Do AI engines cite brand websites or third parties? Set the target per engine
Profound's study of 11.84 billion citations shows each engine mixes owned, earned and social sources differently. How to set an owned-versus-earned target.
The short answer
Both, in proportions that differ by engine. In Profound's study of 11.84 billion citations (16 April to 16 July 2026), about 57% went to company-owned sites overall, 69% on Gemini and 47% on ChatGPT. Set a separate owned-versus-earned target for each engine and your industry, instead of one number for AI search as a whole.
Key takeaways
- In Profound's study of 11.84 billion citations across eight engines (16 April to 16 July 2026), about 57% went to company-owned sites.
- The mix differs by engine: company sites took 69% of Gemini's citations and 47% of ChatGPT's, while earned media took 30% of ChatGPT's.
- Industry moves the mix as much as the engine does: cybersecurity's median brand share was 74%, while pharma and biotech drew a median 59% from earned media.
- Company-owned means any company's site, not yours, so the study tells you where engines look, not how often they cite you.
- Our view: set an owned target for engines that favour company sites and an earned target for engines that lean on media and communities.
In this article
What the Profound study found
Most AI citations go to company websites, but the share swings widely by engine and industry. Profound, which sells AI visibility software, analysed 11.84 billion citations from eight engines between 16 April and 16 July 2026, published on 30 July.1 About 57% of citations went to sites operated by companies.1
The engine split is where planning starts. Company sites took 69% of Gemini's citations and 47% of ChatGPT's, the lowest among the major engines.1 ChatGPT made up the difference with earned media, which took 30% of its citations, against 17.0% for Google AI Overviews.1
Social and community sources vary too. They took a median 15.3% of AI Overviews citations and 14.4% of AI Mode's, which Profound puts at roughly 1.3 times ChatGPT's rate and 4 times Microsoft Copilot's.1 Profound's eight engines included Grok alongside the six we measure.
16 April to 16 July 2026, eight engines
Share of citations going to company-owned sites
What "company-owned" does and does not mean
It means a site run by any company, not your site. When Gemini cites company sites 69% of the time, it may be citing your competitors, a software vendor's help centre or a retailer's product page.1 The study tells you where engines look, not how often they cite you.
That distinction matters for targets. A brand with thin product pages can sit in a category where engines cite company sites heavily and still get almost none of those citations. Our view: read the study as a map of which door each engine prefers, then measure separately whether yours is one of the doors it uses.
Does your industry change the mix?
Yes, by as much as the engine does. Profound reports a median brand share of 74% in cybersecurity and 15.9% for government and nonprofit topics.1 Pharma and biotech drew a median 59% of citations from earned media, against 11.4% for SaaS and software, and fashion drew about 16.5% from social sources.1
The pattern makes sense. Buyers of security software ask about specifications and integrations that vendors document best. Questions about medicines lean on journals, regulators and health publishers, and engines treat a manufacturer's own claims with more caution. Fashion runs on what people wear and say.
Semrush's weekly snapshots from 14 July to 12 October 2025 point the same way for communities. Reddit and Wikipedia were ChatGPT's two most-cited domains, and LinkedIn appeared in about 15% of Google AI Mode responses.2 Our source map for AI Mode by query type breaks down how that mix shifts between product, comparison and informational questions.
Why do engines mix sources so differently?
Because each engine retrieves and ranks pages in its own way, and the studies describe outcomes rather than mechanisms. Profound reports the mix; it does not claim to know why Gemini prefers company pages or why ChatGPT reaches for news and reviews. We will not guess at their internals either.
What a brand can act on is the observed pattern. An engine that favours company pages rewards clear, specific pages it can quote: specifications, prices, policies and comparisons stated in plain sentences. An engine that leans on earned media rewards being written about by others. Our view: the two need different teams, budgets and timelines, which is the strongest reason to plan them per engine.
Two kinds of work, two kinds of target
Owned target
- Measured as your pages cited
- Moved by page rewrites and structure
- Results within weeks of recrawl
Earned target
- Measured as third-party pages naming you
- Moved by coverage, reviews and listings
- Results over months
How to set an owned-versus-earned target per engine
Start from the engine's baseline mix in your industry, then set two targets for each engine: the share of answers that cite your own pages, and the share that cite third-party pages mentioning you. Weight effort by where each engine actually looks.
| Engine pattern | Primary target | Work that moves it |
|---|---|---|
| Favours company sites, like Gemini | Your pages cited | Answer-first product, pricing and spec pages |
| Leans on earned media, like ChatGPT | Third-party pages naming you | Reviews, trade press, analyst and comparison coverage |
| Uses communities, like AI Overviews and AI Mode | Accurate mentions in discussions | Helpful participation, not seeded posts |
Our recommended targets by engine pattern, based on the Profound mix. Check your industry's numbers before you set yours.
Correlational evidence supports investing in earned mentions. An Ahrefs study of 75,000 brands, published in May 2025 and now dated, found that branded web mentions correlated at 0.664 with AI Overview visibility, against 0.218 for backlinks.3 That is a correlation, not proof that mentions cause citations, and the engines have changed since.
Our view: one blended "AI citation share" hides the decision you need to make. A team that reports a single number cannot tell whether to fix pages or win coverage. Two numbers per engine can.
What should you measure to know it worked?
Measure citations by source type for each engine, for your own prompt set, at a fixed interval. For every answer, record whether you are named, whether one of your pages is cited and which third-party pages are cited. Group the third-party pages as media, reviews and directories, or communities.
Our AI answer monitor does this across six engines: ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI. It uses official APIs where an engine offers one and a licensed AI-answer data provider where it does not. In a Sigzen AI audit, each prompt runs on three separate days, because a single run can mislead.
Remember how rarely some answers link at all. In Similarweb's US data, only about 6.8% of ChatGPT answers carried a citation in May 2026, as our note on citation rates by sector explains.4 Profound's study counts citations where they exist; for most answers, being named is still the outcome. Tactics such as schema and llms.txt come up in every planning meeting, and our review of the evidence on schema, llms.txt and Reddit covers what the studies support.
Sources
- Profound, where AI citations come from: 11.84B citations, 8 engines, 16 Apr–16 Jul 2026 (Jul 2026)
- Semrush, most-cited domains in AI search: weekly snapshots 14 Jul–12 Oct 2025 (Nov 2025)
- Ahrefs, brand mentions and AI Overview visibility, correlational (May 2025)(dated)
- Similarweb, AI search stats 2026: ChatGPT answers with a citation, US, Jun 2025–May 2026 (Jul 2026)
