How to Fix Bardeen Playbooks That Return Empty Rows
A Bardeen playbook that stops working does not throw an error. It returns a sheet with 40 rows and blank cells, and you notice three days later.
Silent failure is the defining weakness of visual browser automation. Everything below is about making failure loud.
Why do playbooks start returning empty rows?
Because the page’s structure changed and the extraction step still looks for the old one.
Typical triggers:
- The site shipped a redesign (class names change, containers move)
- A/B test dropped your session into a variant
- The list now loads lazily, so rows exist after your scrape finished
- A login expired and the page is a redirect, not the list
None of these produce an error. The extension executed all six steps successfully and extracted nothing.
How do you rebuild the extraction step?
The fix is almost always on the extraction node, not the whole playbook.
- Open the playbook and find the extraction step.
- Run it in single-step mode on the live page.
- If it returns zero rows, delete and re-add the step while the page is fully loaded.
- Prefer attribute-based selection (data attributes, role, aria labels) over nth-child paths. A list item’s
data-test-idsurvives redesigns;div:nth-child(3)does not. - Add a wait condition — “wait for at least 5 result items” — before extraction.
- Re-run on a second page to confirm it generalises.
How do you make failure visible?
Add a check step right after extraction. If row count equals zero, send yourself a message or stop the run.
Two rules that save real money:
- Cap the batch. Extract at most 25 rows per run and loop; a failed run wastes 25 credits, not 500.
- Log the URL and row count per run. A changing row count over days is your early-warning signal.
If the run feeds a CRM, add a duplicate check before push. A partially failed playbook that re-pushes ten records creates more cleanup than a playbook that does nothing.
How much do silent failures actually cost?
Two ways. Credits first — a five-step playbook run hourly consumes 120 runs a day, and if it is returning empty rows you paid for all of them.
The expensive cost is data. A lead list that silently emptied means a week of outreach missed, or worse, a CRM full of blank records that sales had to clean.
Set the schedule to daily unless there is a real reason. Nobody needs hourly scrape data for a prospecting list.
What should you move out of the browser?
Anything with an API. Bardeen exists to cover sites you cannot reach by other means.
Anything where you need guaranteed runs. The extension depends on a browser being open; a laptop asleep is a missed run. That is fine for a weekly prospecting list and unacceptable for a pipeline feeding a production system.
For agent-style work where the steps are not known in advance, a general browser agent such as AgentGPT plans its own path — worse for repeatable tasks, better for one-off research. Use each where it fits.
Quick Summary
- Broken Bardeen playbooks return empty rows instead of errors — always add a zero-row check.
- Rebuild only the extraction step, using data attributes rather than nth-child paths.
- Add a wait-for-N-items condition; lazy-loaded lists are a common cause.
- Cap batches at 25 rows and schedule daily, not hourly, to contain credit loss.
- Move production pipelines off the browser and onto APIs where they exist.
