Hi AInauts,

Welcome to the latest issue of your favorite newsletter.

A quick note up front: Apple is suing OpenAI because the company allegedly hired away top hardware talent who may have brought company secrets with them.

At the same time, OpenAI's first own device has leaked. It looks like a portable screenless speaker, built as an AI companion that is simply always around. Whether the world was waiting for yet another smart speaker is another question. We are still betting on glasses.

Either way, the direction is clear: AI wants out of the chat window and into your everyday life.

Today we have fewer news items, but more topics that kept us thinking this week.

Here are the details:

Let's go.

You don't send a colleague a three-word brief. You explain the context, the constraints, what you've already tried. But typing all that into ChatGPT takes forever — so you don't.

Wispr Flow lets you speak your prompts instead. Talk through your thinking naturally and get clean, paste-ready text. No filler words. No cleanup. Just detailed prompts that actually get you useful answers on the first try.

Millions of users worldwide. Works system-wide on Mac, Windows, and iPhone.

Try Wispr Flow free

Small confession first: before important meetings, we often ask AI to build us a briefing about the people we are about to meet.

We use our second brain for people we already know, plus web search. Two minutes later, we know who we are talking to, what is on their mind, and what we have discussed before.

It works almost uncomfortably well.

The same question is worth asking the other way around: what can AI actually find out about us?

Exactly that experiment is going viral on TikTok.

A creator asked ChatGPT to find every photo of him on the internet. His takeaway: "It literally found me in photos I didn't even know existed."

More than two million views, and the comments swing between fascination and panic.

Why this should matter to you, even if you are not an influencer:

Your AI search profile exists. The only question is whether you know what it looks like.

Level 1, the visual self-image: Upload a photo of yourself and ask:

Level 2, the research:

The same exercise is worth doing for your company too: "What can a prospect find about us in five minutes, and what impression does it create?"

Googling yourself is an old habit. AI has made the same thing easier, better, and sharper.

If you know your AI search image, you can shape it: clean up old profiles, make the good work more visible, and avoid being surprised by the first question in a meeting.

This connects directly to the next topic. The quality of that kind of briefing depends entirely on how you make AI search.

Quick reality check: how often did you ask AI to research something this week?

Exactly. Same here. Comparing tools, checking markets, researching people before meetings.

The answers usually look good: structured, cited, polished.

But we noticed something: whether it is ChatGPT, Claude, or Perplexity, the result often becomes the same consensus soup. The same 20 blog posts, summarized in slightly different words.

This week we thought about why. The short version: there are two types of research, and your AI usually defaults to only one of them.

Web search, Deep Research, studies, documentation. This answers the question: what is factually true?

The research features from the big providers are very good at this.

The blind spot: they often pull from SEO articles and listicles, published by editorial teams or even by the vendors themselves. Age and incentives are not always clear.

This answers a different question: what are real people saying right now?

Reddit threads including comments, Hacker News debates, YouTube comments, Instagram, TikTok.

The last 30 days, ranked by upvotes and discussion instead of Google ranking.

We tested both types on a simple question we hear all the time: "What is the best AI meeting notes tool?"

The standard research answer was basically: "Use Fathom" or "use otter.ai". Not bad at first glance.

Then we opened the sources: 15 of 24 belonged to a tool vendor. Almost no real user voices. Advertising, disguised as research.

Then came the pulse research, after 86 real posts. The community was talking about something completely different.

Otter.ai had just been sued over hidden recordings. Krisp.ai had trust problems in some discussions. And the trend was clearly moving toward local, bot-free tools.

Language note: one of our favorites is the German, GDPR-friendly tool Jamie.

Two types of research, two worlds. That is why we recommend doing both: first the fact check with the normal research function, then the pulse check through real user voices.

Social signals can be hard to retrieve cleanly. We use one of the most popular skills for this: /last30days.

If you do not yet use Claude Cowork, ChatGPT Work, Codex, or similar tools, you can also try this prompt:

Our rule of thumb since then:

If you ask like everyone else, you know what everyone else knows.

As mentioned above, social signals are not always easy to collect.

We use a free open-source skill called last30days. It searches Reddit, Hacker News, YouTube, TikTok, and more in parallel, with real links and dates.

If you already work with Claude Cowork, Claude Code, Codex, and similar tools, ask your agent to install the skill:

The skill works out of the box. You can improve it further with scraping APIs and an X login.

Claude vs Gemini. GPT-7 vs Llama 5. Which AI lab ships AGI first. These are live Kalshi markets with real money on both sides, updated in real time as releases land. The person who follows model cards and tracks evals has a genuine edge here. If that's you, trade it.

Trade the model race

One final observation from our day-to-day work, because it may affect you too.

The dash has long been treated as an AI tell. Rolling Stone already calls it the "ChatGPT Hyphen", human writers self-censor, and students are suspected because of punctuation.

Nobody really knows why models love that mark so much. The best theory: too many edited quality texts were in the training data.

This whole "AI wrote it" versus "AI did not write it" topic is getting weird. People try to detect AI text, then stop reading it on principle.

We built an anti-AI writing-pattern skill a while ago because this pattern became so common.

But lately we often see the other side. We work with AI all the time and have many outputs prepared as HTML pages or document summaries.

Of course those prepared docs can look like AI. We are not hand-coding every HTML page from scratch.

But behind those AI-prepared reports are strong second brains, clean sessions, and a lot of real work with the AI.

Even when the text was written by AI, the substance can be excellent. Still, many people stop reading as soon as they recognize the AI look and think: "Ah, AI slop."

We now sometimes write it explicitly: "Please read it anyway."

The slop reflex was a useful filter in 2025. In 2026, it is becoming a bug.

Style no longer says enough about quality, in either direction. The better questions are: are the facts right? Is there real thinking behind it? Can I get value from it?

We do not really care who wrote a post, article, or report. Human or AI, the content has to hold up.

That's it for today. As always, thanks for reading. 🫶

Reto & Fabian from AInauten