Do English Pages Prompt AI to Investigate a Business? We Checked 111 Websites

This article was first published in Korean by Intrix Lab on August 17, 2026.

There was no difference. After finding that 14.7% of the search queries AI rewrote for itself were entirely in English, we asked whether AI investigated businesses with English pages more closely. The answer was no. Of the 54 businesses AI specifically investigated using the site: operator, 10 (19%) had an English version. Among the 57 businesses it did not investigate, 11 (19%) had one (measured on 2026-08-17, probing 111 business websites). This article records how we removed a plausible recommendation from consideration.

Why did this question arise?

The starting point was a byproduct of another experiment. When you ask AI a question, it does not enter that sentence unchanged into a search box. It rewrites the search queries in its own way. We examined 1,289 search queries generated by AI in our research.

The language mix stood out. Queries mixing Korean and English accounted for 80.1%, and Korean-only queries for 5.2%, while English-only queries accounted for 14.7%. Even though the original questions were in Korean and sought Korean businesses, one in seven searches used English only.

A recommendation naturally suggests itself: "Wouldn't creating English pages help a business appear for those English queries?" It is something we often hear in practice, and we also found it plausible. But plausible and true are different, so we decided to measure it.

What websites do those English queries actually return?

We first examined the queries. We checked what websites 35 English-only search queries actually returned.

They produced 309 results, of which 27 were on .kr domains, a share of 8.7%. Counting unique domains, that was 7. The rest were overseas sites. The web results returned by English queries were mostly outside Korea.

At this stage, the premise behind the recommendation was already weakening. Creating an English page means trying to enter those search results, but those results were not a space where Korean businesses competed with one another.

Did businesses with and without English pages differ?

The premise alone was not enough, so we also measured the outcome. Our outcome variable in this experiment was whether AI investigated the business more closely.

This needs some explanation. AI works in two stages. It scans candidates through general searches, then searches again within a specific domain when a business interests it. For a person, the equivalent would be hearing a company's name and then opening its website directly.

We divided the business domains AI had read into two groups: 54 that received this targeted investigation, and the top 60 that did not. We then checked each website for an English version. The 111 that responded formed the final sample.

Here are the figures.

	Businesses AI investigated more closely	Businesses AI did not investigate more closely
Had an English version	10/54 (19%)	11/57 (19%)
Median English share of homepage content	6%	6%

The ownership rates were identical. The median share of English in homepage body text was also identical at 6% in both groups. Other factors must distinguish the groups, but having English pages was not one of them.

How far can we trust this result?

This is the most important section of the article. What we can say is limited to this: "In this sample, having English pages did not distinguish whether AI investigated a business." That is not the same as saying "English pages are useless." Those are entirely different statements.

Here are the limitations. First, the sample covered 111 businesses, measured once at a single point in time on 2026-08-17. Without repeated measurements, chance factors specific to that day may have affected the results.

Second, these domains came from 40 queries about the GEO and AI search optimization industry. That is one industry. Industries with a high share of exports or actual overseas customers could easily produce a different picture.

Third, we checked for English versions only on the businesses' websites. We may have missed businesses operating a separate English-language domain. We expect such misclassification to lower the ownership rate in both groups, but we have no basis for ruling out the possibility that it was concentrated in one group.

Fourth, this was an observational study, not a randomized experiment. Businesses with English versions may also differ in size or years in business, and those differences are mixed into the results.

Neither this experiment nor any other action guarantees a citation in a particular response or a search ranking. What we can provide is the observed data within the scope we measured.

What remains?

We removed one recommendation from consideration. That is a removal rather than a new discovery, but we count it as progress. When budgets and time are limited, presenting "things you could do without supporting evidence" alongside "things supported by evidence" lets the former consume resources needed for the latter.

The remaining question is unchanged: what determines which businesses AI investigates? We have not yet established the answer. In subsequent experiments, we are testing other candidates one by one in the same way. When we have an answer, we will publish it here. The same applies if the result is not statistically significant.

If you want to start by checking whether AI crawlers can actually reach your site, our free SEO and GEO audit measures accessibility factors. That comes first.

Frequently asked questions (FAQ)

Q. What is a rewritten search query?
AI does not search using the user's sentence unchanged. It creates search queries in its own way. Those queries, rather than the user's question, become the actual searches.

Q. Does this mean we do not need English pages?
No. If English speakers are actual customers or your business receives overseas inquiries, English pages are needed in their own right. What this measurement does not support is the specific rationale of creating English pages as a way to be discovered by AI. A different purpose calls for a different judgment.

Q. Would our industry show the same result?
We do not know until we check. This sample covered one industry, 111 businesses, and one day. AI may select businesses to investigate differently across industries, so each needs separate measurement.