Automated content pipelines
A quality gate for automated blog pipelines
If your pipeline generates posts on a schedule, generation is the easy half. The hard half is making sure each post is worth a reader's time. DraftFilter sits between generation and publish: score the draft, flag the sentences that sound like AI, rewrite, and only then push to the CMS.
The problem: generated posts converge on the same voice
Automated blogs fail in a predictable way: every post starts to sound the same. The models that make scaled publishing possible also produce recognizable patterns — overlong introductions, bullet lists standing in for argument, conclusions that restate the title. Readers who land on one templated post rarely come back for a second.
Manual review does not scale to a pipeline publishing dozens of posts a week, and skipping review means shipping the misses. What scales is an automated check with a threshold you choose, plus sentence-level detail for the posts that need work.
How it works: scan, heatmap, rewrite
1. Score every generated post
Each draft gets a 0-100 AI-likelihood score. In the web app you paste or upload; in a pipeline you submit the post body to the score API as part of the publish job.
2. Inspect the ones that fail the bar
For posts that score too AI-like, the heatmap shows exactly which sentences read machine-written, so a human editor — or the rewrite step — works on the weak lines instead of the whole post.
3. Rewrite, rescore, publish
The rewriter reworks the flagged sentences and rescores the result. Loop until the post clears your threshold and sounds human, then let the pipeline publish it.
No black box: read how the detector scores text, including what the score means and where it can be wrong. It is a probabilistic signal for review, not proof of authorship.
Wiring it into the pipeline
The Pro and Max plans bundle API access sized for pipelines. A typical setup adds two steps to the publish job: score the generated post, then either publish, loop it through the rewrite API, or park it for human review depending on the is_ai score. Batch scoring handles a backlog, and idempotency keys make retries safe.
# After generation, before publish: score the post.
curl -sS https://api.draftfilter.com/v1/score-jobs \
-H "Authorization: Bearer $DF_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"idempotencyKey": "post-2026-07-09-a",
"items": [{ "id": "draft", "text": "Generated post body..." }]
}'
# Poll the job; publish only when is_ai comes back low,
# otherwise send the flagged sentences through the rewrite API.Score a generated post
Paste one of your pipeline's posts and see what the heatmap flags. Free scans, no account required.