AI content benchmarks 2026: what 2,575 Kompozy outputs show
Original research across 10,000 Kompozy outputs. Platform-by-platform engagement, format-by-format CTR, autopilot vs manual review quality.
Published 2026-04-24 · Updated
Short answer
On October 23, 2026 we measured 2,575 AI outputs from 72 Kompozy workspaces, created April 25 to October 22. 69.1% were published, 16.1% were finished but unpublished, and 14.8% failed. Of 6,256 scheduled posts that reached a final state, 97.3% went out. Quote graphics and carousels were published most often, and clipped shorts failed most often. This study does not measure engagement.
Key takeaways
- 69.1% of 2,575 AI outputs from 72 workspaces were published, and 14.8% failed (measured October 23, 2026).
- Quote Graphics (87.1% of 225) and Carousel Posts (86.3% of 278) had the highest publish rates.
- Clipped Shorts failed 67.6% of the time. Most Blog Article failures happened at the publish step.
- 97.3% of 6,256 publish attempts succeeded. Facebook, Instagram and Threads each reached 99.5%.
- The five busiest workspaces made 77.1% of outputs, so read every figure as directional.
What did this study measure?
We measured what happened to every AI output created in Kompozy between April 25 and October 22, 2026, using the production database on October 23, 2026. The sample is 2,575 outputs from 72 workspaces, plus 6,256 publish attempts from 14 workspaces. Each output ended in one of three states: published, finished but not yet published, or failed.
This is not an engagement study. Kompozy stores post metrics for too few workspaces to report them without pointing at individual accounts, so you will not find likes, views or click-through rates here. An earlier version of this page reported engagement and autopilot approval figures from 10,142 outputs and 500 workspaces. We could not trace those figures to any dataset, so we removed them. A small sample we can reproduce is worth more than a large one we cannot.
How many AI outputs actually get published?
69.1% of the 2,575 outputs were published. Another 16.1% were finished and still sitting in the pipeline on the measurement date, and 14.8% failed. An output counts as failed when generation, rendering or publishing did not complete.
Most workspaces made very few outputs. The median workspace had 2 in the window, the 75th percentile about 13 and the 90th percentile about 48. The five busiest workspaces produced 77.1% of all outputs, so the totals mostly describe heavy users.
Volume peaked in May 2026 at 985 outputs and fell to 123 in October (through the 22nd). June had the most active workspaces: 46 of them created at least one output that month.
Which formats are most likely to be published?
Quote Graphics had the highest publish rate: 87.1% of 225 outputs from 24 workspaces. Carousel Posts came next at 86.3% of 278, then Listicle Video at 80.4% of 158 and Text Posts at 79.9% of 273. Infographic Photo (77.9% of 222), Persona Infographic (76.5% of 119) and Persona Tweets (73.8% of 233) were close behind.
Naturalistic Video (69.9% of 136) and Photo Posts (64.1% of 237) sat in the middle. Persona Photos had the lowest publish rate of the social formats, 51.9% of 185, and failure was not the main reason: a third of them (33.0%) were finished but never published.
We only report formats with at least 10 outputs from at least 3 workspaces. Anything below that line is left out.
Which formats fail most often?
Clipped Shorts failed most often. 67.6% of 108 outputs from 10 workspaces ended in a failed state, and none of the failed ones ever reached a publish attempt, so they broke during generation. Blog Articles failed 33.0% of the time (279 outputs, 14 workspaces), and 70 of those 92 failures came from a failed publish attempt rather than generation. Email Newsletters failed 34.5% of the time, across 29 outputs from 9 workspaces.
How reliable is scheduled publishing?
On the main social platforms, scheduled posts went out almost every time. Of 6,256 publish attempts that reached a final state, 97.3% succeeded. Facebook, Instagram and Threads each succeeded 99.5% of the time, LinkedIn 99.1%, TikTok 98.6%, X 98.0% and YouTube 95.1%.
Two destinations were much weaker. Blog posts to a connected GoHighLevel site succeeded 67.1% of the time (213 attempts), and Google Business Profile posts 62.3% (122 attempts). When a post did go out, the median delay was about one minute after its scheduled slot.
How far ahead do workspaces schedule?
Further than we expected. The median published post was scheduled about 14 days (335 hours) before it went out, and the median gap from generation to publication was about 19 days (462 hours). Each published output went to a median of 4 platforms (mean 3.37): 1,759 outputs produced 6,084 published posts.
What can this data not tell you?
It cannot tell you how AI content performs with an audience. There are no engagement, reach or conversion figures here. It also cannot tell you why a finished output was never published. It may be waiting for review, it may have been passed over, or it may be scheduled for later.
The sample is small and concentrated. Three workspaces account for 90.1% of published posts, and only 14 workspaces made any publish attempt at all. Statuses are a snapshot taken on October 23, 2026, and outputs a user deleted are not counted.
How do public engagement benchmarks compare?
For engagement, use published third-party data and check how each source calculates its rate. Socialinsider measured TikTok engagement by views at 4.20% in 2025, up from 3.85% in 2024. Hootsuite puts average 2026 engagement at 3.9% on LinkedIn and 4.2% on Instagram, calculated as engagements divided by followers. A rate by views and a rate by followers are different numbers, so only compare figures that use the same method.
What should you do with these findings?
If you are picking formats for an AI content workflow, start with the ones that reliably reach publication. Quote graphics, carousels, listicle videos and text posts each published about 80% of the time or more in this sample. Check clipped shorts, blog articles and newsletters by hand, because they failed far more often. The guide on turning long-form into clips covers what a usable clip source looks like.
If you post to many platforms at once, the scheduler is the dependable part: the social platforms succeeded between 95.1% and 99.5% of the time. Connected blogs and Google Business Profile are where to watch for failures. For the setup, see how to cross-post to all platforms and what an automated social content engine actually is.
How was the data collected?
We ran read-only SQL against the Kompozy production database on October 23, 2026. The window covers rows created from April 25, 2026 (the earliest surviving output) through October 22, 2026. An output is a generated content row that entered generation; source items still waiting for approval are not counted. A publish attempt is a scheduled post that ended as published or failed, so queued and cancelled posts are left out.
We excluded 14 of the 214 workspaces in the database: those owned by Kompozy staff or sales reps, and those whose owner email matched an internal or test pattern. We also suppressed any figure based on fewer than 10 rows or fewer than 3 workspaces. The exact queries are saved in the Kompozy codebase as scripts/research/ai-content-benchmarks-2026.sql. Re-running them later will give different numbers, because statuses change and deleted rows drop out.
What does this look like in Kompozy?
A real Quote Graphic generated in Kompozy. Quote Graphics had the highest publish rate in this study: 87.1% of 225 outputs from 24 workspaces, measured October 23, 2026.
Who is this for, and who should skip it?
Best for:
- Teams choosing which AI content formats to rely on
- Operators checking how reliable scheduled multi-platform publishing is
- Readers who want a small first-party sample with a reproducible method
Not ideal for:
- Anyone looking for engagement, reach or conversion benchmarks
- Niche-specific or large-sample performance data
Frequently asked questions
Is this data from real customer accounts?
Yes. It is an aggregate of production Kompozy data from 72 workspaces, measured on October 23, 2026. We excluded 14 staff, sales-rep and test workspaces, and we only report figures based on at least 10 rows from at least 3 workspaces. No account is named or identifiable.
How often do you update these benchmarks?
Whenever we re-run the measurement script, and each update states its run date. This version was measured on October 23, 2026 and covers outputs created from April 25 to October 22, 2026.
Can I access the raw dataset?
No. We publish aggregates only, because row-level data would expose individual workspaces. The method section describes the exact queries, which are saved in the Kompozy codebase so the figures can be reproduced.
How do these numbers compare to public social media benchmarks?
They measure different things. This study tracks what happens to an output (published, failed or still waiting), not engagement. For engagement, Socialinsider measured TikTok engagement by views at 4.20% in 2025, and Hootsuite's 2026 benchmarks put LinkedIn at 3.9% and Instagram at 4.2%, both measured against followers.
Which AI content formats were most likely to get published?
Quote Graphics (87.1% of 225 outputs), Carousel Posts (86.3% of 278) and Listicle Video (80.4% of 158). Clipped Shorts failed most often, at 67.6% of 108 outputs, and none were published from the pipeline.
How do we know this?
Read-only SQL run on 2026-10-23 against the Kompozy production database: 2,575 outputs (72 workspaces) and 6,256 publish attempts (14 workspaces), created 2026-04-25 to 2026-10-22. 14 staff, sales-rep and test workspaces excluded; figures under 10 rows or 3 workspaces suppressed. Engagement not measured. Queries: scripts/research/ai-content-benchmarks-2026.sql. Third-party engagement figures checked against Socialinsider and Hootsuite.
Sources
- Socialinsider — TikTok benchmarks (read 2026-10-23)
- Hootsuite — Average engagement rate benchmarks (read 2026-10-23)
- Kompozy — Pricing (output formats and platforms) (read 2026-10-23)
Get started → · ← All guides · Compare Kompozy vs other tools