Power of the Voice

My wife and I have spent a lot of time lately sitting across a table from home builders. Some of those meetings have run three hours.

That's a lot of talking. Allowances, timelines, floor plans, what's included, what isn't. By hour three, hour one is a blur.

That's where this issue starts. No new model, no big announcement. Just something I think gets far too little attention outside the AI bubble I live in: talking to your tools instead of typing to them.

Quick note before going further. Nobody's paying me for this. No affiliate links, no sponsorship. These are two tools I've been using myself, and I'm telling you what I've found.

Back in 2016, researchers from Stanford, the University of Washington, and Baidu ran a simple test. They had 32 people enter everyday phrases on an iPhone, once by typing and once by speaking.

Speaking was three times faster. And in English, the error rate was 20.4% lower than typing.

To be fair, one of the researchers pointed out that typing still makes more sense in plenty of situations. In a meeting. In bed while your spouse is asleep. Voice won't fit every moment. It fits more of them than you'd think.

The first tool is Wispr Flow. You talk, and it turns what you said into clean text wherever you're working. I use it for prompts. I also use it for text messages, which I didn't expect.

Speed is the obvious win. The less obvious one is what I actually say.

When I type a prompt, I keep it short. Typing is work, so I cut corners. When I talk, I explain things the way I would to a person. The background. What I've already tried. What I'm worried about. What “done” looks like.

That's context. And context is the single biggest factor in whether AI gives you something useful or something generic. We covered that back in Issue 2. Voice gets you there without trying.

It also changes how fast I iterate. If the first answer misses, I don't sigh and start retyping. I just say what's wrong and keep going. It feels like a conversation, because it is one.

The second tool is Otter.ai. It records a meeting and produces a written transcript.

For our builder meetings, here's what that looks like:

1.      Ask first. Everyone at the table knows we're recording before we start.

2.      Let Otter run. Two or three hours, start to finish.

3.      Hand the transcript to Claude with a prompt along these lines: “This is a transcript of a meeting my wife and I had with a home builder. Summarize what was discussed, list every decision we made, every open question, and the next steps for each of us. Flag anywhere we and the builder seemed to be on different pages

4.      Check what matters. Prices, dates, and specs get verified against the builder's written documents. Always.

When I told our builder and his partner I'd be using AI to transcribe and summarize our meetings, I expected a raised eyebrow. Instead, his partner said he'd been looking at a Plaud device to start recording their client meetings too. He figured transcripts would add value on their side of the table as well.

That stuck with me. Here was someone well outside the AI bubble, running a building business in the Lowcountry, landing on the same idea for the same reason: fewer things lost between what gets said and what gets built.

What comes back from Claude is more than a summary. It gives us a starting point for the next conversation. My wife and I can sit down with the same list in front of us instead of two different memories of the same meeting.

That's the part I didn't see coming. I thought I was getting a record of the meeting. What I got was a better next meeting.

Three things I'd want someone to tell me before I started.

Get consent. Recording laws vary by state, and some require everyone's agreement. Even where they don't, asking first is just good manners. Nobody likes finding out later they were on tape.

Watch what you upload. Same rule as Issue 7: strip out account numbers, ID numbers, and anything you wouldn't hand a stranger.

Check the transcript. It will mishear names, drop numbers, and occasionally assign a sentence to the wrong speaker. Treat the summary as a draft, not a contract.

You don't need Wispr Flow or Otter.ai to test this. Your phone already has a microphone button on the keyboard.

1.      Next time you're about to type a question into ChatGPT, Claude, or whatever you use, tap the microphone instead.

2.      Talk for a full minute. Give it the background, what you want, and what you've already tried.

3.      Don't clean up the “ums.” The AI will sort it out.

4.      Compare the answer to what you'd normally get from a one-line typed question.

I'd bet the difference shows up on the first try.

I'm curious whether this is just me. Have you tried talking to AI instead of typing, and did it help?

Hit reply and tell me. I read every one.

Next week: scheduled tasks. Getting AI to do the work while you're doing something else.  

— Eric

Connect with me on LinkedIn: https://www.linkedin.com/in/eric-skelton-ai-over-50-newsletter/