How I Never Run Out of LinkedIn Post Ideas (AI Tools I Actually Use)

For a solid year and a half, I posted on LinkedIn the way most people do it: sporadically, guiltily, and only when something exciting happened at work. Then I'd go quiet for three weeks because nothing "post-worthy" had happened, and I'd tell myself I just wasn't a "content person."

Turns out I wasn't a content person because I was waiting for ideas to show up instead of having a system that generated them for me. That sounds like a small distinction but it changed everything. I now post 3-4 times a week and haven't had a "what do I even write about" moment in months. Not because I got more interesting — I promise my job did not suddenly become fascinating — but because I stopped relying on inspiration and started using a few AI tools that do the unglamorous part of content creation for me.

Most posts about "using AI for LinkedIn content" stay frustratingly vague — "I use ChatGPT for ideas" without ever showing what that actually means in practice. So this one's going to get specific: exactly which tool does which job, what I actually type into it, and what comes out the other end.

The core problem was never "what to write"#

Here's the thing nobody tells you about running out of ideas: it's rarely a creativity problem. It's a capture problem. Interesting things happen to you constantly — a weird client conversation, a mistake you fixed, a hot take you had in the shower — and almost none of them get written down in the moment. By the time you sit down on a Sunday to "plan content," you've forgotten 90% of what actually happened that week.

So the first shift wasn't really about AI at all. It was accepting that I needed to capture raw material constantly, and only then use AI to shape it into something postable. Trying to generate ideas from nothing, cold, in front of a blank Notion page, was always going to fail. Generating ideas from a messy pile of half-formed thoughts turned out to be a completely different and much easier problem — and this is where each tool below actually earns its place.

Tool 1: ChatGPT as my running idea catcher#

I talk to ChatGPT throughout the week the way some people talk to a voice memo app. Not formally — I'm not writing prompts in the moment. I'll open the app on my phone while walking somewhere and just ramble into the voice input: "had a really annoying meeting today where a client wanted three contradictory things and didn't seem to notice, feels like there's a post in here about scope creep but I don't know the angle yet."

Here's the specific setup that makes this actually work instead of just being a chat log I never revisit. In ChatGPT's custom instructions (the settings where you tell it how to behave across every conversation), I've added this:

"When I send you a rough voice-to-text ramble about something that happened during my day, don't respond with advice or a polished summary. Just acknowledge you've logged it in one short line, and ask one specific follow-up question if something's unclear. At the start of any new conversation where I say 'idea review,' pull together everything I've rambled about recently into a numbered list with a one-line theme for each."

That second part is what actually makes it useful. Without it, I'd have forty scattered conversations and no way to see them together. With it, I can type "idea review" on a Sunday and get something like: "1) Scope creep story from Tuesday's client call — theme: clients don't realize what they're asking for. 2) The onboarding doc rewrite — theme: documentation nobody reads is worse than no documentation. 3) The intern's question that stumped you — theme: explaining basics reveals what you actually believe."

I do this capture step four or five times a week. It genuinely takes under two minutes each time, and it's the single habit that made the rest of this system possible. Before I had this, my "content ideas" existed only as a vague feeling that I'd had a good thought sometime that week and then lost.

Tool 2: Claude for pressure-testing the angle before I write#

Once I've got that numbered list from ChatGPT, I don't ask ChatGPT to just "write a post about it." I've found Claude specifically better for the next step, which is figuring out if an idea actually has a point of view or if it's just an anecdote with no conclusion.

What I actually paste in is something like: "Here's a rough story: [the scope creep thing]. I think there's a post here but I don't know what I'm actually arguing. Push back on me — what's the boring version of this take, and is there a less obvious angle?"

The difference from ChatGPT here is specific: Claude tends to actually disagree with me or point out that my first instinct is the most common take on the topic, rather than immediately validating whatever angle I threw at it. On the scope creep example, it pointed out that "clients don't know what they're asking for" is the take everyone posts, and asked whether the more interesting story was actually about the moment I realized I'd been agreeing to scope creep for two years because saying no felt confrontational. That reframing — from "clients are the problem" to "I was the problem" — is a genuinely better, more specific post, and it's not one I would have landed on by asking for "10 LinkedIn post ideas about client management."

I do this back-and-forth for maybe five or six exchanges before I have an actual thesis sentence I'm confident in. That thesis sentence is the only output I need from this step — I'm not asking for a draft yet.

Tool 3: Turning the thesis into a first draft, then killing the AI voice#

With a real thesis in hand, I go back to ChatGPT for the actual first draft, but with a very specific instruction rather than a generic "write me a LinkedIn post about this" prompt:

"Write a first draft LinkedIn post arguing this: [thesis]. Use the scope creep story as the opening hook, told specifically, not generically. No em-dashes. No 'here's what I learned' closing line. No listicle format unless I ask for one. End on the actual point, not a summary of the post."

That last constraint list exists because I got burned early on by exactly the AI-post rhythm everyone's learned to spot: short punchy opener, em-dash-heavy middle, "the lesson? ___" closing line. Once you've read a hundred LinkedIn posts with that shape, it stops reading as insight and starts reading as template.

Then — and this is the step people skip — I rewrite every sentence of that draft in my own words before it goes anywhere near the "post" button. Sometimes that means keeping the paragraph structure and changing most of the actual wording. Sometimes I read the draft, realize it reminded me what I actually wanted to say, and delete it entirely and write from scratch. Either way, nothing goes up that I haven't personally touched line by line. The AI draft's job is to get me unstuck from the blank page, not to be the final text.

One more specific trick: I'll paste my own rewritten draft back into either tool and ask, "does any of this still sound AI-generated — flag specific phrases, not just vibes." It's caught things I didn't notice myself, like an unconscious habit of writing sentences in threes ("faster, cheaper, and more reliable" — that rhythm). Once a tool points it out a few times, you start hearing it yourself, which is a genuinely useful side effect of using AI as an editor rather than a ghostwriter.

Tool 4: Gamma for carousels, without learning design software#

I'm not a designer, and for a long time that meant my only content format was plain text posts. Carousels reliably get more reach on LinkedIn, but opening an actual design tool and building slides from scratch was enough friction that I never did it — I'd think about making one, estimate it'd take 40 minutes, and just post text instead.

Here's exactly what I do in Gamma: I paste my final, already-rewritten post text in, and I use this specific instruction rather than the default "make a presentation" prompt: "Turn this into a LinkedIn carousel — one idea per slide, short enough to read in three seconds each, no more than 8 slides, plain language, no bullet-heavy slides."

The default output without those constraints tends to cram too much text onto each slide, because it's optimizing for a generic presentation, not a scroll-stopping carousel. With the constraint, I get something close to postable on the first try — a title slide, then one point per slide building toward the thesis, then a closing slide with a single question or call to action.

I still go in and manually adjust colors to match my usual palette, occasionally reorder a slide, and sometimes trim a sentence that's still too long. That manual pass takes about ten to fifteen minutes now, versus the forty-plus minutes (and the accompanying dread) it used to take building a carousel from a blank template. That gap — not "zero effort" but "way less effort than the alternative" — is honestly the realistic bar AI tools clear for me across this whole system, not some fantasy of fully automated content.

Tool 5: Napkin AI for the diagram posts that quietly do really well#

Some of my best-performing posts aren't stories at all — they're simple visual explainers of a concept, the "here's a diagram of how X actually works" format. I used to skip this format entirely because I have zero instinct for laying out a diagram by hand.

What I actually do: I write the paragraph explaining the concept normally, as if it were just going to be a text post, then paste that paragraph into Napkin AI and let it generate a visual structure from the text automatically — usually a flow, a comparison, or a simple hierarchy depending on what the paragraph is actually describing. I'll usually generate two or three variations and pick whichever one represents the idea most honestly, since sometimes the auto-generated structure forces a relationship between ideas that wasn't actually what I meant.

The genuinely surprising part is how much better a rough, slightly imperfect diagram performs compared to posting the exact same idea as a wall of text. I don't think the diagrams are winning because they're beautiful — mine usually aren't. I think they win because scrolling past a paragraph takes no effort, but a diagram makes people pause for half a second to parse it, and that half-second pause is enough to get someone to actually read the caption underneath.

Tool 6: Notion AI as the backlog that never lets me start from zero#

The single biggest change to my actual week-to-week routine wasn't any individual tool above — it was moving from "think of an idea, write it, post it" as a one-shot process to keeping a running backlog inside Notion AI. Every rambling voice dump, every angle Claude helped me sharpen, every draft that didn't quite work gets dropped into one running database, tagged loosely by topic (client work, career, hot takes, "not sure yet").

The AI layer inside Notion is what makes this a living system instead of a graveyard of abandoned drafts. When something happens in the news or at work that's suddenly relevant, I can ask it "do I have anything already written that relates to [topic]" and it'll surface a half-finished draft from six weeks ago that I'd completely forgotten about — sometimes it just needed the current event as the missing hook to finally make sense. Without that recall, a backlog is just where good ideas go to die quietly. With it, I genuinely never sit down on a Sunday with an empty page, because there are always at least three or four half-baked things waiting for the piece that makes them click.

What I stopped doing (the honest part)#

A few things that didn't work, because I think they're as instructive as what did:

• I stopped trying to post daily. The pressure to fill a daily slot pushed me toward generic content just to have something up. Three to four solid posts a week beats seven forgettable ones, and nobody's actually counting your streak.
• I stopped asking any tool for "post ideas" cold. "Give me 10 LinkedIn post ideas about leadership" produces the blandest, most recycled advice you've read a hundred times, because it has nothing real to draw from. Every post that's actually landed for me traces back to something specific that happened, run through AI for structure — never the reverse.
• I stopped copy-pasting first drafts. It's slower to rewrite every time, but it's the entire difference between posts that sound like me and posts that sound like everyone else's feed right now — and readers can tell, even if they can't articulate why.

The honest summary#

None of this required an elaborate content calendar or a big stack of paid tools. It required accepting that ideas come from capturing real moments, not staring at a blank page, and then using specific tools for the specific bottlenecks I actually had: ChatGPT for constant low-effort capture, Claude for pressure-testing whether an idea has a real point of view, Gamma and Napkin for visual formats I couldn't build by hand, and Notion AI so nothing gets lost between "I had a thought" and "I actually wrote about it."

If you're stuck the way I was, skip the generic "AI prompts for LinkedIn content" listicles entirely and just start capturing the annoying, funny, or confusing things that happen to you this week. The ideas were never the hard part. Having somewhere to put them before you forgot them was.

Want to explore the tools mentioned here yourself? Browse AI writing tools or the full AI tool directory to find what fits your own workflow.