Most AI content statistics pages summarize other people's surveys. This one publishes our own production data: 524 live posts from Gravity's autonomous blog pipeline as of late July 2026, drafted, scored, pruned, and measured with almost no human writing. We are sharing the numbers because very few teams running fully autonomous content publish theirs, and because the dataset contains an expensive lesson worth getting for free. The pipeline proved that volume is a solved problem, then proved that volume without demand research earns almost nothing, then tripled its search results within weeks of flipping that priority. Every figure below is first-party, cited to gravity.fast production data with its period, including the numbers that make us look bad.
The headline numbers
Here is the dataset at a glance. All figures are gravity.fast production data drawn from publishing logs, rubric scoring records, and Search Console windows for the stated periods; the single external number is marked DataForSEO. For third-party industry numbers on agents themselves, see our AI agent statistics roundup.
| Metric | Figure | Period |
|---|---|---|
| Live posts published | 524 | Late July 2026 |
| Peak output | 355 posts in 39 days, about 9 per day | Early phase, 2026 |
| Current cadence | About 1 new post per day plus refresh cycles | Since late June 2026 |
| Publish bar | 80 or higher on a 100-point rubric | Every post, all periods |
| Posts pruned entirely | 19 | Late June 2026 |
| Clicks, volume-first phase | About 15 to 20 per month across ~473 posts | First half of 2026 |
| Impressions after pivot | About 8,200 to about 27,000 per 28-day window, 3.3x | Late June vs late July 2026 |
| Clicks after pivot | 63 to 188, 3x | Same windows |
| Best single-post rank | About position 4 for its money query | July 2026 |
| Wedge query demand | 40 US searches per month for "cheapest ai agent" | DataForSEO, July 2026 |
Two of these numbers matter more than the rest. The 3.3x impressions jump and the tripled clicks arrived within roughly a month of changing topic selection, with no change to the writing system, the templates, or the publishing infrastructure. The only variable was what we chose to write about.
How much content can an autonomous pipeline produce?
More than any team can usefully review. At peak, the pipeline published 355 posts in 39 days, roughly 9 per day, every day, with quality gates rather than editors deciding what shipped (gravity.fast production data, early phase, 2026). The full build log of that sprint, architecture included, is in the 355 posts in 39 days write-up.
Each post moved through the same chain: a research brief, a draft against a fixed template, internal links into its topic cluster, schema markup, a generated cover, and scheduled publishing. Humans set strategy and reviewed samples; nobody wrote body copy. At that cadence, production capacity stopped being the interesting question within weeks. The pipeline could always make more.
Then we cut it on purpose. In late June 2026 the cadence dropped to about one new post per day plus refresh cycles on existing posts, not because the pipeline broke, but because the click data made clear that output was no longer the constraint. The next two sections explain what that data showed.
What quality gates does autonomous publishing need?
Every post must score 80 or higher on a 100-point rubric before it publishes (gravity.fast production data, all periods). The rubric covers sourcing, structure, internal linking, schema, and readability; a post that misses the bar is rewritten and rescored automatically rather than shipped thin.
The sharper gate works on outcomes, not prose. An automated topic gate banned an entire category, infrastructure and ops explainers, after the category earned impressions but zero clicks over its measurement window. No individual post failed the rubric; the subject itself failed the market, so the pipeline stopped writing about it.
Gates matter more in autonomous systems than in human ones because nobody reads every draft. Review has to be encoded or it does not happen. The honest caveat, proven at our expense: gates like these enforce craft, not demand. A rubric cannot tell you that nobody is searching for the thing you wrote well.
What did the volume-first phase get wrong?
Almost everything except the writing. The first roughly 473 posts optimized for output volume on low-demand topics, and the combined return was about 15 to 20 clicks per month (gravity.fast production data, first half of 2026). Nineteen of those posts were eventually pruned from the site entirely.
The failure mode is worth naming precisely. Each of those posts cleared the 80-point bar; the craft was fine. What the system never asked was whether anyone searched for the topic before generating it. We had built a machine that answered questions nobody was asking, at scale, on schedule.
This is the number most autonomous-content case studies leave out, and it is the most useful one here. Months of near-peak output bought a click total that rounds to a rounding error. In July 2026, demand-research-driven topic selection replaced volume as the pipeline's organizing principle, and the next section shows what that changed.
What changed after demand research replaced volume?
The same pipeline, pointed at researched demand, roughly tripled its results in a month. Comparing late June to late July 2026 windows, search impressions grew from about 8,200 to about 27,000 per 28-day window, a 3.3x increase, and clicks rose from 63 to 188, a 3x increase (gravity.fast production data).
Individual posts moved the same way. One research-targeted post reached about position 4 for its money query within weeks of publishing, a rank the volume phase never touched for any commercial term. Nothing about the writing system changed between the two periods; topic selection now starts from measured search volume and keyword difficulty instead of a list of things worth explaining.
Two caveats keep this honest. The base was small, so multiples came easier than they will later, and 188 monthly clicks is a starting line, not a victory. The point of the comparison is direction and cause: demand research was the single change, and every headline metric moved with it.
Why do small ponds trap volume-first pipelines?
Because a pipeline can rank well in a pond that holds almost no fish. The pipeline's first wedge query, 'cheapest ai agent', gets about 40 US searches per month (DataForSEO, July 2026). Ranking on page one for it, which the pipeline achieved, still produced single-digit weekly clicks, because 40 searches is the whole pond.
Volume-first content drifts toward small ponds naturally. Uncontested queries are easy to win, winning them feels like progress, and nothing in a production metric flags that the prize is tiny. The fix costs one query to a keyword database before generation instead of after: check the pond before automating the fishing.
What should teams take from these numbers?
Five findings from the dataset generalize beyond our pipeline. They apply to any team pointing autonomous agents at content, and most of them apply to agent work far beyond content; for the broader capability picture, see what an AI agent can actually do and our library of AI agent examples.
- Volume is solved. A sustained 9 posts per day proved capacity is not the differentiator; anyone can make more content now.
- Quality gates are necessary, not sufficient. An 80-point publish bar produced consistently well-made posts and still allowed roughly 473 of them to land on low-demand topics.
- Demand research belongs before generation. The 3.3x impressions and 3x clicks gains came entirely from changing what got written, not how.
- Prune without sentiment. Deleting 19 posts and banning a whole category cost nothing measurable and cleaned the site's focus.
- Describe the outcome, not the workflow. The pipeline improved when we specified the result, qualified traffic, instead of the activity, daily posts. That framing shift is the same one that makes agents useful anywhere; we unpack it in describe the outcome, not the workflow.
These numbers are free to cite. Attribute them to gravity.fast production data with the stated period, and link this page so readers can see the unflattering parts in context.
Frequently asked questions
How many posts has Gravity's autonomous pipeline published?
524 live posts as of late July 2026 (gravity.fast production data). Output peaked at 355 posts in 39 days during the early phase, about 9 per day, before a deliberate cut in late June 2026 to roughly one new post per day plus refresh cycles on existing posts.
Does publishing more AI content produce more traffic?
Not on its own. The first roughly 473 posts optimized for volume on low-demand topics and earned about 15 to 20 clicks per month combined (gravity.fast production data, first half of 2026). Traffic tripled only after demand-research-driven topic selection replaced volume in July 2026.
What quality checks does autonomous AI content need?
Every Gravity post is scored on a 100-point rubric and must reach 80 or higher to publish. An automated topic gate also blocks whole categories that fail in search: infrastructure and ops explainers were banned after earning impressions but zero clicks. Gates matter more when no human reads each draft.
What results followed the demand-research pivot?
Comparing late June to late July 2026 windows, search impressions grew about 3.3x, from roughly 8,200 to 27,000 per 28-day window, and clicks tripled from 63 to 188 (gravity.fast production data). One research-targeted post reached about position 4 for its money query.
Are these statistics from a third-party study?
No. Every figure except one is first-party gravity.fast production data from our publishing logs, rubric scoring records, and Search Console windows for the stated periods. The single external number, about 40 US searches per month for 'cheapest ai agent', comes from DataForSEO, July 2026.
Sources
- gravity.fast production data: publishing logs (post counts, cadence, pruning), 100-point rubric scoring records (publish bar, topic gate), and Google Search Console 28-day windows compared for late June and late July 2026. Backs every first-party figure on this page.
- DataForSEO, Keyword Data, July 2026, backs the search volume figure of about 40 US searches per month for "cheapest ai agent".