What AI engines cite in your industry: a synthesis of 2026 studies

What this synthesis covers

This post combines large studies from five publishers, released between November 2025 and July 2026, into one view of what AI engines cite, industry by industry. The measurement is theirs. Profound, Similarweb, Semrush and Conductor each collected the data, and each sells software or data in this category. What we add is the comparison across them, the gaps between them and a first move for each industry.

The studies use different samples, engines, periods and industry labels. Profound counts citations across eight engines; Similarweb measures how often ChatGPT cites anything at all; Semrush measures how concentrated answers are among brands; Conductor measures where Google shows AI Overviews. None of them is wrong because it disagrees with another. They are measuring different things.

Our view: no single citation study tells an industry what to do. Read together, they answer three separate questions. Which source types do engines lean on in my category? How often do they cite at all? How crowded is the answer? Each needs a different fix.

Owned (brand) citation

A link to a site the company controls: its website, documentation, help centre or blog.

Earned citation

A link to coverage the company does not control: news, reviews, analyst and trade publications, directories.

Social citation

A link to a platform where people post: Reddit, YouTube, LinkedIn, forums.

Google-owned citation

A link to google.com itself, such as a Business Profile or a product panel.

The studies do not all draw these lines in the same place, so we use them as a rough guide rather than a strict taxonomy. Where a study uses its own label, we say so.

The headline: brand sites win most citations, but not everywhere

Company-owned sites take the largest share of AI citations. Profound analysed 11.84 billion citations across eight engines, including ChatGPT, Gemini, Claude, Copilot, Perplexity, Grok, AI Overviews and AI Mode, from 16 April to 16 July 2026. About 57% went to company-owned sites.¹

The engine matters a great deal. Gemini sent 69% of its citations to brand sites and ChatGPT 47%, in the same data.¹ ChatGPT leaned most on earned media, at 30% of its citations, while Google's AI Overviews were lowest on earned media at 17.0%.¹

That has a practical meaning. A brand that wants to be cited by Gemini mostly needs its own pages in order. A brand that wants to be cited by ChatGPT needs those pages plus coverage on sites it does not control. We set the per-engine targets out in our note on brand sites versus third parties.

AI citations, April–July 2026

Own sites lead, by a margin that depends on the engine

~57%of citations across eight engines went to company-owned sitesProfound, Apr–Jul 2026 ¹

69%of Gemini's citations went to brand sitesProfound, Apr–Jul 2026 ¹

47%of ChatGPT's citations went to brand sitesProfound, Apr–Jul 2026 ¹

Owned, earned and social shares by industry

Industry moves the mix more than engine does. In Profound's April–July 2026 data, cybersecurity brands drew a median 74% of their citations from their own sites, the highest of any industry. Government and nonprofit organisations sat lowest, at 15.9%.¹

Earned media splits industries just as sharply. Pharma and biotech companies drew a median 59% of citations from earned media, while software companies drew a median 11.4%.¹ Social sources drove a median 16.5% of citations for fashion brands and 5% for pharma.¹

Source: Profound, 11.84 billion citations across eight engines, 16 April–16 July 2026. ¹ Profound reports these industries as examples, not a full table.

Our view: the pattern follows how buyers check claims in each industry. Security buyers read vendor documentation and threat research, so vendors' own pages carry weight. Health claims need independent confirmation, so engines lean on medical and news publishers. Fashion is about taste, and taste lives on social platforms.

Share of citations going to brand sites, by engine

Gemini69%

All eight engines~57%

ChatGPT47%

How often does ChatGPT cite a source at all, by sector?

Rarely overall, and far more in some sectors than others. Similarweb found the share of US ChatGPT prompts whose answer carried a citation rose from about 1.6% in June 2025 to about 6.8% in May 2026.² Most ChatGPT answers still link nothing.

The sector spread is wide. In May 2026, about 23% of travel and hospitality answers carried a citation, about 20% in automotive and under 4% in professional services.² Similarweb's explanation is that categories where people compare specific, checkable options, such as a flight, a car or a hotel, get cited more than categories built on general advice.

Our view: this changes what "winning" means for a law firm or a consultancy. When ChatGPT rarely links anything, being named in the answer matters more than being cited. For travel and automotive, the citation itself is a realistic target, and the page it points to needs to convert. Our note on Similarweb's sector data goes further into why.

How crowded is the answer in each industry?

Very crowded in some categories, open in others. Semrush's 2026 AI Visibility Index, built on 126 million US prompts from January to April 2026, measured how much of each industry's AI visibility the top three brands hold.³

Share of AI visibility held by the top three brands

News and media82.9%

Consumer electronics76.9%

Industrial42.2%

Finance41.4%

The same index found ChatGPT cited about 15 sources per response against about 3 for Gemini (US, January–April 2026).³ Put that next to Profound's finding that Gemini leans hardest on brand sites, and the picture is consistent: Gemini reads few pages and favours the brand's own, while ChatGPT reads widely and mixes in more third parties.

Concentration tells you how hard the fight is. In news or consumer electronics, a mid-sized brand is competing for what is left after three leaders. In finance or industrial categories, there is room for specialists that answer narrow questions well, which is the case we made for financial firms in Singapore.

Where does Google show AI Overviews, by industry?

Mostly where questions are about health and money, least where people are buying a specific thing. Conductor's 2026 benchmarks report covered 21.9 million US Google searches from 15 September to 12 October 2025, and found AI Overviews on 25.11% of them.⁴ That data is now about ten months old.

The industry spread in Conductor's data runs from 48.75% of searches in health care to 4.48% in real estate. Financials sat at 25.79%, utilities at 25.4% and consumer staples at 6.82%.⁴ Conductor sells an SEO and AI visibility platform.

A second study gives a similar shape. Ahrefs found AI Overviews on 43.0% of health searches and 3.2% of shopping searches.⁷ Its sample was 146 million desktop results from September 2025, with the geography not stated, so the data is now close to a year old. Two studies with different samples agree that health sits near the top and buying-intent categories near the bottom.

Google is citing itself more in AI Mode

For local and product questions, Google's own pages now compete for the citation. Profound found that from 15 April to 30 June 2026 google.com's citation share in AI Mode grew 8.4 times, making it AI Mode's second most-cited domain.⁵

The growth came from two Google-owned assets: Business Profiles and Product Knowledge Panels. Profound names hospitality, home services, restaurants, real estate and healthcare as the most affected local verticals, based on more than 32 million Business Profile instances, with product-comparison queries also shifting towards Knowledge Panels.⁵

Our view: for local businesses and retailers, the Business Profile and product data now belong in the same workstream as the website. They are no longer a listings chore. If the profile is incomplete or out of date, AI Mode will cite it anyway. Our source map for AI Mode breaks this down by query type.

Which industries get the most AI referral traffic?

Technology, by a wide margin, at least in older data. Conductor's benchmarks measured US AI referrals across 1,215 of its enterprise customers' domains in ten industries between May and September 2025, data now more than ten months old. AI referrals made up 1.08% of visits overall, 2.80% in information technology and 0.25% in communication services.⁶

In the same data, ChatGPT sent 87.4% of AI referrals.⁶ Later studies put ChatGPT's share lower, and Similarweb reports its share of generative AI web traffic falling over the following year, so treat the engine split as a snapshot. Our comparison of referral-share studies explains why the figures differ.

Referrals and citations are not the same measure. An industry can be cited often and still see few referral visits, because many people read the answer and search for the brand later. Use referral data to prioritise landing pages, not to judge visibility.

What each industry should fix first

Match the fix to the source type your industry's answers lean on. The table below is our reading of the combined studies. It is judgement, not measurement, and it should be checked against your own prompts.

Our judgement from the studies cited above. Illustrative: industries and engines vary, and the studies use different industry labels.

Our industry pages go further for software companies, health care providers and travel brands. Whatever the industry, the order is the same: confirm the source types engines use for your prompts, then fix the pages of that type you control or can influence.

Notice what the example does not do. It does not start with a content calendar or a new blog. It starts with the source types already shaping the answer, which is the cheapest place to change it.

What the studies cannot tell you yet

Four gaps matter for anyone applying these numbers. Most of the data is US or does not state a geography, so patterns in other countries and languages are largely untested. The industry labels differ between studies, so "finance" in one is not "financials" in another. Every publisher sells in the category, which is not a reason to dismiss the data, but it is a reason to note the sample.

How to read any citation study before acting on it

1. Check the period

   When was the data collected, and how much has the engine changed since?

2. Check the geography

   US-only data says little about answers in other markets.

3. Check the engines

   A ChatGPT-only finding may not hold for Gemini or AI Mode.

4. Check the unit

   Share of citations, share of answers and share of visibility are different measures.

5. Check the seller

   Note what the publisher sells, and whose customers are in the sample.

The fourth gap is volatility. Semrush's weekly snapshots from July to October 2025 showed Reddit falling from roughly 60% of ChatGPT responses to around 10% within weeks.⁸ An industry pattern measured in one quarter can shift in the next, which is why we covered which citation tactics have evidence behind them separately.

Our view: use these studies to choose what to test, not what to conclude. The only citation data that settles a decision for your company is data from your own prompts, run on several days.

How to get your own industry's numbers

Write the thirty to fifty questions your buyers ask before choosing a supplier in your category, in the countries you sell to. Run them across engines, on more than one day, and record every source each answer cites. Classify those sources as your own, earned, social or Google-owned. Within a few weeks you will have your own version of the tables above.

At Sigzen AI, our AI answer monitor runs that kind of prompt set across six engines: ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI. It uses official APIs where offered and a licensed AI-answer data provider otherwise. Our audits run each prompt on three separate days and classify every cited source, so the share figures rest on more than a single run.

The studies agree on one thing: there is no single citation strategy that fits every industry. A software company and a hospital face different engines, different sources and different odds of being linked at all. Start from your industry's pattern, then replace it with your own data as soon as you have it.