Last updated: 9 June 2026 · Read the full framework
Plain English Wisp is an AI-scripted, AI-narrated podcast for kids. That's an honest claim, and it's also a responsibility. This page is how we meet it.
1. We never discuss certain topics, ever.
Military and weapons. Violence, attacks, terrorism. Deaths, killings, fatal accidents. Drugs, alcohol, gambling. Sex, dating, relationships. Abuse, hate, court cases. Political party fights. Adult body topics.
We also skip 20th-century world wars entirely — including 'good guys' like Schindler or Anne Frank, and named battles like D-Day or the Blitz. The framing's too heavy for the 5–11 age range and there are plenty of celebratable people in history without going near them. If your child needs to learn about Anne Frank, that's for school and you, not Wisp.
The list is enforced by three layers of automated checks before any audio is generated. If your child hears one of these, it's an editorial failure and we want to know.
2. We never collect data from your child.
The iOS app has zero analytics. No accounts. No tracking. Anything you set up in the app stays on the phone — it isn't sent to us. We genuinely couldn't profile your child if we wanted to, because the data doesn't exist on our side.
On this website only (where parents read, kids don't), we use Cloudflare Web Analytics to count page views, countries, and what people clicked from. It's cookieless, doesn't fingerprint visitors, doesn't follow you across sites, and never reaches the kids' app. Under UK PECR + ICO guidance, cookieless aggregate analytics don't require a consent banner. We disclose it here so you know what's running.
3. We never run adult ads at children.
No third-party ad networks. No behavioural targeting. No gambling, dating, financial, or alcohol ads. When (and if) we accept sponsorships, they will be audience-appropriate, clearly labelled, and chosen for the audience as a whole — not for an individual listener.
4. We keep a copy of every script we've ever produced.
If your child hears something you didn't expect, you can report it from inside the app (tap ⋯ → "Report this episode"). We can pull the exact script that aired, on any day, and decide with you what to do about it.
5. There is a real person responsible.
Ben Heyman, founder. hello@wispdaily.com. Editorial complaints, safety questions, regulatory inquiries — all come to me. Twenty-four hour response target during the working week.
The AI gets real headlines from mainstream news feeds (BBC, BBC Newsround, The Guardian, NPR, Smithsonian, NASA, New Scientist, National Geographic) and is told to report what's there — not invent. But generative AI can still occasionally introduce factual errors: a wrong name, a wrong number, a story that misses a key detail.
We don't currently run automatic fact-checking on every script. We rely on (a) credible sources as input, (b) prompt instructions to stay within those sources, (c) the audit trail that lets us produce the exact script and source set for any past episode, and (d) the Report flow in the app for parents and curators to flag what we missed.
Our v1.1 roadmap adds source-citation enforcement (the AI has to tag which headline backs each claim) and named-entity drift detection (flag scripts that mention people or numbers not in the source material). Until those ship, please report anything that looks off — we'll show you the source set and decide what to do together.
Wisp is AI-scripted and AI-narrated. The brief, the safety rules, and the ban list were all written by a human — me. Each individual episode is generated by AI within those rules, checked by the 3-layer safety filter, and uploaded automatically by an overnight job. I do not personally read or listen to every episode before it publishes.
What I do, every week, is review samples of the output and check the rules are still doing their job. I read every Report sent through the app. If something needs to come down, it comes down within minutes. The audit trail (every script archived) means I can show you exactly what aired on any given day.
I'm telling you this plainly because the alternative would be overclaiming, and a safety framework that overclaims is worse than one that's honest about its tradeoffs. Pre-publish review of every episode is a v1.1 candidate — but it competes with the "fresh before breakfast" promise that's central to what Wisp is. For now, the rules + the filter + sample review + the Report flow is how this works.
This is the technical detail. If you'd rather skip it, the short version is: nothing harmful reaches an audio file, because three independent checks have to pass first.
Every morning Wisp pulls news headlines from a narrow set of trusted feeds (mostly BBC, with a few specialist topic banks). Before any headline reaches the AI scriptwriter, it's checked against a list of banned terms — military, weapons, violence, drugs, sex, abuse, hate, and similar. If a headline contains any of these, it's dropped. The AI never sees it. That's the most important layer — the model can't talk about content it can't read.
The AI scriptwriter works to a long set of instructions: what to lean into (animals, festivals, kid-makers, sport, science discoveries), what to skip entirely, how to age-pitch the language (understandable to a sharp 7-year-old, interesting to an 11-year-old, max one unfamiliar concept per minute), how to end (on wonder, not warning). The full brief is published in our source code repository.
After the AI writes a script, the same banned-word check runs against the finished text. If anything slipped through, the script is automatically rewritten with an explicit note about what was wrong. If it still doesn't pass, the segment is refused — no audio gets generated, nothing is published. Better a missing episode than a bad one.
Wisp is operated from the UK and used worldwide. We anchor against the following regulations and apply the strictest standard across them as our baseline:
Australia (Online Safety Act 2021), Canada (PIPEDA + emerging AI rules), Singapore (PDPA), India (DPDPA 2023) — all addressed by the same no-data-collection architecture and the same content controls.
The detailed mapping of every requirement to a specific Wisp control is in our public Safety Framework.
It can. We use AI; AI surprises people. Here's what happens:
Data protection incidents (e.g. if our backend logs ever exposed something) are reported to the ICO within 72 hours under UK GDPR Article 33. So far we have had none.
None of the above is a marketing claim — it's all visible in the source code:
_HEADLINE_BAN_REGEX, _build_mixed_news_prompt, _scan_script_for_violations)If you're a journalist, curator, regulator, or just a thorough parent and want a specific script from a specific date, email hello@wispdaily.com with the date and we'll send the text.