Does a First Comment Change AI Post Reach on LinkedIn?

After LinkedIn's March 2026 Authenticity Update, the first comment became the most misunderstood variable in post distribution. Across hundreds of accounts we run at SocialNexis, we tracked a clean before and after: accounts that kept a link in the first comment saw median impressions drop 55-65% from their pre-patch baseline on identical content. Accounts that switched to a link-free substantive comment kept their reach.

The Real Effect of a First Comment on LinkedIn Post Reach

The first comment matters because LinkedIn decides most of a post's fate before most of your network has seen it. A new post goes out to 1-2% of the author's network in the first 30 minutes. That sample has to produce a 5-10% engagement rate for the post to advance to the next distribution stage. Below roughly 2% engagement, the post stops moving and never recovers, no matter how good the writing is. So the useful question is not whether a first comment helps in some general sense. It is whether one comment can change the numerator of a very small fraction inside a very short window. It can, and that is why a variable this small ends up shaping the final impression count so heavily.

The mechanism is not mysterious. A comment from the author adds a second content surface to the post, gives an early reader something specific to respond to, and starts a thread that other people can reply into rather than having to open a blank comment box. LinkedIn Engineering's own write-up on feed dwell time confirms that passive attention is a ranking input, not just clicks and reactions. A first comment that carries a real idea keeps a reader on the post longer while they read it, which is scored even when nobody taps anything. A first comment that says nothing keeps nobody anywhere.

The author's presence in the comment section, not just the first comment itself, is where the largest published numbers sit. Posts where the author responds to comments within the first 30 minutes receive 64% more total comments and 2.3x more views. Buffer's analysis of 72,000 LinkedIn posts found that replying to comments within two hours lifts engagement across the post's full lifecycle by about 30%. Those two figures describe the same behavior at different resolutions: the author staying in the thread while the thread is still young. A first comment is the cheapest way to be in the thread from the start, because you do not have to wait for someone else to show up.

Timing inside that window is measurable at a finer grain than most guides admit. Testing first-comment timing across 300+ accounts, we found the 0-5 minute window outperformed the 5-15 minute window by roughly 18% on Stage 2 advancement rate. The 5-15 minute window, though, produced better comment thread depth, meaning more replies from other people rather than more raw comment count. Those are two different wins. Early comments buy you the ticket into the next distribution stage; slightly later comments land while real humans are already reading and are more likely to get answered.

Where conventional advice overstates the case is in treating the first comment as a boost button. It is closer to a qualifier. The comment can push a post over the Stage 1 threshold when the post was already close, and it can add depth to a post that people wanted to talk about anyway. It cannot rescue a post that nobody in the initial 1-2% sample wanted to read. We have watched accounts run a disciplined first-comment routine on weak posts for weeks and produce nothing, because the ceiling is set by the post and the floor is set by the account. If you are trying to separate these effects in your own data, it helps to understand how LinkedIn comments and posts affect reach differently before you attribute a good week to your comment habit.

The failure mode we see most often has a simple shape, and we call it the placeholder comment. The author treats the first comment slot as parking: a link, a bare question like thoughts?, or a repeat of the post's hook. All three occupy the highest-leverage moment in the post's life with content that generates no dwell time, no reply, and no save. The slot gets used, the box gets ticked, and the distribution signal stays flat. The rest of this guide is mostly about what to put there instead, and what to keep out of it.

Putting a Link in Your First Comment Hurts LinkedIn Reach in 2026

The link-in-first-comment workaround is closed. LinkedIn's algorithm now detects what practitioners have started calling bridge behavior: a post with no link in the body that funnels viewers toward a comment containing one. Since the March 2026 Authenticity Update, the platform applies roughly the same 60% reach suppression to that pattern as it does to a link placed directly in the post body. The workaround worked for years because the detection was positional, not behavioral. It stopped working when the detection started reading the pair of objects together rather than scanning the post text in isolation.

The penalty on the comment itself is worse than the penalty on the post. Richard van der Blom's 2026 LinkedIn Algorithm Report, built on analysis of 1.3 million posts, found that comments containing external links see up to 80% visibility reduction. Read that alongside the bridge behavior finding and the tactic collapses from two directions at once: the post loses reach for pointing at the link, and the comment carrying the link loses most of its own distribution, so fewer of the people who did see the post ever see the thing you were routing them to. You pay the toll and do not get the trip.

The published figures do not agree with each other, and that is worth resolving rather than averaging. Van der Blom's 2026 report puts the median reach reduction for a single external link in the post body at 18.8%, which is far more conservative than the 60% number that circulates in practitioner writing. Both can be true. A median across 1.3 million posts flattens accounts that barely get touched and accounts that get hammered into one figure, while the 60% estimate describes what the penalty looks like when it lands hard. In our own data the spread tracks account authority, posting frequency, and content type: a text post from an established account with steady engagement absorbs a link far better than a carousel from an account that posts twice a month. We do not publish a single point estimate for the penalty because we have never observed a single point estimate.

Authorship of the linked comment mattered even before the patch. Kristen Sesto's controlled placement test found that a first-comment link posted by the post author caused roughly a 15% reach drop, while the same link posted in a comment by another user caused only about a 5% drop. That asymmetry is the tell. If the penalty were purely about the presence of a URL in the comment section, both cases would score the same. They did not, which means LinkedIn was already reading the relationship between the post author and the linked comment as a distinct signal, and the March 2026 update escalated enforcement on exactly that relationship.

We tracked the inflection point directly because we had the same accounts running the same content strategy on both sides of it. Accounts that continued the link-in-first-comment tactic after the March 2026 patch saw median impressions fall 55-65% versus their pre-patch baseline on identical content. Accounts that switched to link-free first comments with substantive text held their reach or improved it. This is the comparison the public research cannot produce, because third-party researchers scrape posts rather than operate accounts, and nobody scraping the feed can hold content constant across a policy change. We could, and the curve is not subtle.

The same update explains a second collapse people usually discuss separately. The March 2026 Authenticity Update escalated enforcement against engagement bait, automation pods, and external link spam, and it is the same change that dropped poll engagement rates to 0.07%. If you are trying to work out whether the update hit you, that is a useful cross-check: accounts that lost reach on polls and on link-in-comment posts in the same period were caught by one policy change, not three unrelated ones. It is also a reminder about direction of travel. Every tactic that update targeted was a way of getting distribution without giving the reader something worth reading.

The practical answer for links has not changed as much as the panic suggests. If the link is the point of the post, put it in the body and accept the cost, because the bridge does not save you anything now and it costs your comment section as well. If the link is optional, leave it out and let people ask. We use both patterns depending on whether the post exists to drive a click or to be read, and we no longer run any variant of the workaround on managed accounts.

Does the LinkedIn Algorithm Treat the Author's First Comment Differently?

Yes, and it treats it differently in both directions. A substantive, link-free first comment from the author, posted in the first few minutes, functions as a positive engagement signal: it seeds the thread, gives early readers a reply target, and counts toward the initial engagement rate that determines whether the post advances. That is the upside. The downside is that the same comment is held to a stricter standard than a comment from a stranger, on content and on the manner in which it arrives.

Sesto's controlled test is the cleanest published evidence of the content-side asymmetry. An author's own link in the first comment cost roughly 15% of reach; the same link in a third-party comment cost about 5%. The post author is the party with the incentive to route traffic, so the algorithm scores an author-placed link as promotional intent and a stranger-placed link as ordinary conversation. Every recommendation in this guide about keeping the author's first comment link-free follows from that asymmetry rather than from a general suspicion of links.

There is a second asymmetry that has nothing to do with what the comment says. When we tested automated first comments posted by an author-controlled agent within seconds of publication, LinkedIn's engagement-velocity classifier read the pattern as inauthentic. The consequence was not a ban, not a warning, not a restricted account. It was silent suppression: the post was served to 1-2% of its normal initial sample and never advanced past Stage 1. From inside the account, that looks exactly like a post that flopped. Manual first comments posted two to four minutes after publication did not trigger the pattern, on the same accounts, with the same comment text.

That failure mode is worth naming because it is invisible by design. Call it the zero-second comment. The post publishes, the comment lands before a human hand could have typed it, and the distribution curve goes flat with no notification attached. Authors who hit it conclude their content was bad and rewrite the hook. The hook was fine. The timestamp gap between publish and first comment was the problem, and no amount of editing fixes a timing signature. LinkedIn's own Help Center is explicit that comment visibility may be limited for comments created through automation tools or excessive comment creation, which is as close to a public confirmation of this behavior as the platform gives.

The origin of the comment also appears to carry weight. Accounts we operate on real-browser, home-IP agents consistently showed better comment attribution in reach outcomes than accounts running through browser extensions or datacenter infrastructure, independent of what the comment said. The same text, from the same account persona, scored differently depending on how the session reached LinkedIn. Our read is that comment signals are discounted when they arrive from a session that already looks automated, which means high-quality comment writing does not recover reach that the infrastructure has already lost. The author's first comment is more exposed to this than third-party comments, because it is the one comment that is always posted from the author's session, on schedule, seconds or minutes after a publish event that LinkedIn also logged.

Put those together and the answer to the question gets more specific than yes. The algorithm treats the author's first comment as a stronger positive signal, a stronger negative signal when it carries a link, and a more heavily scrutinized event for authenticity than any other comment on the post. It is the highest-leverage comment in the thread in both directions. If you want the broader map of which behavioral patterns get scored this way, we have written separately about LinkedIn activity patterns that trigger suppression, and comment timing sits in the same family as connection pacing and session fingerprinting.

LinkedIn's 4-Stage Distribution and the 30-Minute Window

LinkedIn runs every new post through a staged distribution pipeline, and the stages are narrow at the start. Stage 1, from 0 to 30 minutes, shows the post to 1-2% of the author's network. Stage 2, from 30 to 90 minutes, expands to 5-10% if the post clears the engagement threshold. Stage 3, from two to six hours, reaches 20-40%. Stage 4 is broader distribution beyond the direct network, which is where posts that go wide get their numbers. Nothing about this pipeline is generous with second chances. Each stage is gated on the one before it.

The gate is an engagement rate, not an engagement count. A post must produce 5-10% engagement from the Stage 1 sample within the first 60 minutes to advance, and posts that come in below roughly 2% are effectively killed from further distribution. On a small network that means a handful of interactions decides everything. On a large one it means the sample is big enough that a first comment is a rounding error and the post has to carry itself. Knowing which situation you are in changes how much effort the first comment deserves, and it is the reason we look at LinkedIn engagement rate benchmarks for B2B accounts before deciding whether an account's problem is distribution or content.

The timing data on comments is stark. Comments posted within the first 30 minutes after publication generate 3.8x more impressions on average than comments posted after 24 hours, 391 average impressions versus 104, based on analysis of 261,137 real LinkedIn comments. A separate figure sourced to van der Blom's methodology puts the reach boost for comments made within the first five minutes at 4.2x. Both numbers describe the same physical fact: a comment written into a post that is still being distributed rides the distribution, and a comment written into a post that has finished being distributed reaches whoever happens to scroll back. There is no third option where a great late comment catches up.

This has an implication most people skip. Those impression figures apply to comments you leave on other people's posts as well as your own. If you comment on a large account's post four hours after publication, you are writing into a thread whose distribution has already been allocated. The same effort spent in the first half hour of that post is worth several times more in exposure. Commenting strategy and first-comment strategy are the same clock problem viewed from two sides, and accounts that treat their own posts as time-critical while treating their commenting as a leisurely afternoon activity are leaving most of the available reach on the table.

The window is short and front-loaded, which argues for a specific shape of author behavior rather than a general instruction to be responsive. Across 300+ accounts, the 0-5 minute window beat the 5-15 minute window by about 18% on Stage 2 advancement, while the 5-15 minute window produced deeper reply threads. The pattern that captures both: post one substantive first comment near the front of that range, then move into active reply mode from 15 to 60 minutes while incoming comments are still arriving. That sequencing gets the early engagement rate that the Stage 1 gate reads, and it gets the threaded depth that pays off later in the pipeline.

What does not work is spreading author engagement evenly across the day. We have run that pattern deliberately, with the same total number of author comments and replies distributed across eight hours instead of concentrated in the first hour, and it does not reproduce the effect. By the time the author shows up at hour four, the post has already been graded. Stage 3 traffic is a consequence of Stage 1 and Stage 2 performance, so engagement that arrives after the gates have been decided adds conversation without adding distribution. If you only have one hour a day to spend on LinkedIn, spend it in the hour after you publish.

Most LinkedIn Advice Gets First Comment Timing Wrong

The single most repeated piece of first-comment advice is to post your comment immediately after publishing, ideally within seconds. That guidance was written for an earlier version of the platform and it now describes a failure mode. A comment that appears seconds after a publish event creates a timing signature no human commenter produces, because a human has to read the post, decide to comment, and type. LinkedIn's engagement-velocity classifier reads the gap, and the gap is the evidence.

When we tested automated first comments posted seconds after publication across multiple accounts, the posts were suppressed at Stage 1. They received 1-2% of normal initial distribution and never advanced. No notification, no flag, no account restriction, just a flat impression curve indistinguishable from a post that failed on its own merits. That indistinguishability is the reason the bad advice survives: nobody who follows it gets told they were penalized, so the tactic keeps getting recommended by people whose posts quietly underperform and who blame the content.

At the coordinated end of the spectrum, the numbers are worse. LinkedIn's AI detects engagement pod activity with 97% accuracy as of 2026. Pods of 15 or more people commenting within 90 seconds using short generic phrases are identified and suppressed. The detection is reading several things at once: comment velocity, network proximity between commenters, language patterns across the comment set, and IP clustering. A pod is a first-comment strategy with more participants, which means it fails for the same reason a zero-second solo comment fails, only with more signal for the classifier to work with. The posts a pod is meant to lift are the posts it buries.

Between those two errors sits most of the bad outcomes we see. Too fast on your own and you look automated. Coordinated with a group and you look like a pod. The advice that produces both is the same advice: treat the first comment as a mechanical step to be executed as quickly and reliably as possible. Reliability and speed are exactly the properties an authenticity classifier is built to notice. The behavior that scores well is the behavior that looks like a person who wrote something and then thought of one more thing to add.

There is a dimension of this that public guides do not cover, because covering it requires operating accounts rather than analyzing scraped posts. Comment attribution behaves differently depending on how and where the comment originates. Real-browser, home-IP activity is scored differently from datacenter or browser-extension activity, which means the infrastructure behind a first comment affects reach independently of what the comment says. We run on real-browser, home-IP agents, so we see the contrast directly: identical comment text, identical timing, different distribution outcomes depending on the session behind it. A guide written from the outside cannot see this variable, so it gets attributed to content quality or to luck.

The correction to the standard advice is small and specific. Do not post in the first seconds. Do not coordinate the comment with a group. Do not automate the comment from infrastructure that already looks automated. Post it yourself, a couple of minutes in, with something in it worth reading. Every part of that is slower and less scalable than what the popular guides recommend, which is the point: the tactics that scale cleanly are precisely the ones LinkedIn built classifiers to find.

Substantive First Comments, Not Generic Ones, Move the Reach Needle

LinkedIn's natural language processing reads what comments say, not just how many there are. Substantive comments, meaning 15 or more words with personal experience, specific details, or a real question, carry 3-5x more algorithmic weight than generic comments under 10 words. A first comment that reads Great post! is close to free in signal terms. It occupies the slot, it produces no dwell time, it invites no reply, and it teaches the ranking system nothing about whether the post is worth showing to more people. Length is a proxy here, not the mechanism, but it is a proxy the platform is measuring.

The shape of the comment section matters as much as its size. AuthoredUp's analysis of 621,833 LinkedIn posts found that posts with indirect comments, meaning replies inside comment threads, see 2.4x more reach than posts with only direct top-level comments. That reframes what a good first comment is for. It is not a comment that collects likes. It is a comment that produces a reply, which produces a thread, which produces the depth signal the algorithm pays for. A first comment ending in a real question outperforms a first comment ending in a summary for exactly this reason, and the difference shows up in reach rather than in comment count.

Research from van der Blom cited by Forbes in May 2026 points the same direction from a different angle: posts with fewer but deeper comment threads reached 3.2x more people and drove 8x more profile visits than posts with many generic reactions. The profile visit number is the one worth sitting with, because it describes what happens after distribution. Deep threads are read by people who then want to know who is talking. Reaction-heavy posts are scrolled past by people who tapped and moved on. If the point of the post is business rather than vanity, the thread is the asset.

The engagement hierarchy adds one more consideration. Saves carry roughly 5x the reach weight of a like and 2x the reach weight of a comment, per AuthoredUp's analysis of 3M+ posts. That makes a first comment designed to prompt a save more valuable than one designed to prompt agreement. A dense summary that a reader wants to come back to, a checklist compressed into a paragraph, a number they will want to quote later: these produce the save. A rhetorical question that invites nodding does not. Most first-comment advice optimizes for the cheapest signal in the hierarchy because likes are the easiest thing to ask for.

Three formats work reliably in our accounts. A TL;DR that compresses the post for people who will not read all of it, which earns dwell time and saves. A follow-up question specific enough that a stranger can answer it without rereading the post, which earns replies and thread depth. A bonus data point that did not fit in the body, which earns both, because it gives skimmers something to take and gives commenters something to argue with. All three are link-free, all three clear 15 words comfortably, and all three are written to be read rather than to be counted.

The format to avoid is the one that shows up most in templates: the first comment as a repackaged call to action. Follow me for more, DM me for the guide, check out our newsletter. Self-promotional calls to action in the author's first comment carry their own reach penalty, and that penalty compounds with the link penalty when both are present in the same comment, which they usually are. The comment slot is a place to add value to the post, not a place to bolt a conversion step onto it. Posts and comments do different jobs in the distribution system, and we have written more on how LinkedIn comments and posts affect reach differently for readers who want the mechanics rather than the tactics.

How to Write a First Comment That Boosts LinkedIn Reach Without Triggering Suppression

Timing first, because it is the constraint with the sharpest edges. Post your first comment two to four minutes after publishing, manually. The 0-5 minute window outperforms the 5-15 minute window by about 18% on Stage 2 advancement rate, so the early end of the range is where the distribution advantage lives. But anything in the first seconds trips the engagement-velocity classifier and caps the post at 1-2% of its normal initial sample, which costs more than the 18% is worth. Two to four minutes clears both constraints. It is early enough to seed Stage 1 and slow enough to look like a person.

Content next. Write a TL;DR, a bonus insight that did not fit the post body, or a direct question aimed at the first readers. Keep it to 15 words or more so it clears the substance threshold that carries 3-5x the algorithmic weight of a short generic comment, reference something specific in the post so it reads as a continuation rather than a bolt-on, and keep two things out of it: external links and self-promotional calls to action. Each of those carries an independent penalty, and when a comment contains both the penalties compound. This is the whole recipe. It is short because the failure modes are what need explaining, not the format.

Then watch your self-comment ratio, which is the constraint almost nobody discusses. The number of sequential comments you post on your own post without receiving replies from other people is tracked as a per-account quality signal. Our operational data shows accounts that posted three or more sequential self-comments without intervening replies from other users saw 20-30% lower reach on subsequent posts for 7 to 10 days. Read that carefully: the penalty landed on later posts, not on the post where the self-commenting happened. It behaves like a per-account quality score that decays over time rather than a per-post penalty, which means first-comment strategy has to be evaluated at the account level. One well-timed first comment is the target. Three quick follow-ups because your first one did not get traction is the thing that costs you next week.

After the first comment lands, change modes. Stop commenting and start replying. Actively responding to incoming comments between 15 and 60 minutes after publication deepens the thread, spreads engagement signals across the distribution window instead of clustering them at minute three, and produces the threaded reply structure worth 2.4x reach. This is where most of the sustained gain comes from. Posts where the author responds within the first 30 minutes receive 64% more total comments and 2.3x more views, and Buffer's 72,000-post analysis puts the lifecycle engagement lift from replying within two hours at about 30%. The first comment opens the door; the replies are what walks through it.

A few operational notes that come from running this at volume rather than on one account. Write the first comment before you publish, not after, so the two to four minute gap is spent posting rather than thinking. Vary the gap between posts instead of hitting the same interval every time, because a consistent delay is itself a signature. Do not schedule the comment through the same automation that published the post if you can avoid it, since comment signals are discounted when the session behind them already looks automated, independent of what the comment says. And if a post is clearly dead by the 30 minute mark, leave it alone. Piling author comments onto a post that failed Stage 1 does not revive it and does feed the self-comment ratio problem.

The version of this that fails quietly is worth stating once more, because it is the version most accounts are running. Publish, immediately drop a comment with a link in it, post two more comments when nothing happens, then check back in the evening to reply to whatever arrived. That sequence collides with the bridge behavior penalty, the velocity classifier, the self-comment ratio signal, and the closed distribution window, in that order, over about eight hours. Every individual step came from a guide that used to be right. Reversing it is not complicated: one substantive link-free comment a couple of minutes in, then an hour of replying to people, then leave it. That is the whole discipline, and on the accounts we manage it is the difference between a post that stalls in Stage 1 and a post that gets to find its audience.

Frequently asked questions

Does putting a link in the first comment still boost LinkedIn reach in 2026?

No. LinkedIn's March 2026 Authenticity Update closed this workaround. The algorithm now detects 'bridge behavior' (a linkless post that funnels viewers to a comment with a link) and applies the same roughly 60% reach suppression as a link in the post body itself. SocialNexis accounts that continued the tactic after the patch saw median impressions fall 55-65% on identical content compared to their pre-patch baseline.

What is LinkedIn's bridge behavior penalty and does it apply to links in the first comment?

Bridge behavior is LinkedIn's term for the pattern where a post omits an external link in the body but directs viewers to find one in the comments. As of March 2026, LinkedIn explicitly detects and penalizes this pattern. Richard van der Blom's 2026 Algorithm Report found that comments with external links see up to 80% visibility reduction. Yes, the penalty applies directly to links placed in the first comment.

Does the LinkedIn algorithm treat the post author's first comment differently from other commenters?

Yes, in two ways. The author's first comment is weighted more positively as an engagement seed. It is also penalized more heavily when it contains a link. Kristen Sesto's controlled test found an author's first-comment link caused a 15% reach drop versus a 5% drop for the same link posted by a third-party commenter. The author's comment also draws more scrutiny for authenticity signals such as posting speed and session origin.

How long after posting should you add a first comment on LinkedIn for maximum reach?

Two to five minutes after publishing is the optimal window. The 0-5 minute range outperforms the 5-15 minute range by about 18% on Stage 2 advancement rate. However, comments posted within seconds of publication trigger LinkedIn's engagement-velocity classifier, which flags them as inauthentic and silently suppresses the post to 1-2% of normal initial distribution. Posting manually at two to five minutes avoids both failure modes.

Do comments in the first hour increase LinkedIn post reach, and by how much?

Significantly. Comments posted within the first 30 minutes generate 3.8x more impressions on average than comments posted after 24 hours (391 versus 104 average impressions), based on analysis of 261,137 real LinkedIn comments. Posts where the author responds to comments within the first 30 minutes receive 64% more total comments and 2.3x more views. The first hour sets most of a post's distribution trajectory.

Should you write your own first comment immediately after publishing a LinkedIn post?

No, not immediately. A first comment posted within seconds of publishing creates a timing pattern that LinkedIn's authenticity systems detect as automated behavior. The post may be silently capped at 1-2% of its normal initial distribution with no visible warning. Post your first comment two to five minutes after publishing. At that point, a substantive, link-free comment from the author reads as a positive engagement signal, not a flag.

What should you write in the first comment on a LinkedIn post to boost engagement?

A TL;DR that condenses the core point for readers in a hurry, a follow-up question that invites specific replies, or a bonus data point that was too long for the post body. Keep it 15 words or more and avoid external links and self-promotional calls to action. Substantive comments carry 3-5x more algorithmic weight than generic comments under 10 words, so content quality directly shapes reach.

Does a first comment with a link hurt reach more than a first comment without a link?

Yes, substantially. A link-free first comment from the author is treated as a positive engagement signal. A first-comment link from the author triggers the bridge behavior penalty (up to 60% reach suppression) and draws a heavier per-author penalty than the same link in a third-party comment. Van der Blom's 2026 data puts comment-link visibility reduction at up to 80%, making the two scenarios entirely different in reach outcome.

How does LinkedIn detect and penalize engagement pods and coordinated first-comment strategies?

LinkedIn's AI detects engagement pod activity with 97% accuracy as of 2026. Groups of 15 or more users commenting within 90 seconds using short or generic phrases are identified and suppressed. The detection weighs velocity, commenter network proximity, comment language patterns, and IP clustering. Coordinated first-comment strategies are counterproductive: they suppress the post they are meant to help, with no mechanism for the account to know it happened.

Sources and further reading

• LinkedIn Engineering on feed dwell time and passive engagement scoring
• LinkedIn Help Center on comment visibility limits for automated or excessive activity
• Richard van der Blom's LinkedIn Algorithm Insights Report (1.8 million posts analyzed)