197,558 of the 419,998 app videos we classified carry no call to action
That is 47.0% of the posts our classifier read, not 47.0% of app videos. A test we ran on our own pipeline says the real share is lower.
All-time classified organic, plus a 30-day window to 14 August 2026
- 419,998
- Videos we classified
- 47.0%
- Carry no call to action
- 9.5%
- Ask for the download
419,998 classified videos · Snapshot of 14 August 2026
197,558 of the 419,998 app videos our classifier has read carry no call to action. That is 47.0% of them.
The denominator moves the claim more than the share does. 419,998 is the count of videos in our library carrying an AI classification, out of 565,898 videos in total. The other 145,900 have never been read. Measured against the whole library the no-CTA count is 34.9%, and measured against app video in general it is a number we do not have.
Among the 222,440 posts that do carry an ask, the most common one is not the install. It is the profile. 87,004 posts point at a link in bio against 40,103 that ask for the download, which is 2.2 to 1.
Every classified organic post in the library, by the one call-to-action value the classifier assigned it. The seven values are a closed vocabulary and each post gets exactly one, so the column sums to the denominator.
| Call to action | Videos | Share of 419,998 |
|---|---|---|
| None found | 197,558 | 47.0% |
| Link in bio | 87,004 | 20.7% |
| Follow the creator | 46,269 | 11.0% |
| Visit a website | 43,831 | 10.4% |
| Download the app | 40,103 | 9.5% |
| An offer or discount | 5,163 | 1.2% |
| Other | 70 | 0.02% |
Other is a residual bin and not a destination. It holds 70 posts, 0.02% of the column, and is left in so the figures sum.
419,998 classified videos · Snapshot of 14 August 2026
- 565,898
- Organic videos in the library
- 419,998
- Carrying a classification
- 145,900
- Never classified
565,898 organic videos · Snapshot of 14 August 2026
What none means here, and what it does not mean
The classifier reads a post's caption, plus its spoken transcript where we resolved one, and picks one of seven values. None means it found no ask in the text it was given. It does not mean the post contained no ask. The rest of this page measures how far apart those two things are.
Link in bio outnumbers download the app by 2.2 to 1
The 222,440 classified posts that carry an ask, split by what they ask for.
Only the posts carrying an ask. The 70 posts in the residual other bin sit inside the 222,440 denominator. They are left off the bars, where they would draw a line one pixel high.
The figures behind this chart
| Link in bio | 87,004 |
| Follow the creator | 46,269 |
| Visit a website | 43,831 |
| Download the app | 40,103 |
| An offer or discount | 5,163 |
222,440 classified videos carrying a call to action · Snapshot of 14 August 2026
Creators promoting apps mostly send people to a profile. 87,004 posts point at a link in bio and 46,269 ask for a follow. That is 133,273 of the 222,440 posts with an ask, 59.9%, routing to the creator and not to a store.
40,103 posts ask for the download outright. That is 18.0% of the posts carrying an ask and 9.5% of everything we classified.
A further 43,831, 19.7%, send people to a website. We do not resolve where those links land, so a website ask is not evidence of a store visit in either direction.
The 47.0% is not an Instagram artifact. We had the direction backwards
Exact counts per platform over one uniform 30-day window, with no cell near the count cap.
Here is the hypothesis we went in with. On TikTok we read the caption and the spoken transcript. On Instagram we read the caption alone. A creator who says the ask out loud in a Reel is invisible to us and scores as no call to action. So we expected the caption-only platform to under-report calls to action, and to push the pooled figure up.
We ran the count. Instagram reports 39.0% no call to action against TikTok's 59.3%, over the same 30 days. The caption-only platform reports more calls to action, not fewer, and by 20 points.
The mechanism is ordinary once the result is in front of you. An Instagram caption is typed, and the words link in bio are typed in it. A TikTok transcript is speech, and a creator who taps the screen while it reads link in bio says nothing our classifier can read.
Classified posts published in the 30 days to 14 August 2026, counted per platform. Each cell is an exact count from the live index, not a sample.
| Call to action | TikTok | Share | Share | |
|---|---|---|---|---|
| None found | 9,749 | 59.3% | 2,547 | 39.0% |
| Link in bio | 2,868 | 17.5% | 1,743 | 26.7% |
| Follow the creator | 1,692 | 10.3% | 645 | 9.9% |
| Visit a website | 1,211 | 7.4% | 994 | 15.2% |
| Download the app | 661 | 4.0% | 487 | 7.5% |
| An offer or discount | 250 | 1.5% | 112 | 1.7% |
| Other | 1 | 0.0% | 1 | 0.0% |
| All classified posts | 16,432 | 100% | 6,529 | 100% |
The window is 30 days because 60 and 90 days both pushed the largest cell past the 10,000 count cap. Past that cap the counter stops reporting a number. At 30 days no cell is capped, so every figure in this table is exact.
22,961 classified posts · 30 days to 14 August 2026
Two things follow, and the second is the one a sceptic will reach for.
The pooled all-time 47.0% is not an Instagram artifact. Instagram pulls it down, and the pooled figure sits between the two platform figures, which is what a genuine mix looks like.
The two measurements also do not reconcile arithmetically, and we are not going to pretend they do. Over the same 30 days the two platforms pool to 53.6%, 12,296 of 22,961 classified posts, above the all-time 47.0%. The all-time counter and the windowed counter run over different populations. The platform mix of everything we have ever classified is not the platform mix of the last month.
The test found a limit we were not looking for
The same posts split by whether a transcript resolved, inside each platform. 1,245 classified rows out of a 2,796-row sample.
Platform was the wrong cut. The cut that moves the number is whether we resolved a transcript at all.
Inside TikTok, posts with a resolved transcript report 53.4% no call to action and posts without one report 68.3%. Inside Instagram the same split runs 21.6% against 42.4%. When we can hear the video, we find more asks: 14.9 points more on TikTok and 20.8 more on Instagram.
Pooled across both platforms the effect disappears. 49.9% with a transcript against 50.5% without is a gap of 0.6 points, pointing the other way. The pooled comparison hides the effect because the two platforms resolve transcripts at opposite rates, and the within-platform split is the measurement.
Classified rows from the sample, split by whether a transcript resolved for the post. The gap column is the no-CTA share without a transcript minus the share with one.
| Platform | Rows with transcript | No CTA | Rows without | No CTA | Gap |
|---|---|---|---|---|---|
| TikTok | 416 | 53.4% | 243 | 68.3% | 14.9 points |
| 51 | 21.6% | 535 | 42.4% | 20.8 points | |
| Both pooled | 467 | 49.9% | 778 | 50.5% | 0.6 points the other way |
The Instagram cell of 51 rows is small, and it is also one mechanism. All 51 came through a single rescue path, so it is not a random 51 Reels. The 416 TikTok rows split into 356 machine transcripts and 60 creator-supplied captions. Read the Instagram gap as directional and the TikTok gap as the measured one.
1,245 classified rows in a 2,796-row sample · Published 10 July to 14 August 2026
The honest ceiling sits below the headline
Two sentences are both needed and neither works alone. Of the posts our classifier read in the last 30 days, 59.3% of TikToks carried no call to action, against 39.0% of Instagram Reels. The rows counted are 16,432 on TikTok and 6,529 on Instagram. That number is partly a measurement limit. Posts where we resolved a transcript report a no-CTA share 15 to 21 points lower than posts where we did not, within the same platform. Nothing in the pipeline reads on-screen text either. A share of the no-CTA bucket is our classifier not hearing the ask. The real-world figure is therefore lower than 47.0%, and we cannot say by how much.
This page counts what exists, and nothing about what works
Four things these numbers cannot see, starting with the one that would change what you do.
There is no performance column here. We counted how often each ask appears. We did not measure what any of them produced, because we hold no attribution data: no installs per post, no link clicks, no store visits traced back to a video. Nothing on this page says a post carrying a call to action does better or worse than one without. Nothing here supports telling a creator to add one or drop one.
To answer that you would need installs or store visits attributable to individual posts. They would have to be compared inside the same creator and the same audience, so that the ask is the one thing that differs. No public content library carries that, ours included. Read the 47.0% as a description of what creators currently do, and stop there.
Nothing in the pipeline reads on-screen text. One text-only pass runs over the caption and the transcript. There is no video download, no frame extraction and no image model. An end card reading download now is invisible on both platforms. That puts a floor under the no-CTA count, and neither the platform split nor the transcript split can lift it.
Classification lags collection, heavily. 145,900 of the 565,898 videos in the library carry no classification and the backfill has never been run. The lag is worst at the head of the feed: a draw of the 1,995 newest videos returned 13 rows with a classification. That is why the sample above starts 40,000 rows down each platform's feed. Those rows still carry publish dates from 10 July to 14 August 2026, across 2,216 distinct handles.
The library is a sample of our customers' questions. Apps enter it because somebody asked about them. Content is linked to apps by store links where we can find them and by name matching otherwise. Name matching is deliberately generous and it is not perfect. These figures describe Apptonic's own library as of 14 August 2026. The library is a growing sample of the app market, not a census.
How we counted
Every figure comes from one frozen snapshot of the library, harvested on 14 August 2026. Nothing on this page is an estimate. Every count is a row count over the library or over a sample of it, with no model behind it.
The all-time distribution is the seven-value call-to-action facet over every classified organic video, summing to 419,998. Library totals come from a separate whole-library counter, which is where 565,898 comes from. The 145,900 unclassified videos are a subtraction across those two counters, 565,898 minus 419,998. The two are computed differently, so it is a residual and not a counted set. The per-platform figures are exact counts over one uniform 30-day window. The 60-day and 90-day candidates both pushed the largest cell past the 10,000 count cap, which returns a saturation value and not a number. 30 days is the widest window in which every cell is exact.
The transcript split is computed locally over a 2,796-row sample, 1,245 rows of which carry a classification. The sample starts 40,000 rows down each platform's feed because classification lags collection: a draw of the 1,995 newest videos returned 13 classified rows, so a newest-first draw answers nothing. At 40,000 rows down, 59 of 100 TikTok rows and 61 of 100 Instagram rows carried a classification. Those rows carry publish dates from 10 July to 14 August 2026 across 2,216 distinct handles. A row counts as having a transcript when it carries a resolved subtitle source.
Every video is classified by a language model into a fixed vocabulary of formats, hooks, tones and CTAs. It is a classification, not a human tag, and it is spot-checked, not benchmarked, so we publish it as a distribution across hundreds of thousands of posts, never as a judgement about any single video.
On TikTok we read the caption and the spoken transcript. On Instagram we read the caption alone. A call to action that is only spoken aloud is invisible to us. The sample refines that. 51 of the 586 classified Instagram rows did carry a resolved transcript, all of them through one rescue path. That is why the Instagram side of the transcript table is directional and the TikTok side is the measured one.
Content is linked to apps by store links where we can find them and by name matching otherwise. Name matching is deliberately generous and it is not perfect. These figures describe Apptonic's own library as of 14 August 2026. The library is a growing sample of the app market, not a census.
- Organic videos held in the Apptonic library
- 565,898 rows · Snapshot of 14 August 2026
- Organic videos carrying an AI classification
- 419,998 rows · Snapshot of 14 August 2026
- Classified posts counted per platform in one uniform 30-day window
- 22,961 rows · 30 days to 14 August 2026
- Sampled video rows carrying a classification, drawn 40,000 rows down each platform feed
- 1,245 rows · Published 10 July to 14 August 2026
Frequently asked
- What does no call to action mean in your data?
- It means the classifier read the post's caption, plus its spoken transcript where one resolved, and picked the value none out of seven. It does not mean the post had no ask. A creator who points at an end card lands in the same bucket. So does one who says the ask out loud in a video we could not transcribe.
- Is 47.0% the share of all app videos with no call to action?
- No, and the difference is large. 419,998 of the 565,898 videos in the library carry a classification, 74.2% of them. Against the whole library the 197,558 no-CTA posts are 34.9%. Against app video in general it is a number nobody has, ours included.
- Why is TikTok's no-CTA share higher than Instagram's when you read more of a TikTok?
- Because a caption is typed and speech is not. On Instagram we read the caption, where the words link in bio are written out. On TikTok we read the caption and the spoken transcript. A creator who gestures at on-screen text while saying nothing about it leaves no words for the classifier. Over the 30 days to 14 August 2026 the counts were 59.3% no CTA on TikTok against 39.0% on Instagram. Those come from 16,432 classified TikTok posts and 6,529 Instagram posts.
- Do you read text shown on screen in the video?
- No. One text-only pass runs over the caption and the transcript. There is no video download, no frame extraction and no image model. An end card reading download now is invisible on both platforms. That puts a floor under the no-CTA count, and neither the platform split nor the transcript split can lift it.
- Why was the sample drawn 40,000 rows down the feed instead of from the newest posts?
- Because classification lags collection badly enough to make a newest-first draw useless. A draw of the 1,995 newest videos returned 13 rows carrying a classification. At 40,000 rows down, 59 of 100 TikTok rows and 61 of 100 Instagram rows carried one, so that is where the sample starts. Those rows still carry publish dates from 10 July to 14 August 2026, across 2,216 distinct handles.
- Is link in bio the most common ask?
- Yes, among the posts that carry one. 222,440 of the 419,998 classified posts carry an ask, and 87,004 of those point at a link in bio against 40,103 asking for the download. That is 2.2 to 1.