you do now?

Start by identifying whether your category has any of this self-referential dynamic. If you compete with, partner with, or get compared against a platform's own products, benchmark your visibility against that platform specifically, not just against AI engines in general.

Check where your category's citations actually come from. If you're in a research-heavy or technical space, code repositories, academic papers and primary documentation may matter more than press coverage. If you're in a consumer space, Reddit, reviews and editorial press are likely doing more of the work.

Track your own citation share across multiple engines rather than relying on just one. Because bias direction differs by platform, a brand that looks weak on ChatGPT might look stronger on Claude, and vice versa. You need the full picture, not a single snapshot; a free audit from Sited is a straightforward way to see how your brand actually shows up across different AI engines before you commit budget to fixing anything.

Finally, don't treat this as a one-off study to file away. The 5W research library is explicitly designed as an ongoing series, with quarterly releases planned through 2026, according to 5W's own research page. Self-citation bias, category-specific citation patterns and platform-level quirks are all things that will keep shifting, so this needs to be a recurring check, not a one-time read.

Frequently asked questions

What is self-citation bias in AI search?

It's when an AI assistant recommends or cites its own parent company's products more often than competing AI engines recommend those same products when asked identical questions. A lift of 2.0x means the engine recommends its parent company twice as often as other engines do in identical prompts.

Which AI assistants show this bias?

According to the 5W research, ChatGPT, Gemini and Google AI Overviews all showed measurable self-citation lift toward their own parent companies. Claude shows the lowest self-citation lift of the major assistants studied.

Does this affect brands outside the AI industry?

Directly, the strongest effect is in the AI-companies category itself. But the wider point, that citation sources and platform behaviour vary sharply by category and by engine, applies to any business trying to build AI search visibility.

How was this data gathered?

The findings come from 5W AI Communications' 2026 research library, including a benchmark that analysed 32,200 prompts across ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews, run in two independent waves, plus a wider synthesis covering citations across the same five engines.

What should marketing teams do with this information?

Treat platform bias as a known variable, similar to outlet lean in traditional PR, and measure your brand's citation share across each engine separately rather than assuming uniform treatment. Auditing your AI visibility regularly, rather than checking once, is the only way to catch shifts as engines and their citation preferences change.