AI answers by language: the market in the query decides

In global categories, AI search visibility depends on whether the question names a market, not on the language it's asked in. That's what 15 Perplexity queries I ran in September 2026 showed. Most AI visibility reports boil everything down to one score. I don't, because three separate things happen in an AI answer and each takes different work: a brand gets named, its page gets cited as a source, and someone visits the website afterwards. I come from SEO and performance marketing, and there a number with no source behind it carries no decision. For any company selling into a non-English market there's a fourth question: which language market do the sources come from when an AI search answers a question in the local language?

Named, cited and visited are three different wins.

A brand can show up in a generated answer without anyone visiting its site. It can be linked as a source without being recommended. And an answer can recommend a vendor whose own page never appears as a source. Each of these has its own value for marketing.

A report that rolls everything into one score hides exactly these differences. When the score goes up, nobody knows whether the brand is being recommended more or just showing up more often as a footnote. Track mentions, source links and visits in separate columns.

A German question doesn't guarantee German sources.

An AI search builds its answer from pages it finds on the web. Which pages those are decides who can show up at all. For a German question it seems natural to assume German sources get used. That's unproven. For software and other products sold internationally, most comparison content is written in English, and the German pieces are often derived from it.

Whether local-language content is worth the money depends on this. Check which sources get cited in your category before you put budget into content for AI answers.

Without a log, every measurement is an anecdote.

Answers vary from one run to the next. A screenshot from the CEO's browser isn't a measurement. It only becomes reliable with a fixed query set, run the same way over several days and logged line by line:

An example: asked for suitable vendors, an AI search names a company and links to a trade article. On the next run the source link is gone, but the company is still named. Only a separate log shows that change. If you pass on numbers like these, check them yourself first. Measure in series over several days, never from a single run.

More direct visits don't prove an AI effect.

When direct visits to a website go up, that doesn't prove a particular AI platform is behind it. Not every visit from an AI answer can be identified as one. A report that credits AI for the jump anyway is selling a guess as a result. Report missing referral data as a gap.

The September 2026 measurement

Between September 9 and 11, 2026, I ran 15 queries through the Perplexity API, without a location signal. They covered six categories, among them CRM software, ERP for mechanical engineering and supplemental dental insurance, in up to three languages: German, English and French. Some questions came in versions with and without a market reference. I analyzed the source lists at domain level and, separately, the vendors named. The full article with all tables was published in German by AFS-Magazin on October 15, 2026: KI-Sichtbarkeit messen: Warum deine deutsche Frage englische Quellen bekommt.

The clearest result: asked for the best CRM for mid-sized companies, the German and the English version returned the same source list, 19 of 19 domains identical, none of them German. For supplemental dental insurance, the German and English versions shared not a single source, and the German answer drew 88 percent of its sources from German sites.

The categories split into two groups. In CRM and project management software, English comparison portals, trade magazines and forums dominated. In ERP for mechanical engineering, software for trade businesses and supplemental dental insurance, the AI search found its own body of German sources. By my analysis, what makes the difference is whether a category has a dense body of sources of its own in the German-speaking market.

Naming the market in the question brings in far more German sources. Without a market reference, the same CRM question returned no German source. With "in Germany" added, 6 of 17 (35 percent) were German, and with "GDPR-compliant, from Germany", 13 of 19 (68 percent). The English question asking for a CRM for companies in Germany returned 7 of 17 German sources (41 percent), more than the German question without a market reference. The market in the question pulls in German sources, while the language alone doesn't.

Two more findings concern the measurement itself. In CRM and project management, the German and English versions cited identical sources, but the vendors they named only partly overlapped. That confirms the case for keeping mentions and sources apart. And the French answers shared only 3 percent of their sources with the German ones for CRM and 6 percent for project management. You can't infer one language market from another.

The way into the answer runs through a niche.

In global categories, large international brands win regardless of the source pool. A mid-sized company gains little by going after the top-level category, whether it writes in German or in English. Its way into the answer runs through a niche where it's strong: a market, an industry, or a requirement like GDPR compliance. Once the question said "GDPR-compliant, from Germany", German vendors that never showed up in the general version suddenly appeared among the sources.

If your category sits in the English source pool, well-kept profiles on the comparison portals cited there, like Capterra and TrustRadius, count for more than another German blog post. If it sits in the German pool, it's the other way around. Queries with and without a market reference belong in separate reports. Measure only questions that name the market and you see a rosier slice than reality. Report both numbers side by side. The gap between them is the metric worth watching.

A 20-minute check for your category.

This comes from the AFS article and needs nothing more than an AI system with live web search.

1. Write four questions the way a customer would ask them, two without a market reference and two with one, for example "Which software for X do you recommend?" and "Which software for X works for companies in Germany?".
2. Ask each one and note the cited sources, not the brands named. The brands are the result. The sources are where you can do something about it.
3. Count the share of domains with content in the local language. Below 20 percent you're working in an English-dominated source pool. Above 50 percent, local source work is the right move.
4. Repeat after two weeks. If the domain list changes a lot, the category is contested and needs more runs before you act on it.

Fifteen queries are a sample.

This isn't a study. I tested one system, Perplexity. ChatGPT, Gemini and Google's AI Overviews draw on other search indexes and may behave differently. It's still open whether systems with a location signal reach for German sources sooner. I didn't measure Austria or Switzerland. The results support a working hypothesis, not a general rule. The picture only becomes reliable once the same queries are run again.

I measure AI visibility through the sources, because that's where the work starts. Run the four questions from the check for your own category before you sign off on the next content budget. How that turns into ongoing measurement is on the page Measuring AI visibility: GEO in practice.

Further reading

• The guest article in AFS-Magazin (German)
• Publications and press