Where Your Words Go When You Use AI
For seven years, a big part of my work as a postpartum doula was making sure people understood what they were saying yes to. A tired new parent agreeing to something at 3 a.m. deserves to understand it at least as well as someone signing a contract in daylight (probably more). So it's a little awkward to admit how many times I've clicked "Got it!" on a software update without reading a single word of what changed.
Lately, those updates have a theme. A little sparkle icon appears in the email app, a "summarize this call" button shows up in the video tool, and the design app starts offering to write your caption for you. None of it asked permission in a way that felt like a real question... it just arrived.
So if you've never opened Chat or Gemini or Claude in your life, AI is probably in your business anyway. The useful question isn't whether you should start using AI. It's where AI is already touching your clients' words, and whether you ever actually agreed to that.
The AI you choose, and the AI that comes with your tools
There are two ways AI ends up in a small business, and only one of them feels like a decision.
The first is the AI you choose on purpose: Chat, Claude, Gemini, or a writing assistant you signed up for. You opened that door yourself, so at least you know it's there (even if you haven't read the fine print, which, I get it).
The second is the AI your tools add for you: meeting transcripts, smart replies, thread summaries, "magic" design features, and the website builder that offers to write your About page. These arrive inside software you already pay for, usually in an update you didn't read, under names that sound like pure convenience.
Underneath almost all the tech we use today there is an AI layer that's been slowly growing features for years. Even a business that proudly "never uses AI" probably runs tools with AI features because of the spam filter sorting their inbox, the fraud check on every card payment, and the algorithm deciding who sees their posts are all automated systems making calls about our business. I'm not pointing that out to catch anyone. I'm pointing it out because "do you use AI?" turns out to be the wrong question, and a better one is "which AI, doing what, with whose information?"
What the feature promises, and what it needs
The announcement email always says some version of "New AI-powered features to save you time!" What it rarely mentions is that most of those features have to read your content to work. A meeting summary needs the whole conversation, smart replies need to scan your inbox, and an assistant that "understands your notes" has to go through every one of them.
What happens to that content next depends on settings we technically control but most of us never open. It might be processed and discarded, stored for a while, reviewed by people checking quality, passed to a separate AI company that powers the feature, or used to train future models. Every tool handles this differently, and the answer can change with the next policy update, which is exactly why I'm not naming names here.
For most businesses this is a privacy question. For client-facing work, and especially for birth workers, it's a confidentiality situation. The content moving through our tools is intake forms, birth stories, feeding struggles, mental health check-ins, and the things clients only told us because they trusted us. That kind of information deserves a real decision, not the default setting the company chose.
Where AI is probably already in your tech stack
If you want to look (and that's really the whole point), here's where I'd start. You're not deleting anything yet, just getting a clear picture.
Email: smart replies, writing suggestions, thread summaries, and events pulled out of messages all mean the feature is reading what's in your inbox.
Video calls: transcripts, AI meeting notes, and recap emails are made from the full recording of the conversation.
Scheduler or client portal: suggested responses and client "insights" are built from the client records you keep there.
Notes and docs: auto-tagging, smart search, and "chat with your notes" features work by indexing everything you've written.
Design tools: generated images, AI copy, and magic resizing often send what you type or upload to a model for processing.
Website builder: AI copy assistants, SEO suggestions, and chatbot widgets may read your site content and, with a chatbot, your visitors' questions too.
Specific company policies change constantly, so the goal isn't to memorize who does what. It's to build a reflex: when a new feature shows up, you check before you shrug.
Three questions for any AI feature
When I'm deciding whether a feature gets a place in my business, I come back to the same three questions. They work for any tool, not only the AI ones.
Where does this go? Does your content stay on your device, go to the company's servers and back, or get passed to a separate AI provider? Is it kept, and for how long?
Who decides? Was the feature switched on by default, or did you turn it on? Is there a setting to keep your content out of model training, and who gets to change those terms later?
Would I know? If the policy changed next month, would anyone tell you? And if a client asked how their information is handled, could you explain it in a sentence or two?
You won't always get a clean answer, and that's information too. A tool that makes these questions hard to answer is telling you something about how it sees your data.
Where I stand (gaps included)
I'm not writing this from outside AI. I use it every day, I teach small business owners to use it well, and I'm currently working on a Master of Science in AI at Purdue. I'm also the person who moved my whole business from Chat to Claude partway through a year because the company's values mattered to me, which tells you how seriously I take the question of who's holding my words.
In practice, that means I've turned off training data sharing, I adjust and touch everything before it goes out, and I've published an AI policy on my website. What's still under construction is saying it out loud to every client, every time, before I open a chat window on their behalf. On some projects I remember up front, and on others I catch myself halfway through and circle back.
Birth work taught me that consent isn't a single yes you collect once. It's something you keep building with the person in front of you, and that's the standard I'm holding myself to here. I'd rather tell you where I actually am than pretend I've arrived.
Your one move this week
Pick one tool that touches client information, which for most of us is email, client records, or video calls. Open its settings, search for "AI," "smart," "improve," and "data," and run whatever you find through the three questions above.
You don't have to change anything yet. The goal is to swap a default you never chose for a decision you actually made, even if that decision is to leave everything exactly as it is.
And if a client ever asks whether you use AI with their information, you'll have a real answer instead of a nervous one. If you've supported families through birth, you already know how to have an informed-consent conversation, and this uses the very same muscle.
Private is a promise, and control is what you keep
Nearly every tool in your stack promises privacy somewhere in its marketing. That promise is worth something, but it's still the company's promise, written in terms they can update whenever they like. Control is the part that stays with you: knowing what's switched on, choosing what goes where, and being able to explain it to a client without flinching.
The alternative to intentional AI use isn't no AI. The honest alternative is accidental AI.
The next post in this series (Day 5 of the email course) is about what to do when you look at your stack and don't love what you see. It's a spectrum rather than a purity test, and nobody has to move to a cabin in the woods.
If you'd like a structured way to do the looking, take the How Private Is Your Business? self-check, and join the free 5-day email course Your Private Business. And if you find something in your settings that surprises you, I'd genuinely love to hear what it was.
