5 Substantive LinkedIn Comments a Day: The Perplexity + ChatGPT Loop I Run Instead of Posting

I dropped my own LinkedIn posting schedule to once a week and filled the other four slots with substantive comments on other people's posts. Within a month, profile views were up 47%, and I'd turned two inbound DMs from strangers into booked calls. I would not have predicted that the comment box was the higher-leverage slot of the week.

That result is not a fluke. It is what the LinkedIn algorithm actually rewards in 2025. A working paper circulated by Chris Donnelly in early 2025 — repeated by Hootsuite, Agorapulse, and LinkedIn's own research — shows that commenting on 10–20 posts per week is associated with a roughly 50% lift in profile views and a 10% lift in your own post reach. The same paper shows that AI-generated comments get about 5x fewer author replies and 7x fewer audience engagements than human-written ones. Translation: comment more, but do not let the model ghost-write for you. The combination that actually compounds is what I call the Perplexity-research / ChatGPT-draft / human-edit loop.

Here is the loop I run every weekday morning, in five steps.

1. Pick your 5 leaders the night before (Perplexity)

The reason most comment strategies die in week two is that people open LinkedIn, scroll, and react to whatever the algorithm surfaces first. That gives you a day of commenting on whoever happens to show up — usually the same three creators, plus a couple of recruiters and the obligatory "I agree with X" replies. The compounding effect comes from commenting on a curated shortlist, every day, for 90 days.

The shortlist is the part most people skip. I keep a Notion table of 12–15 industry leaders (CMOs, growth operators, paid media leads, content heads at B2B SaaS companies — Business-to-Business Software as a Service) whose comments I want my name under every day. Each morning, I open Perplexity (an AI search tool that returns cited answers, like a chat-based Google with sources) with a single prompt:

Find the top 3 posts published in the last 24 hours by [name 1], [name 2], [name 3], [name 4], [name 5]. For each post, give me: the URL, the first 3 lines verbatim, and the 3 strongest comments already on it. I want to find posts where the existing thread is missing something.

Perplexity returns a 90-second digest. I pick the five posts I want to engage with — usually the ones where the existing comments are weak (lots of "great post!" or sycophancy, which is flattery that doesn't add information). A weak comment thread is an open lane. Commenting on a popular post with 200 replies is mostly wasted; commenting on a post with 4 mediocre replies is a stage.

2. Read the original post cold before you touch the model

Spend 90 seconds reading the actual LinkedIn post. Not the Perplexity summary — the real post. You need to know what the author actually said, because the model does not have that context, and the moment your comment references something the author did not write, you look like a bot.

I learned this the hard way. Last year I pasted a post URL into ChatGPT, took the output, posted it, and got a polite but devastating reply from the author: "Hi Song, I appreciate the engagement, but I never said CAC (Customer Acquisition Cost — the total marketing spend to acquire one new customer). I think you're responding to a different post." The comment was technically good, but it was responding to a different conversation. That kind of miss is unrecoverable in public, and once an author publicly corrects you, every reader in the thread sees the correction.

3. Draft with ChatGPT using a contrarian-first prompt

Now, and only now, you bring in ChatGPT. The prompt below is the one I have refined over the last six months. It is long on purpose — prompts that ask for "a thoughtful LinkedIn comment" are useless, because the model returns the same five-paragraph balanced-acknowledgment essay every single time. The prompt is opinionated on purpose.

You are helping me draft a LinkedIn comment on a post by [author name], who is a [role] at [company]. Here is the post:

[paste post]

Here are the top 3 existing comments so I do not repeat them:

[paste top comments]

Write a 40–80 word comment that does ONE of the following:

1. Pushes back on the strongest claim with a specific counter-example or data point
2. Adds a second-order consequence (a less obvious follow-on effect) the author missed
3. Names the one thing everyone in the comments is wrong about

Tone: peer-to-peer, not fan-to-creator. No flattery. No "great post." No "this resonates." Start with the substance, not a compliment. Use 1–2 short paragraphs. Do not use bullet points. Do not use hashtags. If you cite a stat, name the source in 4 words or less (e.g., "per HubSpot's 2024 report"). If you cannot support a claim, do not make it.

The instruction set matters. "Pushes back," "names the one thing everyone is wrong about," and "no flattery" are the three phrases that flip the model out of its default balanced-acknowledgment voice and into something an actual operator would write. The length cap (40–80 words) is the other lever — it forces the model to compress instead of hedging, because hedging (saying "on the one hand... on the other hand...") is what makes AI comments read as AI.

4. The human edit (this is the part everyone skips)

Treat the ChatGPT output as a first draft from a junior colleague. It is 70% right. Your job is the 30% that makes it yours.

Three edits I make every time:

• Replace the first sentence. The first sentence of an AI comment almost always opens with a soft acknowledgment ("This is a great reminder..." or "There's a lot of truth here..."). Real operators open with the substance. I delete the first sentence and rewrite it.
• Inject a specific number, name, or client story. AI comments are abstract. "We have seen CTRs (Click-Through Rates — the percentage of people who click a link) drop 23% in Q2 for SaaS clients in the $50K–$200K annual spend range" is a comment. "CTRs often drop" is not. The number does not have to come from a peer-reviewed study. It can be from your own work. Specificity is the only thing that makes a comment sound like it came from a person who has done the work.
• Read it out loud once. If it does not sound like something you would say in a real conversation, rewrite. If it sounds like a thought-leadership post compressed into a comment, it is too long.

This step is the difference between the 7x engagement lift and the 7x engagement penalty. Skipping it is the single most common reason AI-assisted comment strategies fail.

5. Post, then track three numbers every Friday

The output is not "I commented five times this week." The output is the change in three numbers, measured every Friday at the same time of day:

1. Profile views (LinkedIn analytics → who viewed your profile). The headline number. Five substantive comments per day, sustained, should add 30%–60% week-over-week by week three.
2. Inbound DMs from non-mutual connections. This is the quality signal. Anyone who DMs you from a comment thread is already pre-sold; the comment itself was the trust handshake.
3. Post impressions on your own post that week. A side benefit I did not predict: when you show up in 25 comment threads a week, your own posts reach more people. The algorithm appears to treat consistent commenters as community members and gives their own posts a small but real distribution boost.

Track these in a spreadsheet. After four weeks, you will see whether your specific leader list, your specific prompt, and your specific topic mix are working. If profile views are flat, change the leader list, not the prompt. The list is the variable with the most leverage.

What this is not

This is not a "comment more, win more" pep talk. Commenting on 50 posts a day of "love this!" is worse than commenting on five. LinkedIn's algorithm specifically downranks generic engagement, and a heavy volume of low-quality replies can trigger the same "spammy engager" pattern that gets new accounts throttled. The compounding benefit comes from depth, not breadth.

The other thing this is not: a replacement for posting. You still need a post of your own every week, because posts are how the algorithm learns what you are about. Comments are how you surface in the right conversations; posts are how you anchor them. The five-comments-a-day loop is a force multiplier on whatever else you are doing on LinkedIn — not a substitute for it.

If you have a small audience and a full day job, the comment loop is the highest-ROI thing you can do on LinkedIn. A 1,200-word post costs me 90 minutes to write well. Five 60-word comments cost me about 25 minutes total, and they put my name in front of more ICPs (Ideal Customer Profiles — the type of buyer you most want to reach) than my own post did. Once you have done it for a month, you stop arguing with that math.