Why your AI content gets 300 views (and what actually fixes it)

You put AI on your content, posted for weeks, and every video lands near a few hundred views. It is rarely the AI doing that. Here is where the reach actually leaks, with numbers from my own accounts, and a one-week plan to get one video past the floor.

Published August 15, 2026 · By Dmitry Tolstoi · 9 min read

You know the number. You publish, refresh, and the counter crawls to two hundred and something, then stops. Do it ten times and it feels like a verdict on the AI you used to make the thing.

I run a content operation as one person with AI on most of the production, so this is no anti-AI speech. The tools work. But I watched my own produced clips die at a few hundred views while a plain phone video of me at my desk pulled tens of thousands.

My receipts first. A run of produced English clips on one account landed at 144, 176, 231, 315 and 316 views. Live footage of me doing real work, same account, pulled 16,000 and 17,900. Then one produced clip broke to 10,100, and it still used an AI voice clone. The gap was never the AI. It was five other things, and every one is fixable.

Why does every video stall near the same low number?

In short: reach is earned from how your first viewers react, not handed out by your follower count, so a weak early signal caps a video before most people see it.

The floor is not a quota anyone set. It is the point where a video stops earning its way to new people, and follower count barely helps you across it. TikTok says plainly that "neither follower count nor whether the account has had previous high-performing videos are direct factors in the recommendation system", and most of the people a clip reaches are strangers anyway: Adam Mosseri, the head of Instagram, notes that on Reels "much like Explore, the majority of what you see is from accounts you don't follow".

So the number comes down to what those first strangers do. TikTok says a strong interest signal, "such as whether a user finishes watching a longer video from beginning to end, would receive greater weight". Mosseri lists the same levers for Reels: the top predictions are "how likely you are to reshare a reel, watch a reel all the way through, like it, and go to the audio page". YouTube says that for short videos, relative watch time matters more than the raw seconds. Finish rate and shares are the whole game, and none of these platforms lists "made with AI" in it.

The algorithm decides in the first two seconds

In short: the early audience only expands your reach if people stop scrolling, so a dead first frame caps a video before the content gets a chance.

I lined up my worst clips against my best one, frame by frame, and the difference showed in the opening half second, before a word landed. The produced clips opened on a person at a flat grey wall, talking. My 17,900-view clip opened on me busy with a real task in a real room.

Three things killed the produced openings, all of them visual. The frame was empty, so a scrolling viewer learned nothing about where they were. The frame was static, so the thumb had no reason to stop. And the hook was spoken, not shown, but feeds play muted by default, so the picture has to earn the stop alone. TikTok's own ad research puts 90 percent of ad-recall impact inside the first six seconds. I made it worse for myself: I shot the base footage to the flat, head-and-shoulders spec a lip-sync model wants, which is the exact recipe for a dead frame in a feed.

It is usually the audience, not the AI

In short: the same clip shown to the wrong audience gets buried, and a buried clip lands near the floor no matter how good the production was.

Here is the measurement that reframed it for me. Every one of those 144-to-316-view clips lived on an account whose real audience is personal and speaks a different language than the videos did. English content about building software with AI was never aimed at those followers. They scrolled past without finishing, the system read a weak signal, and it showed the next clip to fewer people still. That is how you get a few hundred views. I had not measured that AI is worse than live footage; I had measured a content-to-audience mismatch.

This is the cheapest fix on the list and the one almost nobody makes. A clip for solo founders belongs on an account that solo founders follow, not on a feed built for family and friends. Mixing two audiences on one account hurts both, because a run of clips that get scrolled past drags down how the system treats everything else you post there, including the one video that would have flown.

Do the platforms actually punish AI content?

In short: the crackdowns target mass-produced and unoriginal uploads whoever made them, not AI itself, so the fix is to be original, not to hide the tools.

The fear underneath a low number is that the platform can smell the AI and buries you for it. The written policies say otherwise. Google's spam rules call out "scaled content abuse," pages "generated for the primary purpose of manipulating search rankings and not helping users", and Google draws the line on purpose, not method: the target is producing content at scale to game ranking, "whether automation, humans or a combination are involved".

Video platforms land in the same place. In July 2025 YouTube renamed its "repetitious content" rule to "inauthentic content", and coverage stressed it was a minor clarification of a long-standing policy, not a new ban on AI. TikTok and Meta went the other way, choosing to label AI content rather than remove it: TikTok auto-tags it through Content Credentials and Meta adds "AI info" labels across video, audio and images. What distribution rewards is originality. In 2024 Instagram said accounts that repeatedly repost others' work they "didn't create or enhance in a material way" would stop being recommended, and Mosseri put it straight: "if you create something from scratch, you should get more credit than if you are resharing something that you found from someone else." What the platforms bury is one more copy of what they have already shown a thousand times, machine-made or not.

The clip that finally broke out still used AI

In short: my one video that jumped to 10,100 views kept its AI voice, so what changed was making something real enough to finish, not removing the machine.

The most useful clip I own is the one that jumped to 10,100 views after a long run stuck between 144 and 316 on the same account. It did not win by dropping the AI. It kept a cloned voice on top and word-level captions underneath. What changed was everything around the AI: I shot the footage on a camera instead of generating it, the story was a real process with its real snags, and it ran long enough to actually tell that story.

Sameness is what sinks AI content, and sameness is a choice, not a property of the tool. The failure mode has a name now: Merriam-Webster made "slop," low-quality content produced in quantity by AI, its 2025 word of the year, after developer Simon Willison predicted it would stick "the way that 'spam' became the term for unwanted emails". A feed learns to skip the tenth identical template. I wrote about the visible version in the piece on why $1 AI websites look generated, and the fix is the same: put something specific where the template put something generic.

Why don't the views turn into customers?

In short: a view with no path to your product moves nothing, so a small clip that sends ten people to what you sell beats a viral one that sends none.

Even after you beat the floor, a view count is vanity until it connects to something you sell. I have two clips on one account whose view counts sit orders of magnitude apart, and neither moved my product a step, because neither gave a viewer a road from the video to the thing I make. Views are stars on a repo. A product needs a path, and building the path from content to product is its own system, work the AI did not do for you.

The missing piece is usually the same: no clear reason and no route to a next step. I have posted clips with no ask and seen only a couple of real comments, while the creators who pull comments by the thousand almost always asked for one outright. A thread like that is a request, not luck.

Check what the platform actually published

In short: the tool that posts your video can quietly crop or mute it, so a clip can run to zero with nobody ever seeing the call to act.

One trap sits past everything else you fixed: the platform does not always publish what you sent. A posting tool once auto-cut one of my videos from 103 seconds to 60, chopping off the ending, the one line telling viewers where to go. It ran zero views there, and nobody saw the ask, because nothing in my own files flagged it when the master was fine. Open the live post on every platform and read what actually shipped.

A one-week plan to get one video past the floor

In short: match the account to the audience, reshoot one opening as a real event, keep the AI but kill the sameness, give it one ask and one path, then verify what published.

None of these fixes asks you to abandon AI. This is the plan I would run to drag one stalled clip off the floor; it fits a week of evenings, and one step leans on the playbook on being the check on AI output.

1. Day one, pick the audience. Name the one audience the video is for and the account they actually follow. If the clip was dying in front of the wrong crowd, that was your whole problem.
2. Day two, fix the first two seconds. Reshoot the opening as a real event with motion, and put the hook on screen as text. Judge it muted, at arm's length.
3. Day three, cut the sameness. Keep the AI voice and captions, but replace any default-template look with something that is yours: your desk, your process, your footage.
4. Day four, add one path. Write one clear ask and one route to your product, inside the video where it plays. One ask, not three.
5. Day five, post and check the cut. Publish on the matched account, then open the live post on every platform and confirm nothing was cropped, muted or trimmed.
6. Day six and seven, read the curve. See where viewers drop in the first three seconds and whether anyone took the path. Fix the opening or the ask, then post the next one.

Where this comes from

Every view count above is from my own accounts, and I say so where the example is mine. The clips that stalled and the one that broke out ran through the same pipeline, one person with AI on operations, changed in the ways described here, so read the numbers as a before and after.