Do Our Findings Apply Only to Healthcare? Repeating the Experiment in Other Industries

This article was first published in Korean by Intrix Lab on August 14, 2026.

In our previous article, we reported that 87% of the sites AI cited were outside Google's top 10 search results (measured on 2026-08-14). But that experiment covered dermatology only. Healthcare is a so-called YMYL field, where search engines and AI are known to give individual businesses less prominence while favoring public institutions and professional societies.

That means our findings could reflect a response to healthcare regulation rather than a characteristic of AI. If so, giving non-healthcare clients the same recommendations would be wrong. We therefore repeated the same experiment. We kept the number of questions, the number of question formats, the repetitions, and even the models unchanged, and changed only the industries (measured on 2026-08-15, 960 calls, 0 failures).

The results were split. One finding held up, and another did not.

How did we repeat the measurement?

The comparison industries were vacation rentals and wedding photography studios. We had two requirements. First, they had to be clearly outside YMYL. Second, they had to be industries where businesses actually have their own websites. Had we chosen hair salons or cafes, most of which use only map services, we could not have distinguished whether a result showing only platforms reflected the industry or simply the lack of websites to compare.

We used 48 local recommendation questions, such as "Recommend vacation rentals in Gapyeong." This matched the original experiment's question count and mix of question formats. We also chose locations suited to each industry: tourist destinations for vacation rentals and cities for wedding studios. An unsuitable combination such as "vacation rentals in Seoul" would make the question itself unnatural and change search behavior.

What stayed the same across industries?

The sites AI cites still had little overlap with Google's top search results.

Share outside Google's top 10: Healthcare / Vacation rentals and weddings. Model A: 87.1% / 85.7%. Model B: 62.5% / 71.4%.

The overlap indices were also in a similar range: 0.087 versus 0.095, and 0.294 versus 0.200. Changing the industries barely changed the results.

This replication mattered most to us. It meant our response to the argument that "AI search optimization is ultimately SEO" was not a chance result from a single industry.

Which finding did not hold up?

Citations of public institutions and professional societies. These accounted for 41.8% of Model A's citations in healthcare, but fell to 5.1% outside healthcare. For Model B, the share fell from 11.8% to 0.2%.

The difference is clear when the most-cited sources are placed side by side. In healthcare, Model A's two most-cited sources were dermatology-related professional societies. Outside healthcare, accommodation and wedding booking platforms took their place.

But the combined shares were the same

This was the clearest pattern in the experiment. When public institutions, professional societies, and platforms were grouped together, Model A's figures barely changed.

Healthcare: Public institutions and professional societies 41.8% + platforms 6.3% = 48.1%. Vacation rentals and weddings: Public institutions and professional societies 5.1% + platforms 42.2% = 47.3%.

The difference was 0.8 percentage points. The total stayed the same; only the composition changed. Model A consistently drew about half its citations from sources compiled by third parties. Professional societies filled that role in healthcare, while booking platforms filled it outside healthcare.

Our interpretation is that it looks for authoritative sources first, then turns to platforms if those sources are unavailable. Professional societies and platforms appear to serve as substitutes for one another.

For Model B, the platform share barely changed, from 28.6% to 29.6%. This model appears to use Google search directly. We suspect the reason is that Google places platforms near the top in both industry groups. Instead, the share of media sources such as blogs and videos rose sharply, from 6.5% to 35.5%.

What went against our expectations?

The share of Model B responses that skipped search entirely rose from 21% in healthcare to 37% outside healthcare. We had expected it to be more cautious in a less regulated field, but the result was the opposite. Model A never skipped search in either industry group.

Runs without search also had 0 sources. That means it recommended vacation rentals without any supporting sources one time in three. However, we could not establish why. We record this only as an inference.

Our recommendations now follow two paths

Three recommendations apply across industries: make your site reachable through search, do not use search rankings to estimate AI visibility, and check each model separately.

By contrast, the recommendation to "build connections with professional societies and public data" applies only to healthcare. Outside healthcare, that route is effectively unavailable.

For non-healthcare businesses, check instead whether your business is properly listed on booking and intermediary platforms. These account for 42% of Model A's citations. For vacation rentals, accurate listings on major booking platforms may have a greater effect on AI visibility than refining the business's own website.

Our free SEO and GEO audit measures whether your site is reachable through search. However, no action guarantees a citation in a particular response. What we can provide is a measurement of the prerequisites.

Limitations of this experiment

This covered two industries at one point in time. Vacation rentals and wedding studios do not represent all non-healthcare industries. The locations also differed by industry, so some location effects were mixed into the results.

We also did not report the share of individual businesses. When classifying sources by type, 35–53% remained unclassified, too large a share for us to report that figure in this round. Assigning unverified sources to a category would make that category's entire figure false.

Frequently asked questions (FAQ)

Q. What is YMYL?
It stands for Your Money or Your Life and refers to topics where incorrect information could cause people material harm, such as health, money, and safety. Search engines and AI are known to place more weight on source authority in these fields. Healthcare is a typical YMYL field.

Q. Why vacation rentals and weddings?
We needed non-YMYL industries where businesses actually have their own websites. If we chose industries where most businesses use only map services or social media, we could not distinguish whether platform-only citations reflected the industry or the lack of websites to compare.

Q. Which group does our industry belong to?
If incorrect information can lead to material harm, as in health, healthcare, finance, or law, it is closer to the first group. However, this experiment measured only healthcare, accommodation, and weddings. We have not yet measured which way intermediate fields, such as education or interior design, behave.

Q. Why do you not name the two models?
To avoid having the results read as a comparison of vendors. Recently, we ran the same experiment separately with older and newer models. We found that two of our earlier conclusions reflected differences between generations rather than vendors, and withdrew them.