Formal or Informal With AI? A Politeness Study of 10,000 Conversations

When OpenAI boss Sam Altman was asked on X in April 2025 how much it costs the company that people write “please” and “thank you” to artificial intelligence, he answered: tens of millions of dollars — and money well spent. But nobody has properly measured who is actually polite to AI. And in particular, there’s a question the English-speaking world can’t even ask itself: in Europe, where — unlike English — most languages distinguish informal from formal address, do we address AI formally?

We run an AI chat in 22 European languages, alongside an alternative called free ChatGPT with no sign-up. It’s data from this free version that we anonymously analyzed — 10,000 conversations from July — to measure how Europeans actually behave toward artificial intelligence. Whether they greet it, thank it, ask it politely, say goodbye — and whether they use informal or formal address.

A note on privacy up front: we worked exclusively with aggregated, anonymized data — no IP addresses, no publishing the content of a single conversation. Paying customers’ data never enters our studies. Details are in the methodology at the end of the article.

Key findings at a glance

Overall, at least one polite expression (a greeting, thanks, or a request) appears in 21% of all conversations — roughly one in five people writes at least one to the AI.

The politeness ranking: where “please” and “thank you” show up most

For each language, we measured the share of conversations with at least one polite expression. Rankings include only languages with a sufficient sample (50 or more conversations).

Czech leads by a wide margin — and the comparison with Slovak (15.9%) is telling: between two of Europe’s closest languages, there’s a nearly two-fold gap. At the bottom sit the Greeks and the Dutch, who deal with AI in the most businesslike way.

But politeness isn’t a single trait. Broken down into individual expressions, it turns out every nation is polite in its own way — and it’s the shape of these profiles that’s most interesting:

Czech is the only language evenly strong across all three expressions. German and English rest almost entirely on “please,” Spanish on greetings, and French greets and thanks but rarely asks politely. Detailed rankings for all twenty languages follow in the next sections.

Who greets the AI

Spanish speakers greet the most — “hola” opens every seventh conversation. But Spanish has an odd profile: it leads in greetings, yet ranks among the weakest in thanks (3.9%) and requests (3.8%). Say hello, sure — but then straight to business.

Who thanks the AI

Czechs thank the AI more than anyone in Europe — nearly every seventh Czech conversation includes “thank you” or “thanks.” Poles, by contrast, who greet above average, thank it the least of all (2.5%).

Who asks AI politely

Germans and English speakers lead in polite requests — with the caveat that German “bitte” is ambiguous, and the metric may slightly overstate it. Italians almost never ask politely (1.1%): in Italian, you command the AI.

And who says goodbye? Almost nobody

We found a farewell in just 0.8% of conversations — in eight of the twenty languages, not even once. People greet the AI, ask it politely, thank it… and then simply close the window. Part of the explanation lies in the interface itself: chat has no natural end-of-conversation signal — closing the window is the ending. Still, it’s striking that of all the rituals of politeness, the one that never made it into communication with AI is precisely the one with no practical function.

The longer we talk, the more polite we get

Politeness isn’t just a feature of language — it also changes with how a conversation unfolds. We split conversations into one-off commands (a single message) and longer exchanges (two or more messages): one-off conversations contain a polite expression in 17.4% of cases, longer ones in 24.9%. The pattern holds in 16 of 20 languages.

Part of the effect is purely mechanical — more messages mean more chances to say thanks. But even when we normalize politeness per message, it doesn’t decline in longer conversations — and the language ranking barely changes (see methodology). The real exception is interesting: English is the only major language where politeness doesn’t change with conversation length at all (23.3% vs. 24.0%). For Czech, by contrast, a longer conversation lifts politeness from 20% to 35% — as if people build a relationship with the AI over the course of the conversation.

Do we address AI informally, or formally?

This is a question English itself can’t ask — there’s only one “you.” But Czech, German, French, Spanish, and most other European languages force every user to choose: informal, or formal?

Important context first: most people don’t address the AI at all. Roughly three-quarters of non-English conversations are bare commands and prompts (“pancake recipe,” a pasted text to translate) — no informal or formal address appears in them. For most people, artificial intelligence is still more of a search engine than a conversation partner. The figures below therefore cover only conversations where the user actually addresses the AI directly — exact counts are in the methodology.

Europe addresses AI informally. In German, Slovak, Polish, and Swedish, we didn’t find a single case of formal address. Czechs use formal address in 5% of cases — so while they thank the AI more than anyone in Europe, it still doesn’t earn their formal “you.”

Lithuanians — the only ones who address AI formally

The exception is so pronounced that we gave it a dedicated audit: Lithuanians use formal address with AI in 43% of cases — four times the runner-up, Greek. A manual check of all Lithuanian conversations confirmed it isn’t a measurement error: Lithuanian users typically bring official and professional texts to the AI and keep a formal register throughout — polite verb forms like “pataisykite” or “parašykite” are aimed straight at the assistant.

Bonus: who gets frustrated with AI, and who’s just testing it

In the analysis, we also measured three further behaviors:

Frustration (“that’s wrong,” “you don’t understand me”) appears in 3.1% of conversations — and Slavic-language speakers get frustrated with AI the most.

The top four spots belong to Slavic and Baltic languages. The calmest are the Greeks (0.4%) — who are also the least polite. They simply don’t engage with AI emotionally, for better or worse.

Test queries (trivial questions like “what’s 2 + 2,” probing what the AI can do) make up 5.4% of conversations.

English does the most testing (8.2%) — casual global visitors mostly just poke at the AI. Czech (2.9%) and Lithuanian (2.2%), by contrast, actually put it to work.

Treating AI like a person — asking about its opinions, feelings, or identity (“what do you think?”, “are you human?”) — is, after a strict double review, rare: just 0.8% of conversations. Differences between languages are small enough to be noise, so we’re not publishing a ranking here. The audit did turn up a phenomenon we hadn’t expected, though: roleplay sessions. Some users have the AI play fictional characters and then talk to the character, not the assistant — for them, the AI chat isn’t a tool but a stage. We’ll measure how common this is in our next study.

How we measured this

• Sample: 10,000 anonymized records from the free ChatGPT alternative by GuideGlare (no-registration version), June 23 – July 24, 2026, 22 language versions; after removing 6 technical duplicates (the same conversation saved twice), we worked with 9,994 conversations. Rankings include languages with at least 50 conversations (Danish and Finnish were excluded). Sample sizes range from 53 conversations (Slovenian) to 2,775 (English); full counts by language are in the downloadable data. The informal/formal metric covers 1,873 conversations with a determinable form of address (of which 54 are Lithuanian); its ranking includes languages with at least 20 determined cases.
• Polite expressions are measured with dictionaries built separately for each language — including colloquial variants, internet abbreviations (thx, pls), and tolerance for typos in diacritics.
• Informal/formal address was classified by a language model, and every case found was then verified by a stronger model using language-specific rules (verb morphology, distinguishing address to the AI from address inside pasted text). Both striking results (Lithuanian formal address, a French anomaly in anthropomorphizing) also went through a targeted manual audit — one was confirmed, the other we discarded as an artifact.
• Statistics: all percentages are shares within a given language; for key values, we compute Wilson 95% confidence intervals (given in the text as ± percentage points), and we only publish rankings where differences exceed the statistical margin of error.
• Robustness: a control recalculation of metrics per individual message (instead of whole conversations) shifts rankings by at most 1–3 places and leaves the leaders unchanged — so longer conversations don’t mechanically skew the rankings. In the comparison by conversation length, we only include languages with at least 50 conversations in both groups.
• Privacy: we worked exclusively with aggregates — no IP addresses, no user identification, no publication of conversation content. Paying customers’ data never enters our studies.
• Limits: the sample consists of users of one free tool — the numbers describe behavior on our chat, not the population of an entire country. This is a one-month snapshot; repeating the study will show the trend.
• Data downloads and citation: download the complete aggregated data (22 languages × all metrics × sample sizes × confidence intervals) as a CSV. You’re free to reuse the figures and charts in articles, blogs, and presentations — just credit “Source: GuideGlare” with a live link to this page; details in our citation guidelines. We’re happy to send print-quality charts — just email info@explicaire.com.

Try it yourself

The data for this study comes from GuideGlare’s AI chat — it works in all 22 languages covered by this study, and you can try it as free ChatGPT with no registration. Whether you address it informally or formally is up to you. We’ll keep tracking the statistics, though — check back in six months to see if Europe has gotten more polite to AI.