The vendor: Voibe. They make a privacy-oriented voice-to-text tool. Full disclosure: I do NOT use their product and never have.

But now I know the brand and am happy to give them free marketing. All because of this: the AI Privacy Tracker I needed and can't seem to find elsewhere.

What I like:

• A clear up-front methodology. No guessing what they rank as "good".

• Broad range of tools tracked. They list 34 currently, ranging from ChatGPT and Claude to Zoom, Granola, and Otter.

• Very frequent updates. Needed to match the monthly cadence of policy shifts across major AI tools.

• Industry-specific context. It guides users to where privacy matters more - think healthcare and legal - vs. what you might consider low-stakes use.

• It's highly readable, clearly sourced, and promises rapid responsiveness to feedback. Unusual. Are there other vendor rankings that openly welcome corrections and promise fast turnaround?

What I absolutely love:

• They rank themselves. And not automatically at #1. I'm sure plenty of marketing experts might squirm, but that's how they earned my attention and initial trust.

What surprised me: no neutral third-party tools that delivered this or better. What I found were either annual ("the 2025 AI Transparency Rankings") or episodic ("our latest review of this niche category") but nothing as routine or well-sourced or widely applicable. But why? The basic structure seems straightforward enough for someone with Claude Code tokens and a human-in-the-loop for review.

I know the real answer combines "there's not enough money in protecting privacy" and "doing this right would take more time than we have". But a girl can dream. And until there's a neutral version, I'll happily take advantage of Voibe's investment for my own benefit.

"Most of the sentences you make will need to be killed. The rest will need to be fixed. This will be true for a long time."

An unforeseen side effect of all this time I'm spending iterating with AI tools: I desperately want to become a better writer.

I always loved to write but somehow, over the years, never made as much time for it as I wanted. Journal entries lay half-drafted. Story ideas accumulated but didn't magically start composing themselves. And so so many LinkedIn posts never saw the light of day.

But now I'm having a counter-reaction to the AI slop that surrounds us every day. And which I'm sometimes tempted to share out myself, despite the AI-obviousness of it all. After all, it's so easy. The full draft is right there in front of me. And maybe no one will notice? But I notice.

One starting place: an absolutely fantastic book by Verlyn Klinkenborg that's been sitting on my shelf for years. "Several short sentences about writing." 149 pages of perfectly crafted and concise sentences structured into tidy little stanzas, like a self-help poetry book for the wanna-be better writer in us all.

This lovely little book won't protect me from every temptation to lean on AI drafting help when I'm tired or in a hurry. But it will help. And at least this post was 100% human-drafted. It's a start.

It started as a jaded take on AI and privacy. Or lack thereof. With stats like:

• ~25% "never redact" sensitive personal info, according to one study of UK residents last year.
• ~40% uploaded personal health information such as test results or doctor's notes, of those who used AI for health guidance, according to a US study this year by health policy organization KFF. This despite more than 3 in 4 saying they're concerned about the privacy of their health information.

Then there's the frustration I'd been saving up: the privacy 'gotchas' buried in every Terms of Service. Karelia Whitt named it in a comment on my last post: how powerless these Terms leave us feeling. The gotchas are the clauses they know we'd hate — if we read them. Most of us don't. And the few who do rarely feel free to walk away, because what's the better alternative anyway?

But I wanted to believe some providers were better than others. Especially for AI tools where the risks are so much higher (yes I'm talking to you, people who upload your medical records; or, fellow finance folks, your full financial spreadsheets…!)

Which left me wondering: who really does data privacy better?

My takeaways when looking only at US-based providers and how they handle your data by default:

• Yes there are differences but they're minor. For example Meta AI doesn't have a training opt-out in the US, but all models train and none keep you anonymous.
• Differences are diminishing. For example, Claude completely flipped their data training policy in late 2025 from opt-in to opt-out.
• New "temp chat" options from ChatGPT appear to offer more privacy, similar to incognito browsing mode, but only after a court fight. Before it was over, OpenAI handed over 20 million ChatGPT logs - including deleted conversations - to a judge. Phewwww.

But then I found duck.ai from DuckDuckGo and fell in love. Data privacy love. Because it sits on top of other models but takes the opposite stance to every gotcha the AI model providers put in their terms. No training. No account creation. No data storage.

And here's the gosh-darn-it part: I've spent months building with these models, heads-down at the model layer, comparing them provider by provider and choosing the lesser of the evils. And I nearly missed that the privacy I sought was sitting one layer above.

It's still early days for my experience with duck.ai, and I'm not about to push truly sensitive info into a third-party product no matter what. But I'm grateful for the reminder that we still have real choices here — not just the binary "let an LLM mine your data for years" or "skip the AI tools that overnight became table stakes for modern life."

Well, at least until the Terms change again.

Not really sure why. Even people deep into building with AI, as I am, often react defensively when I ask questions, like they're afraid I'm going to take away their favorite LLM toy.

So I was grateful to have a conversation over a lovely Mexican dinner last Sunday where someone who truly wanted to understand asked me the follow-up questions instead.

"What do you really mean by privacy?"
"What are you most worried about?"
"Do these AI companies really handle privacy differently or is it all just marketing?"

I loved these questions so much I tried to answer them even while happily munching on fantastic duck confit flautas.

I couldn't do all three justice between bites, so let me start with the first — what "privacy" even means. For me, data privacy wraps together a few issues, especially in a world of AI chatbots that you can pour your deepest secrets and household budgets into. At a minimum, privacy can include questions about:

• Training: are they using my input to improve the model? (Because doing so can expose your sensitive information.)
• Retention: how long are they keeping my prompts and information around?
• Sharing: who else can they pass my data to? (If their policy says "third party partners" then my info can be shared to a LOT of parties.)
• Control: can I make my settings more private? Assuming I can find the right options deep in the settings menu?
• Accountability: who keeps them honest if they break the rules?

Now that I've written it out, I get why this isn't easy dinner conversation — there's a lot to unpack. What I realized somewhere between the flautas and the check: once you split "privacy" into these five questions, it gets easier to talk about, not harder. As a former management consultant I should have known to find a framework and whip out some bullet points. For now, it's given me more to think about, and to keep writing about in the coming weeks.