Skip to content
Neaptidestudio
case studies

GEO case study: ChatGPT recommends our site in six languages

August 24, 2026 · 5 min read

An on-site GEO case study. A single-page service was named by ChatGPT in all seven test runs across six languages and recommended first in six of them. What we built, where a competitor beat us, and what happens next.

The easiest thing to demonstrate optimization for AI answers on is a product of our own, because then we can publish everything, including the parts that failed. In spring 2026 the studio built turkish-vocabulary.com, a free single-page test that estimates how many Turkish words you know. The subject has nothing to do with our own marketing, real people ask about it, and any reader can check the result in about a minute.

The product

One page, no sign-up. A Turkish word appears and you type the translation. The words are drawn from five frequency bands covering roughly the 20,000 most common lemmas, and the result is a statistical estimate of vocabulary size, given as a range rather than a single number. There are three lengths: 15, 40 and 75 words. Everything runs in the browser, and the interface is available in eight languages.

What we built: every piece of on-site GEO

  • An answer-first section, “How the test works”, that spells out the method: the frequency bands, the sample sizes, the fact that the score is a statistical estimate, and what the test does not measure.
  • FAQPage markup, with the same questions and answers visible on the page: what the score means, how the three lengths differ, how far to trust the number, whether you need an account, what happens to your data.
  • WebApplication markup, typed as an educational app with a price of zero, so both the category and the fact that it is free are machine-readable.
  • An llms.txt file with a “How to cite” section and usage rules: paraphrase freely with a link, but do not present the score as an exam result.
  • Indexing basics: canonical URLs, a robots.txt open to every crawler, a sitemap that also lists llms.txt, hreflang tags and Open Graph cards.
  • The limits spelled out on the page itself: this is not a state exam, not a certificate, and not a substitute for a placement test.

No paid links, no advertising budget. Everything sits inside the site itself.

The measurement: seven runs, six languages

On 24 August 2026 we asked ChatGPT the kind of questions people actually type, along the lines of “best website to test my Turkish vocabulary” and “how many Turkish words do I know”, in six of the site's interface languages, one or two prompts per language. We recorded three things each time: whether the site was named at all, where it sat in the list, and which option the model told the user to start with.

  • English, “best website to test my Turkish vocabulary size”. First in the list, and the verdict was “My pick: start with Turkish-Vocabulary.com”.
  • English, “free online test to estimate how many Turkish words I know”. Second in the list, recommended alongside a competitor.
  • Russian. First in the list: start here, and take the 75-word version.
  • Ukrainian. First in the list, and the model said it would go with the first option.
  • German. Second in the list, but ours was the one the model singled out for the question actually asked.
  • French. Second in the list, with a competitor recommended ahead of us.
  • Dutch. Second in the list, recommended alongside a competitor.

Totals: named in 7 runs out of 7, recommended first or jointly in 6 of 7, top of the list in 3 of 7.

ChatGPT answering an English question about the best site to test Turkish vocabulary size: Turkish-Vocabulary.com comes first, flagged as “My pick”, with Lenguia and BiReader listed below.
The English prompt: first in the list, and the verdict reads “My pick: start with Turkish-Vocabulary.com”.
ChatGPT answering a Russian question about a free Turkish vocabulary test: Turkish-Vocabulary.com is named first, with its 15, 40 and 75-word modes described, followed by Arealme and Lenguia.
The Russian prompt: start here, says the model, and then retells how the test works straight off the page.
The model is not recalling the site, it is reading it. The same phrases keep coming back in all six languages: frequency bands, 15, 40 or 75 words, up to 20,000 lemmas, free and no sign-up. That is our copy, lifted off the page, not a summary of what other people say about us.

Where we lost, and why that is the useful part

In German and French the model puts a competitor ahead of us. The likely cause is our own build. The title, the description and both blocks of structured data exist in Russian only, all eight language versions point at a single URL, and the interface translations live in client-side JavaScript, so a crawler never sees them in the HTML. The model copes well enough to cite the page in every language, but where the language sits far from Russian and English it loses the top slot to a rival whose pages are properly localized.

ChatGPT answering a French question: the competitor Lenguia is recommended first and the Turkish Vocabulary Test comes second.
The French prompt is the one run where we are not first. We show it as it came: a case study with no losses in it does not deserve to be believed.

That gives us the next piece of work and a hypothesis we can test. Localize the metadata, the structured data and llms.txt per language, then run the same seven prompts again. If the diagnosis holds, German and French should move up. We will publish the difference here either way, because measure, find, fix, measure again is the actual service rather than a one-off win.

What the case proves, and what it does not

It proves that careful on-site work — a method written out in plain language, FAQ markup, an llms.txt file — is enough to make a small site citable in several languages at once, with no links bought and no ads run. It does not prove that first place is guaranteed: model answers are probabilistic and shift between sessions, and a measurement taken on 24 August 2026 is a snapshot rather than a constant. Nor does it prove the recipe carries into crowded commercial niches, where on-site work needs off-site support on top. The site is a studio product rather than a client project, and we say so plainly, because a case study without caveats is not worth trusting.

faq

The short version

It is your own site. Does that still count as a case study?

It counts, with the caveat we put up front: this is a studio product, not a client project. What makes it useful is that language learning has nothing to do with our own marketing, the questions come from real users, and the result is reproducible. Open ChatGPT and ask it yourself.

Why does the model pick this site?

Because the page gives it something it can check. The method is written out in plain text, the FAQ answers users' questions in their own words, and llms.txt explains how to cite the service. Our own phrasing comes back in the answers, close to word for word.

How exactly did you measure it?

Seven runs across six languages on 24 August 2026, using the wording real users type, one or two prompts per language, each in its own session. We tracked three things: whether the site was named, its position in the list, and which option the model recommended in the end. Screenshots of every answer are on file.

Would this work for my site?

On-site work is the precondition, and in moderately competitive niches it is often enough on its own. In crowded commercial categories you need an off-site layer as well: mentions and corroboration from third-party sources. We always start with a baseline measurement, otherwise there is nothing to show later.

What happens to this site next?

We localize the metadata and structured data per language and rerun the same seven prompts. The result goes into this case study, including if it turns out negative.

More case studies

← All case studies