1. A tenth of launches take most of the views
The top 10% of launches took 71% of every view measured. The top 1% took 31%.
The median launch did 261,897 views and the mean did 1,509,685, a gap of 5.8 times, because a handful of results carry the average. Launch outcomes follow a power law rather than a bell curve, and any agency quoting you an average is quoting a number that describes almost none of its work.
The practical reading is about where effort goes. Adding a few percent to a launch that would have done a median result changes very little. Getting one launch into the top tenth changes everything, which is why the package and the first hour matter more than polish on the asset.
Cite as: Doomers, The State of AI Launches on X, Q3 2026, finding 1. https://doomers.ai/research/state-of-ai-launches-2026#concentration
2. Accounts over a million followers underperform
Launches from accounts with over 1M followers did a median of 102,955 views, below accounts with 1K to 10K followers at 258,841.
Reach rises with account size up to the 100K to 1M band, where the median launch did 1,213,812 views, and then falls.
The largest accounts in the directory are mostly established companies posting routine updates to an audience that has stopped reacting, and the ranker grades a post on early engagement against that account's own baseline. A big following raises the baseline.
The share of launches passing a million views tells the same story: 56% for the 100K to 1M band, 23% for accounts over a million. The over-1M band is 35 launches, small enough that a few outliers move it, so treat the direction as the finding rather than the exact figure.
Cite as: Doomers, The State of AI Launches on X, Q3 2026, finding 2. https://doomers.ai/research/state-of-ai-launches-2026#account-size
3. Small accounts still break out
17% of launches from accounts under 10,000 followers passed one million views: 59 of 347.
A launch earns almost all of its reach from people who did not follow the account that morning. The ranker tests a post on a small audience and widens distribution when the early response is strong, and it does that for a 900-follower account the same way it does for a 90,000-follower one.
Follower count on the day is a weak predictor of where a launch ends up. What it predicts is the size of the first test audience, which is the one thing an amplified first hour can substitute for.
Cite as: Doomers, The State of AI Launches on X, Q3 2026, finding 3. https://doomers.ai/research/state-of-ai-launches-2026#small-accounts
4. Funding announcements outperform product launches
Funding announcements did a median of 353,695 views against 242,227 for product launches, 1.5 times as many.
A round is news that other people carry. Investors quote it, the press writes it up, peers congratulate it, and every one of those is a post that did not have to come from the company.
A product launch is carried by whoever the company can get to post about it in the first hour, which is usually nobody.
This is a reason to treat a funding round as a launch and plan it like one, and a reason to build the first hour deliberately for a product release, because the round comes with a crowd and the release does not.
Cite as: Doomers, The State of AI Launches on X, Q3 2026, finding 4. https://doomers.ai/research/state-of-ai-launches-2026#funding-vs-product
5. Long video beats short
Launch videos over two minutes did a median of 405,295 views, 2 times the median for videos of 30 seconds or less, which did 200,853.
The advice to keep launch video short is everywhere and the data runs the other way. A ranker that optimises for watch time rewards a video that holds attention for minutes over one that is over in seconds, and a demo that shows the product doing the whole job holds attention better than a teaser.
What short video is good for is a single checkable claim, which is a different launch and a different post. The launches that led with one number and a 30-second clip sit in this data too, and they are the exception the median hides.
Cite as: Doomers, The State of AI Launches on X, Q3 2026, finding 5. https://doomers.ai/research/state-of-ai-launches-2026#video-length
6. A Wednesday launch does 3 times a Friday one
Launches posted on a Wednesday did a median of 344,653 views. Friday launches did 111,674, and weekend launches under 40,000.
The best three weekdays were Wednesday, Tuesday and Monday, with Wednesday on top. Friday was the worst weekday at less than a third of Wednesday, and the weekend sits at a small fraction, on a sample of 23 launches, which is small because almost nobody launches then.
The day matters less than the first hour, and a Wednesday launch with nobody lined up will still stall. What the day decides is how many of the right people are on the feed when the post goes out.
Cite as: Doomers, The State of AI Launches on X, Q3 2026, finding 6. https://doomers.ai/research/state-of-ai-launches-2026#day-of-week
7. Repeat launchers earn more per launch, and later launches do no better than first ones
Companies that launched two or more times did a median of 567,274 views per launch, against 231,359 for one-time launchers. Their later launches did no better than their first: 567,274 against 568,056.
42 companies launched more than once, 99 launches between them and 36% of every view in the directory, at a median of 25 days between launches. Their biggest launch came first for 21 of the 42, a coin flip.
Part of the per-launch difference is who they are: repeat launchers post from larger accounts, a median of 15,174 followers against 4,945.
The rest is that each launch is a fresh draw from a very wide distribution, and companies that take more draws hold more of the top. Cadence multiplies the number of launches. It does not compound any one of them.
Cite as: Doomers, The State of AI Launches on X, Q3 2026, finding 7. https://doomers.ai/research/state-of-ai-launches-2026#cadence
8. Y Combinator batch launches sit well under the median
75 launches from accounts stating a YC batch did a median of 59,937 views, against 326,931 for everything else, from accounts with a median of 1,062 followers.
This is an account-size effect before it is anything else. A batch company launches from an account opened for the batch, often the founder's, with a few thousand followers or fewer, and the distribution of its launches looks like every other small account's: a few travel far and most do not.
What the batch does not do is change the first hour. A demo day, a batch hashtag and a crowd of peers posting the same week put a launch in front of other founders, who are not usually the buyer.
Cite as: Doomers, The State of AI Launches on X, Q3 2026, finding 8. https://doomers.ai/research/state-of-ai-launches-2026#yc
9. What a launch earns beyond views
The median launch earned 12.2 saves per 10,000 views and 251 reposts per million views.
Views are the number the ranker controls. Saves and reposts are the numbers people control, and they are the ones to grade a launch on when the product has a considered purchase behind it.
A launch that reaches seven times the median and saves at a third of the median rate reached a crowd that scrolled past.
These two baselines are the ones we use to read every launch in the directory, and they are published here so anyone can read their own launch against them.
Cite as: Doomers, The State of AI Launches on X, Q3 2026, finding 9. https://doomers.ai/research/state-of-ai-launches-2026#engagement
Methodology
The Doomers launch directory collects public posts on X in which a company or founder announces a product, a release, a feature or a funding round, and reads each post's views, likes, reposts, replies, bookmarks and the posting account's follower count from X.
Each launch is published as its own page with those numbers. This edition freezes the directory as it stood on 14 September 2026: 592 launches from 535 companies, posted between March 2026 and September 2026.
Medians are used throughout because the distribution is heavy-tailed and a mean describes almost no individual launch. "Funding announcement" means the post itself states a round or an amount raised. Video length is X's own duration for the attached media.
"YC batch" means the post or the account's bio states a batch, and alumni mentions do not count. Company grouping follows the posting account, with a small hand-maintained list mapping founder accounts to their companies.
Limits worth knowing. The directory is collected by hand and by search and does not claim to be every AI launch on X. Engagement is read at collection and again when a page is rebuilt, so recent launches carry fewer days of accumulation than older ones.
The over-1M and weekend bands are small. Nothing here is a controlled experiment; every finding is an observed pattern with the confound named where we can see one.
The directory since this edition
The findings above are frozen at 14 September 2026 and their anchors will not move. The directory kept growing. On 7 October 2026 it holds 1,636 launches from 1,125 companies, posted between 7 January and 6 October 2026, read from X every day and every 30 minutes while a launch is moving.
The mix changed with it. In early October we imported the launches from two other directories that the harvest had missed, keeping only posts above 500,000 views, so the directory now leans towards launches that were already noticed.
The median launch in it today is 806,048 views, against 261,897 in this edition, and the top tenth takes 66% of all views against 71%. Those two numbers are not comparable, and the next edition will be cut on a cohort collected the same way as this one.
Three launches since this edition belong in the record. TypeSafe AI, out of stealth from a 1,500-follower account on 15 September, stands at 40M views, bigger than every OpenAI launch of the year except GPT-6 Astra.
Tavus Griffin on 1 October stands at 21.1M from 11,500 followers. Mistral Large 4 on 6 October passed 5.7M in its first day.
How to cite
Doomers Media, The State of AI Launches on X, Q3 2026, published 14 September 2026. https://doomers.ai/research/state-of-ai-launches-2026
Each finding has a stable anchor, for example https://doomers.ai/research/state-of-ai-launches-2026#concentration. The figures on this page will not change. The directory the report is drawn from updates daily at doomers.ai/launches, and the next edition will be cut at the end of the quarter.
The directory is published as a dataset under CC BY 4.0: use it freely with attribution to Doomers. Download it as CSV or as JSON; both files rebuild with the directory.
Questions we get asked
How many AI launches does this report cover?
592 launches from 535 companies, posted on X between March 2026 and September 2026, each with its views, saves, reposts and the posting account's follower count read from X.
How big is the directory now?
1,636 launches from 1,125 companies as of 7 October 2026, posted between January and October 2026. This report is frozen at 592; the directory at doomers.ai/launches carries the current set and every launch page updates from X daily.
What counts as a launch in this data?
A public post on X from a company or founder account announcing a product, a release, a feature or a funding round, with a video or image attached. Reposts, replies and posts from people who do not work at the company are excluded. Every launch in the report has its own page in the Doomers launch directory.
Is the data available?
Every launch is published with its numbers at doomers.ai/launches, and each finding above has a stable anchor for citation. The figures on this page are frozen as the Q3 2026 edition and will not change; the directory itself updates daily.
Why do accounts with over a million followers do worse?
The ranker grades early engagement against the account's own baseline, and a very large following raises the baseline. The large accounts in the directory are mostly established companies posting routine updates to an audience that has stopped reacting. The band is also small, so read the direction rather than the exact number.