Skip to content

Review of 32 AI search studies finds fewer clicks and uncertain brand visibility

Kevin Indig’s review of 32 studies finds that AI answers reduce website clicks, while citations frequently fail to name the source brand. AI recommendations can influence buyer shortlists, but research has yet to establish which optimization methods reliably improve visibility.

Review of 32 AI search studies finds fewer clicks and uncertain brand visibility

AI-generated search answers are reducing visits to external websites while introducing new ways for brands to appear in purchasing decisions, according to a review of 32 research studies by organic growth adviser Kevin Indig. The analysis published by Search Engine Land examines changes in search behavior, the selection of sources for AI citations and the influence of chatbot recommendations on consumer decisions.

The review draws on academic experiments, browsing data and industry research rather than presenting results from a new experiment. Its findings suggest that established search rankings are an incomplete measure of brand visibility in ChatGPT, Gemini and other AI interfaces, where a website may be retrieved without being cited and a brand may be recommended without a link to its own content. Indig also identifies substantial differences between studies, cautioning against universal claims about traffic losses or the effectiveness of individual optimization techniques.

01 / 03

AI search reduces external clicks, although estimates vary considerably

Research reviewed by Indig provides increasingly consistent evidence that AI-generated answers can reduce the frequency with which users visit websites from search results. The size of the reduction varies according to the search interface, the type of query and the study's methodology. The evidence is strongest for particular search experiences rather than a single industry-wide estimate of traffic lost to artificial intelligence.

A Pew Research Center analysis of browsing activity found that users clicked a traditional search result in 8% of visits when a Google AI summary appeared, compared with 15% of visits without one. Links included inside the AI-generated summary received clicks in approximately 1% of visits featuring that summary. The study examined browsing behavior among 900 US adults and did not establish that the same differences apply to every search category.

More recent experimental work has provided evidence of a causal relationship between AI-generated search features and reduced outbound traffic. In a randomized field experiment by researchers Agarwal and Sen, the appearance of an AI Overview reduced organic outbound clicks by 39.8% and increased zero-click searches by 34.5%. A separate experiment involving 1,100 US participants found that directing users into Google's AI Mode reduced external click-through by 18.8 percentage points compared with ordinary Google search.

Other research complicates attempts to apply those figures across the wider market. Semrush and Datos examined more than 200,000 keywords and found a decline in the measured zero-click rate for one subset of searches that began displaying AI Overviews. Indig notes that zero-click rates and click-through rates measure different aspects of behavior, while variations in keyword selection can produce apparently conflicting results.

The analysis also distinguishes the adoption of AI search from the displacement of conventional search engines. A Pew survey found that 49% of US adults had used an AI chatbot at least once, while a separate Bain survey found that 56% of respondents still primarily or consistently began their research with search engines. Those findings describe different populations and behaviors, making it difficult to calculate how many conventional searches have actually been replaced.

For publishers and companies dependent on organic discovery, the findings establish a growing separation between appearing in search and receiving website traffic. They do not show that every site will experience the same decline or that all traffic associated with AI-assisted research can be identified through conventional referral analytics.

02 / 03

Being retrieved or cited does not guarantee a brand will appear in an AI answer

The review identifies important differences between a website being discovered by an AI search system, selected as a cited source and named in the final response. Traditional SEO reporting generally measures the position of a page within search results. AI-generated answers introduce additional decisions about which retrieved pages to use and whether the resulting response explicitly identifies the organizations behind them.

An AirOps study of 15,000 prompts and 548,534 retrieved pages found that 85% of the pages ChatGPT retrieved were never cited in its final responses. Among cited pages appearing in the top 20 search results, 32.9% were found only through additional searches generated during the answering process, rather than the user's original query. The findings suggest that tracking the initial prompt alone can miss important parts of how AI systems discover information.

Source position was also associated with citation frequency. In research conducted by Indig and AirOps, the first retrieved page had a 58% citation rate, compared with 14% for pages further down the retrieval results. Other experiments found that topical relevance and document position affected selection, although the magnitude of those relationships depended on the testing environment.

Even an earned citation may provide limited visibility to the organization responsible for the source. In a study conducted with Semrush, 61.7% of 3,981 recorded domain appearances involved citations without the associated brand being named. Another 25.1% were brand mentions without a citation, while only 13.2% combined both forms of visibility.

Those results distinguish source attribution from brand recognition. A user can receive information supported by a company's research without seeing the company's name, while a competing brand may appear prominently in the generated recommendation without receiving a direct citation. For corporate communications and reputation monitoring, a citation count alone therefore provides an incomplete account of how an organization is represented to users.

The review also questions widely promoted optimization tactics. Some studies found associations between citations and features such as structured content or research originality, but controlled benchmarks produced inconsistent results when documents were modified to include additional statistics or other supposedly advantageous features. Indig concludes that topical relevance and retrieval performance have more consistent research support than claims that a particular formatting change reliably produces citations.

These findings require methodological caution because several experiments supplied documents directly to language models rather than measuring whether the systems would discover those documents independently. Results from such tests can help explain source selection once material is available, but they cannot establish that the same changes will improve visibility in live search products.

03 / 03

AI recommendations influence buyer consideration before the website visit

The reviewed research indicates that AI recommendations can affect which companies enter a buyer's consideration set, even when the interaction produces no immediate click. In a Growth Memo survey conducted with Wynter, 145 of 200 respondents reported visiting vendor websites after using an AI chatbot for product research. Among respondents who already had potential vendors in mind, 62.7% said the chatbot introduced another company they subsequently considered.

Other research observed more decisions taking place within the AI interface. In a study involving 224 ChatGPT shopping tasks, researchers at Profound found that 92.8% of codable tasks ended without a meaningful visit to the open web. Brands selected by participants had appeared more frequently in AI answers than brands that were rejected, although the observed relationship does not prove that visibility caused the purchasing preference.

A separate Profound analysis of an opt-in browsing panel found that visits to brand websites within seven days were 2.07 percentage points higher after a ChatGPT brand mention than a forecast based on those users' previous behavior. The result suggests that some AI influence may appear in later website activity, but the observational design cannot establish whether the mention caused the visit.

The evidence also identifies limits to consumers' willingness to delegate purchasing decisions. A Gartner survey found that 31% of respondents would allow AI to narrow household-supply choices and 28% would do so for personal electronics, while willingness to let the system make the final purchase decision remained much lower. In Visa's benchmark of 22 models, AI systems passed 57.2% of shopping-related test attempts, with the assessment covering product selection decisions rather than completed payments or orders.

The implications for brand measurement extend beyond website visits attributed directly to AI platforms. A buyer may encounter a company in an AI recommendation, investigate it through traditional search and later visit its website independently. Indig recommends examining brand mentions alongside qualified visits and subsequent business outcomes, while acknowledging that existing attribution methods cannot reliably isolate every contribution made by an AI-generated answer.

The review's methodology places limits on the conclusions that can be drawn across its 32 included studies. The research covers different countries, consumer populations and AI products, with evidence drawn from surveys, experiments and simulated tasks. Indig did not combine the findings into a single statistical estimate, and several studies were conducted by commercial providers of AI visibility or analytics services.

The article is an updated synthesis of existing research rather than an announcement of a platform change or a new independent dataset. Its findings indicate that AI interfaces are altering how companies are discovered and evaluated, while the reliability of visibility measurements remains uneven. Whether individual optimization practices produce sustained improvements in brand recognition or commercial performance requires further testing under comparable conditions.

Related

Latest

Reputation Insider

The first media publication dedicated to reputation management, covering industry news and professional practice.