Getting Reflect Chatty With GPT in a Blank Note
So Reflect has this pretty chill feature where you can start a new note and just talk directly to GPT by typing @GPT and then whatever you’re thinking. The thing is, it doesn’t always feel intuitive unless you’ve already played with it. It kind of blends into the page like it’s just another bullet point, not a thing you can actually use to trigger AI writing. The weird part? If you hit Enter after @GPT and don’t type anything yet, you’ll still get a little flashing cursor like it’s waiting… just staring back like “uh, well?” ¯\_(ツ)_/¯
In one case I had, I wanted to brainstorm newsletter name ideas, so I typed:
– @GPT suggest 10 playful newsletter names for a weekly automation workflow roundup
And it delivered! Like 5 seconds later, it just appeared underneath. No loading bar, no drama, it was just like “here’s your serverless magic.”
But here’s what I kept forgetting: GPT in Reflect doesn’t see anything above that bullet. Nothing in the same note, nothing in nearby bullets. If I gave background in bullets earlier like:
– Audience: indie operators and zappers
– Tone: casual, hands-on
– Values: honesty, weird workflows, low-no-code
…and then somewhere below I did @GPT write a tagline — it had zero context. So I had to remind myself (and now you), GPT in Reflect only reads what you give it right there. Like a text message, not a memo thread.
One easy fix is to write the request as a fully detailed ask each time, like:
– @GPT write a 1 line tagline for a weekly newsletter for indie operators and zappers, tone casual and clever, focused on messy wins in automation
Yeah it’s more typing, but the results are way smarter.
Building Collection Lists for Weekly Topics
This part tripped me up at first — if you want an inbox of topic ideas you can pull from each week, you’d think maybe tagging the ideas might be enough. But that gets chaotic fast. Tags get overloaded or show up from the wrong origin. What actually worked way better for me was just starting a new note with something like:
“Newsletter Topics Brainstorm” and then nesting GPT prompts under bullet points. So visually, it looked more like:
– Workflow writeups
– @GPT list 10 underrated Zapier + Airtable use cases for small teams
– GPT playgrounds
– @GPT give 8 odd-but-useful prompts people can try with GPT in Slack
– Reader submissions
– @GPT suggest clever ways to ask readers for their own automation stories
And then every few weeks, I duplicate that note, run a few of the GPT lines again (they give new answers each time), and pull ideas into my writing queue.
Quick note: if you don’t hit Shift+Enter after the @GPT line, it’ll sometimes output directly into the next bullet — which messes up your formatting later if you wanted to collapse it. I started always adding a new sub-bullet and then prompting inside that so I can collapse sections and keep it tidy 🙂
Also, if you’re wondering: GPT doesn’t remember anything across lines. So if it misunderstands, you have to restate everything. Don’t rely on threading.
Using GPT in Reflect Daily Notes Without Clutter
It’s so tempting to use GPT in the daily note — especially when I don’t want to switch apps just to test an idea. But the clutter it creates got overwhelming fast.
I used to write things like:
– @GPT write a tweet about that automation bug from earlier
And then I’d get:
– Sure! How about:
– “When your onboarding Zap decides everyone’s name is ‘Test User’ 😬 It’s like calling your whole mailing list ‘Mom.’ #zapierfail”
Cute. But then it’s all sitting right in my daily log, so when I go back later to find actual meeting notes or ideas, there’s just these weird GPT fragments scattered everywhere.
Eventually, I adopted a pattern:
– In the daily note, write:
– [ ] GPT prompt: come up with alt headlines for today’s post
Then later in the week, I go around and collect all the GPT prompts and dump them into a “GPT for the week” note that’s just my scratch pad. It keeps my dailies clean and centralizes all the randomness.
If Reflect had a way to “hide GPT responses unless I collapse them” that’d be cool, but until then, I just extract the gems manually. Kind of like foraging.
Sorting Bad GPT Responses and Trying Again
OK, real talk: a lot of GPT’s early responses are trash. Not like outrageously bad, but boring bad. Which is worse.
When I asked:
– @GPT list newsletter titles for automation enthusiasts
I got textbook clichés like “Automation Weekly” or “The Flow Update” — like, come on. So what I started doing was refining the prompt *after* reading the first batch.
If the first result was too boring, I’d try:
– @GPT try again but with more playfulness, and avoid the word ‘automation’
That helped a little. But the actual best trick? Adding examples to the prompt. Like:
– @GPT indicate the style I want with examples like “Inbox Shenanigans” or “Zapped and Loaded,” and then generate 10 fresh newsletter name ideas
Boom. That works.
Also, sometimes GPT gets stuck in loops. It’ll keep generating versions like:
– Weekly Automation
– The Weekly Flow
– Weekly Something
If that happens, switching the word “generate” to “invent” or “playfully riff unpredictable titles” can reset the tone.
And if all else fails: copy the same @GPT line to a new note or a different time of day and paste it again. I swear it changes tone slightly depending on who-knows-what.
¯\_(ツ)_/¯
Filtering What GPT Output Is Worth Keeping
Not gonna lie: the vast majority of raw GPT output gets deleted. You kinda have to accept upfront that most outputs are rough drafts you’ll mine for a good line or phrase, but not copy/paste as-is.
Here’s my filtering system:
1. If it made me laugh or pause — I keep it.
2. If it gave me 10 ideas where 1 is actually decent — I grab that 1, delete the rest.
3. If it got something obviously wrong (like recommended APIs that don’t exist), I stop trusting anything else in that list.
What works okay is turning GPT outputs into a vault of rough stones. Think of them less as answers and more like prompts that created interesting terrain. When I revisit my GPT notes later, I usually just bullet the 1 or 2 best bits, and the rest get trashed.
Also… bold tip: never forget the GPT line you used. If you just keep the answer but lose the prompt, you’ll never be able to recreate the logic later. Always leave the @GPT bullet directly above the response. Otherwise, you’ll be squinting at your own words wondering “why did I think this was good again?” 😛
Embedding Reflect Note Links Into Your Prompt Flow
Guess what sucks? GPT in Reflect can’t actually open other Reflect notes. So linking them doesn’t give context in real-time. But I still embed Reflect [[backlinks]] or use === divider titles when I want to show GPT the structure indirectly.
For instance:
– === GPT Experiments: Newsletter Names ===
– @GPT generate clever newsletter name ideas given the style notes below
– STYLE:
– audience: indie automation lovers
– tone: clever, 90s internet, human, not AI
Sometimes I even add fake mini headnotes like:
“From the journal of a chaotic micro-founder testing GPT for creative naming workflows”
It just helps GPT catch the vibe. No idea if it actually changes things — but results feel better. More weird. More usable.
I’ve probed a few times adding note titles like [[newsletter experiments]] but, again, GPT doesn’t seem to interpret that as anything real unless you paste in actual note content. Sooooo maybe don’t rely on internal link structure alone.
Accidentally Prompting GPT in ReadOnly Journal Notes
This one messed me up and wasted like 20 minutes.
I created a journal entry template in Reflect that had some weekly templates written in — like:
– Weekly Theme: [type here]
– @GPT suggest 3 tasks I could take on this week based on the theme above
But I had forgotten that after I made it a “template,” the base note became read-only for GPT somehow. No error showed. It looked like GPT was loading — little pause, the bullet was still blinking — but nothing ever came out.
Turns out Reflect doesn’t actually throw any warnings. It just… does nothing. Clicking on the bullet didn’t help. There was no spinning wheel or sad trombone.
I only figured it out because I pasted the same prompt into a fresh note and boom — instant response.
Soo… word to the wise: GPT can’t write in templates or protected notes. If you’re ever stuck waiting on a blank output for more than 15 seconds, try copying into a new normal note and try again. Doesn’t make sense, but it works.
¯\_(ツ)_/¯
Pacing Yourself Before Your Notes Become Chaos
Yeah. This happened to me around week three. GPT in Reflect was magic at first and then kind of became too much too fast. I had 60+ notes with random bullet points titled things like “better subject line” or “name ideas 2.” None of them were finished. Half weren’t labeled. A few were just copied prompts that didn’t go anywhere.
My fix? I created a note called “GPT Output Crate” and started moving all scrap GPT responses into it — grouped by vague topic. It wasn’t clean, but at least it was contained. Once a month, I reread that note and salvage anything good. Most get deleted.
Also, pro move: if something’s not useful immediately, collapse it.
Reflect’s collapsing bullet view isn’t flashy, but it’s gold when you’re dealing with GPT mess. Better than tagging, better than canvases — just collapse the chunk… forget it exists for a while. Come back later with new eyes.
Most stuff’s better after sitting anyway.