How many views is viral, in numbers
The short answer, measured rather than guessed: on X, a launch post that passes 1.8M views is in the top quarter of every AI launch we track, and the median launch does about 690K. Every figure below comes from the launch directory, which records the views, saves and reach of each launch we can find, and is rebuilt daily.
The word gets used without a threshold behind it, so every launch report can claim it. We measure every AI launch on X we can find and publish each one, which makes the threshold checkable rather than rhetorical.
Here is the whole corpus, sorted by how far each launch got. Read it as the odds on any given launch day.
| Views on the launch post | Share that get there |
|---|---|
| 100K | 80% |
| 250K | 69% |
| 500K | 59% |
| 1M | 40% |
| 2M | 23% |
| 5M | 10% |
| 10M | 5% |
A million views is the number founders mean when they say viral. In the directory it now sits near the middle: the median launch does about 690K, and 1.8M puts a launch in the top quarter.
4.8M puts it in the top tenth. The tier above that is rare enough that most companies will never have one, and planning on it is planning on weather.
Why a few launches take most of the views
That split is the single most useful thing in the dataset. Launch outcomes are not distributed around an average with a bit of spread either side. A small number of results carry almost everything, which is why an agency quoting you a mean is quoting a number that describes almost none of its work.
The practical consequence is about where effort goes. Adding a few percent to a launch that was going to do 200,000 views changes very little.
Getting a launch across the line where other accounts start repeating it changes the outcome by an order of magnitude, because reach on X comes from who repeats a post rather than who receives it first. Everything below is about that line.
Your follower count does not decide your reach
The most common reason a company delays a launch is a belief that it needs an audience first. The data says the audience is close to irrelevant on the day.
| Measure | Across every launch measured |
|---|---|
| Median followers on the posting account | 17,133 |
| Median reach against that following | 26 times |
| Launches posted from under 10,000 followers | 843 |
| Of those, share that passed a million views | 23% |
The median launch reaches many times the following of the account that posted it, which means almost all of that reach arrived from somewhere other than the followers. Roughly one in four launches posted from an account with under ten thousand followers still passed a million views.
Two of ours sit at the extreme of that. Visko introduced Orbis 1.0 from an account with 20 followers and reached 1,432,752 people, then grew ninety times over in 24 hours.
That case study is here. Wonderful announced a $550M round at a $5 billion valuation from an account that started the week near a thousand followers, and the post reached 1,290,522.
That one is here. A company can be worth five billion dollars and still be speaking into a small room, because valuation is built by selling to a few buyers and an audience is built by posting.
Why X and not somewhere else
Two things make X the venue for launches in this category, and neither is about audience size.
The first is who is there. The people who evaluate and buy technical products, founders, engineers, investors and operators, still conduct their professional reading on X.
They are also the population most likely to be running an ad blocker, which removes the obvious way of reaching them and leaves exactly one route: the accounts they already chose to follow.
The second is that X still moves a post between audiences. A post that performs can be shown to people with no connection to the author, which is the mechanism a launch depends on. Platforms that rank almost entirely on existing graph connections cannot produce that jump regardless of budget.
What the ranker is optimising for
Since 2026 the For You feed has been ranked by Phoenix, a Grok-based transformer that replaced the older multi-service stack. It narrows roughly 500 million posts a day to about 1,500 candidates per request, then scores each candidate by predicting how a given user will respond to it.
Those predictions cover both directions. Some heads predict positive actions, replies, reposts, profile visits, dwell.
Others predict negative ones, including hiding the post or reporting it. A post that draws strong reactions of both kinds does not simply net out; the negative predictions carry real weight.
Three consequences matter when planning a launch:
- Replies outrank likes. A post making a specific, disputable claim generates conversation. An agreeable post collects likes and stops moving.
- Dwell time is predicted directly. A dense post someone reads through beats a one-liner they scroll past, which is why launch posts have got longer rather than shorter.
- Your own baseline is the comparison. Accounts are measured partly against their own recent performance, so a quiet month lowers the bar everything afterwards is judged against.
What is worth stating plainly: the numeric weights are not public. The parameters module was never shipped with the open-sourced code, so any exact multiplier being quoted is a leak or an invention. The full breakdown is here.
The window
The single largest controllable variable in an X launch is how tightly the posts are grouped in time.
Posts landing inside the same few hours compound, because engagement on the early ones raises the signal the later ones are measured against, and because a reader seeing the same product from three different trusted accounts in one session reads that as consensus rather than advertising. The same posts spread across a week are unrelated posts and perform like unrelated posts.
A creator posting four hours late has not delivered slightly less value. They have delivered outside the window that made the campaign work.
The second surface
A launch post is the opening of a conversation. The replies are where it either becomes credible or falls apart, and they are ranked highly enough to carry their own distribution.
This is where founders matter more than budget. An engineer asking a sharp question in the replies is the highest-intent reader the launch will get, and a founder answering them well converts more than the original post did. Launches where the team disappears after posting leave most of the available value unclaimed.
What viral means when you sell to companies
Consumer launches and B2B launches are graded on different things, and using one scoreboard for both is how enterprise companies end up buying reach that does nothing.
For a consumer product, the number of people who saw it is close to the whole answer. For a product sold to a committee, the buying population is small and specific, and the question becomes how many of the right people saw it and how many were willing to put their name to it in public.
That second part is measurable. Reposts per million views is the cleanest version of it, because a repost costs the sender something a like does not.
It also produced roughly four times the quote posts and shares per view of a comparable high-reach consumer drop. It reached a fraction of the audience a consumer launch does, which was the trade it was built to make. The full numbers are here.
Bolt is the other half of the argument. Security releases are the category everyone assumes cannot travel, since they are defensive, have no demo and are usually written for auditors.
Bolt's agentic security release did 1M views, because it described a role rather than a process and named the threat in the reader's own terms. That case study is here.
The honest limit on all of this: reach and endorsement are what a launch report can show. What happened in the funnel afterwards is the client's data, and most agencies never ask for it. We wrote about why that half goes missing.
The launch where we saw both halves
Codos launched the first virtual Chief AI Officer from co-founder Dima Khanarin's account, which had about 1,800 followers. The post reached 1.6M people, around 900 times its own following, which would normally be where the report ends.
Underneath it: 2,031 bookmarks, and a founder account that went from 1,800 followers to 3,983.
Two weeks after the post, the founder's account of the inbound was many VCs and many, many customers. We are overwhelmed. The full case study is here.
An enterprise product sold to CEOs, and the thing that filled the inbox was people posting about it. That is the argument for treating distribution as the first decision rather than the last one.
What repetition does that one launch cannot
A single viral launch is an event. Companies that ship every two or three weeks get something else, because each release lands on an audience the last one warmed.
Higgsfield went from launch to a $5 billion valuation in about fifteen months and ran every product release in that window through our network. Between March and June 2026 that produced 40M views and 1.8M engagements across six releases.
The interesting part is the floor rather than the 14.7M-view outlier: four further releases, a prediction tool, an agent, an MCP connector and a set of editor plugins, each landed between 1.3M and 1.9M views. The drop-by-drop numbers are here.
Against the distribution at the top of this page, a 1.3M floor is a top-quarter result treated as a bad month. That is what cadence buys, and no single launch can produce it.
What kills a launch on X
- Identical copy across accounts. Visible instantly, and it collapses the credibility of every post in the set at once.
- Audience overlap. Thirty creators drawing from one pool reach fewer distinct people than ten who do not.
- Launching into a news event. Attention is finite, and a funding round announced during a larger story is a wasted round.
- No artefact. Text-only launches in a category where everyone ships a demo lose before they start.
- Undisclosed paid posts. A liability that lands on the client, and the FTC's rules are not ambiguous about it.
The step-by-step version is here.
Questions we get asked
How many views is considered viral?
There is no platform-wide number, which is why every agency can claim it. Measured on X across every AI launch we track, 1.8M views puts a launch in the top quarter and 4.8M in the top tenth, while the median launch does about 690K. Treat a million as the working threshold for a tech launch on X and anything above five million as rare enough not to plan for.
Is viral the same number on TikTok, YouTube and Instagram?
No. The thresholds people quote for those platforms are guesses that vary by who is quoting them, and our data covers X only. What transfers is the shape: on every feed a small share of posts take most of the views, so the useful question is where a post sits in the distribution for its platform, not whether it crossed a round number.
Why do product launches work better on X than other platforms?
The buyers for technical products still read professionally on X, and they are the population most likely to block ads, which leaves the accounts they already follow as the only route to them. X also still moves a post beyond the author's existing followers, which is the mechanism a launch depends on.
How does the X algorithm decide what spreads in 2026?
Phoenix, a Grok-based transformer, narrows roughly 500 million posts a day to about 1,500 candidates per request, then predicts how each user will respond. It predicts negative reactions such as hides and reports alongside positive ones, so a divisive post does not simply net out. The numeric weights are not public.
How tightly should launch posts be timed?
Inside the same few hours. Engagement on early posts raises the signal later ones are measured against, and a reader seeing a product from several trusted accounts in one session reads it as consensus. The same posts spread over a week behave like unrelated posts.
Does replying to comments actually affect reach on X?
Replies are weighted more heavily than likes and carry their own distribution, so a thread where the author engages keeps moving longer than one where they post and leave. Replies are also where the highest-intent readers appear, which makes them the most valuable part of a launch to answer well.
What counts as a viral launch on X?
1.8M views, which puts a launch in the top quarter of every AI launch we have measured. The median launch does far less than that, and the rungs above it thin out fast: roughly one in four passes two million and 5% passes ten million. The threshold is published rather than asserted, because we measure every launch we can find and show the numbers on each one.
Do you need a lot of followers to go viral on X?
No. The median launch we measure reaches many times the following of the account that posted it, so almost all of that reach came from somewhere other than the followers.
Roughly one in four launches posted from an account with under ten thousand followers still passed a million views. One of ours reached 1,432,752 people from an account with 20.
How is a B2B launch measured differently from a consumer one?
By who saw it and who was willing to put their name to it, rather than by raw reach. Reposts per million views is the cleanest measure of the second part, because a repost costs the sender something a like does not. One enterprise launch we ran produced around four times the quote posts and shares per view of a comparable high-reach consumer drop, on a fraction of the audience, which was the trade it was built to make.
Can an agency prove a viral launch drove revenue?
No, and be careful with anyone who says otherwise. What can be shown is the pre-launch baseline, the change against it, and what buyers said when asked how they heard about you. On one Series A launch, demo requests moved from a single-digit daily baseline to roughly seventy on the day and X has been the largest named source of booked demos since.