British Callers Say Please to AI Phone Assistants: What the fonio Data Shows

Seven in ten British phone calls handled by fonio.ai voice assistants include “please” or “thank you,” according to TechRadar Pro reporting on November 7, 2026. More than half of callers greet the system first; about one in eight apologizes when the bot misunderstands; and a small fraction asks how the AI is doing. The story is less about Britain alone than about what happens when voice AI stops feeling like a gadget and starts feeling like a caller on the line — and what that costs in tokens.

TL;DR: numbers from fonio call analytics

fonio sells AI phone receptionists for businesses — the kind that answer after hours, screen calls, and book appointments. The November figures are aggregated call behavior on fonio’s platform, not a government statistics office poll. Treat them as vendor telemetry with a clear selection bias toward companies that bought voice AI, not as a census of every Brit with ChatGPT.

What fonio and TechRadar actually reported

TechRadar attributes the study to fonio.ai and quotes co-founder Daniel Keinrath: even when callers know they are speaking to AI, they “bring the same manners they’d use in a conversation with another person.” Keinrath frames the pattern as evidence that voice technology is embedding into everyday life — which matches how fonio positions its product (receptionists that sound natural on the phone).

The article also cites a February 2026 Future PLC survey of more than 1,000 people (parent company of TechRadar) finding about 70% politeness toward AI in general, with 12% polite specifically in case of a robot uprising. The earlier, longer TechRadar breakdown of that poll — “Are you polite to ChatGPT?” from February 2025 — says Future fielded it in December 2024 with 510 US and 518 UK respondents. UK users who were polite split 83% “because it’s nice” versus 17% uprising insurance. So the November voice numbers rhyme with the chat numbers but add voice-only rituals: apologies, goodbyes, and faux small talk.

TechRadar rounds out the piece with the politeness-and-power meme: OpenAI CEO Sam Altman has publicly joked that user “thank you” messages cost enormous electricity, a line TechRadar already covered in its February AI etiquette article. We treat that figure as Altman’s colorful cost framing relayed by TechRadar, not as an audited utility bill.

Why voice politeness hits different than chat politeness

Two chat bubbles in conversation, illustrating British politeness habits on AI voice calls

Text chat users can be blunt because latency is low and edits are free. Phone callers inherit telephony etiquette: you say who you are, you thank the person who helped, you hang up cleanly. Our AI interviewer loop story showed what happens when two agents trained for politeness cannot end a call — comic on paper, expensive in production minutes.

For builders shipping voice agents, fonio’s breakdown is a product map:

1. Greeting detection — more than half of users open formally; skipping straight to intent may feel rude.
2. Apology loops — 12% apologize when the ASR or model fails; your recovery script should accept “sorry” without derailing.
3. Closing rituals — 13% say goodbye; cutting the line instantly can tank satisfaction scores.
4. Anthropomorphic slips — 2% ask “how are you”; harmless, but don’t pretend the bot has wellbeing unless your brand wants that fiction.

None of that requires believing the model experiences gratitude. It is UX mirroring, the same reason IVR menus sound human even when everyone knows it is a recording.

Token cost versus user satisfaction

Soft voice waveform with a green peak, for extra tokens spent on please-and-thank-you on AI phone lines

HN’s thread on the TechRadar headline (item 50410796) stayed small but hit the trade-off directly. User ggm noted that politeness costs tokens but still aids user satisfaction and contentment (comment 50410929). SillyUsername, identifying as British, argued the habit is not UK-exclusive and recalled deliberately triggering American “you’re welcome” loops as tourists (50411186). eucryphia snarked about unrelated culture-war headlines (50411191) — skip that as signal.

For engineers, ggm’s point is the actionable one: courtesy is a billing line item. Every “thank you” is another few tokens on a speech-to-text plus LLM plus TTS chain. At fonio scale those add up; at ChatGPT scale Altman’s joke lands because marginal tokens × hundreds of millions of users is real money. That does not automatically mean you should train users to be rude. It means you should measure whether politeness correlates with task success or NPS on your product.

Our guide to working with AI agents argues the same for coding agents: specific beats polite, but hostility does not buy reliability. The dedicated insults-and-tips evidence post is mixed — tone nudges some benchmarks, not all. Voice is a third lane: users may not optimize tokens, and brand risk from sounding cold can exceed the compute saved.

Policy and culture threads builders should watch

Politeness is not only UX. Labs increasingly write user behavior into policy. Anthropic’s 2026 usage policy tightened rules on abusive behavior toward Claude — the mirror image of mandatory courtesy. Whether “be nice to the model” becomes a Terms-of-Service theme is still open; fonio’s data suggests many users self-police without a ban.

The 12% uprising insurance figure from Future’s poll is meme-ready, but it also shows trust anxiety in consumer AI. That connects to broader verification-tax surveys and to safety politics like Sanders’ superintelligence ban framing — different axes, same user worry that systems more capable than us might remember slights.

BBC Future’s chatbot etiquette explainer situates the manners debate in decades of people saying “please” to Alexa. Pew Research noted majority “please” use on smart speakers years earlier. The fonio slice updates that story for UK SMB phone lines in 2026.

What this means if you build or buy voice AI

If you integrate fonio or similar: expect transcripts littered with social glue. Train analytics to strip or tag courtesy tokens before intent classification, but keep them in CRM notes if humans call customers back — hearing “they thanked the bot” signals tone.

If you run your own voice stack: budget 10–20% extra tokens on consumer lines versus stripped-down internal tools. Compare against open realtime TTS options only if you publish a public assistant; fonio customers mostly buy outcomes (answered calls), not raw tokens.

If you deploy coding agents only: this story is cultural context, not a tuning knob. Still, teams debating whether to forbid “please” in prompts should read it as a reminder that non-engineer users import human rituals into every new interface.

If you research alignment: user politeness does not prove models deserve moral standing. It proves interfaces inherit social scripts. Our Claude “soul” pretraining feedback essay is the cautionary flip side — whatever users and journalists say about AI ends up in training data eventually.

Limits of the data (read before you cite the 70%)

1. Vendor telemetry, not a national sample — fonio callers are businesses that paid for AI answering; demographics skew toward UK trade and services using fonio’s numbers.
2. No public fonio whitepaper at crawl time — we could not find a standalone methodology PDF on fonio.ai; TechRadar is the readable secondary source with named executive quotes.
3. Future poll date mismatch — the November article says “February 2026”; the underlying Future poll TechRadar documented was December 2024 with a February 2025 write-up. Use the older article for methodology.
4. Altman cost quote — treat as executive color, not audited energy accounting.
5. HN traction was low — four points at crawl; developer sentiment is illustrative, not representative.

Product teams localizing for UK English should A/B test whether the bot saying “thank you” at hang-up reduces abandoned calls. fonio’s data implies many humans already treat the line as reciprocal; matching their script may recover more in completion rate than it spends in extra synthesis seconds.

Practical checklist for voice-AI product teams

Related on explainx.ai

• How to work with AI agents: communication guide
• Does insulting your AI agent change output quality?
• AI interviewer meets AI interviewee: the politeness loop
• Anthropic usage policy on abusive behavior toward Claude
• AI verification tax at work
• Bernie Sanders Ban Artificial Superintelligence Act, explained
• Kevin Boone on AI web content and trust lists

Sources: TechRadar Pro, November 7, 2026 · TechRadar, Future politeness survey, February 2025 · fonio.ai product site · BBC Future on chatbot etiquette