Sponsored by
In this issue:
🧾 Copy my 7 lines so AI tells you when it's guessing
🗂️ Granola: how Lokesh preps for repeat client calls
📣 Buffer: post to social from inside ChatGPT or Claude
⌨️ Goldfish: AI replies that use what's on your screen
🧠 Trivia: who drew the "Bloper" cartoon?
⚡ The 7 lines I gave ChatGPT and Claude
A few months ago I put a statistic from ChatGPT into a client report.
The client's team checked it before presenting it internally. The study ChatGPT cited did not exist. The number was completely made up.
The client was understanding. But that conversation is not something I want to repeat. Sitting there, knowing I handed someone something an AI invented.
So I put 7 short instructions into my ChatGPT and Claude memory. Now they tell me when they're guessing.
🔍 9 out of 10 details
Sometimes I give AI 9 out of 10 details, and it invents the 10th. Only sometimes. And so well I don't notice.
Most of my checks find nothing. Then one finds a made-up detail.
I can't tell which answer it will be. So I started checking everything. That eats the time I was using AI to save.
I want AI to say "I don't know" instead.
The last of the 7 lines stops most of that.
State of Product reveals what AI still hasn’t solved
80% of product professionals say AI helps them ship faster, but customers aren’t seeing value any sooner. Atlassian’s State of Product 2027 explores the gains AI is delivering and the challenges that remain, from decision-making that hasn’t kept pace to gut instinct overriding customer evidence.
🧰 3 tools worth a look
🗂️ #1 Granola: meeting notes, no bot on the call
Lokesh, our community manager, uses it before every repeat client meeting.
Before every repeat client meeting, I had to rebuild the client's history from scattered notes. Last call's notes, the call before that, then my email for a deadline from three weeks ago. After five minutes I remembered the project. I still didn't remember what had happened in the relationship.
So I gave each client their own folder, and every meeting the same six headings. Granola listened in the background and wrote the summary. Then I connected it to Claude and told it which folder it could read. The rule that mattered most was the one about what it must not read.
Before the next meeting, I asked for the last decision, every unfinished promise and its owner, concerns the client raised more than once, and three questions I had to ask.
What the brief missed told me what to ask for next time. Now it keeps separate lists for decisions, my actions, the client's actions and dates. Anything the meeting never settled is marked "not agreed" instead of guessed.
The first meeting gave me good notes. The second gave me the right context. After four meetings, it started to show repeated problems and changes.
⚠️ Where it breaks: Lokesh is on Business ($14 per user per month, unlimited history). On the free plan, Granola keeps 30 days of history, and Claude can only read your notes from the last 30 days. With monthly meetings, Claude sees one.
📣 #2 Buffer: post from inside ChatGPT or Claude
We use ChatGPT and Claude to draft social posts. What if they could post too, from right inside the chat? With Buffer, now they can: draft, schedule, post.
Connect it in Buffer: Settings → Integrations → pick ChatGPT or Claude → connect.
Free plan: 3 channels, 10 scheduled posts per channel at a time. A slot opens again when a post goes out.
Try one loop: take a campaign idea you already have. Get two caption options from Buffer's AI Assistant. Save the better one as a draft, then schedule it.
⚠️ Save it as a draft first. A scheduled post goes out at its time, checked or not.
⌨️ #3 Goldfish: replies that use what's on your screen
It watches your work on your computer (threads, docs, tabs) and keeps a memory of it. In any text box, like Gmail or Slack, tap the right Option key on a Mac or right Shift on Windows. It reads the screen, pulls what it remembers, and drafts a reply. You edit or send. Nothing goes out until you decide.
Mac and Windows. Free during beta, no price yet for after.
⚠️ It captures what's on your screen. Goldfish says the memory stays on your computer, and AI requests are never stored or used for training.
🧠 Quick trivia
This cartoon went viral on X in August. Look at the signature in the corner.
Quick trivia: the signature says "Bloper". Who drew this cartoon?
👋 Back to that client report
Owning it out loud, in that moment, was very uncomfortable.
The thought in my head was: "I'm the human in the loop. I should have checked it."
I clearly apologized to the client.
I am the one to blame. What I deliver is mine. The tool is just a tool.
The 7 lines I use are here:
Dan Rice · AI Signal · Every Tuesday
Read once. Use AI better all week.