## TL;DR
Survey pages are assembled from third-party scripts that load slowly and unevenly, plus consent and overlay elements that intercept the agent's actions. Wait for the survey's first question element, dismiss overlays by their close controls, and if the survey platform offers response data via its own dashboard or API, collect results there instead of driving the respondent page.

```text
browser automation failed on the NPS survey page: agent timed out on load
```

## Use this when
- The agent times out on the survey page while other pages work
- The page contains embedded third-party widgets or consent banners
- A human can complete the survey, slowly

## Not for this skill when
- The survey link is expired or the survey was closed
- The agent's network blocks the survey domain entirely

## Steps
1. Load the page with tracing and note which elements the agent waits on and what blocks them. Verify: the hanging element and any overlay are identified
2. Replace page-load waits with a wait for the first question's input element, and handle the consent overlay explicitly. Verify: the agent reaches the first question reliably
3. Check whether the task needs the respondent flow at all: if it needs results, find the platform's results export or API. Verify: a non-browser collection path is confirmed or ruled out
4. Move result collection to that path, keeping the browser flow only for a periodic respondent-experience check. Verify: scheduled collection no longer depends on the slowest page in the stack
5. Record load timings for the survey page in the experience check and alert on degradation. Verify: survey page decay is noticed before it blocks again

## Variant phrasings
### NPS survey page load timeout
A related phrasing for the same underlying problem.
### browser agent stuck on survey page
A related phrasing for the same underlying problem.
### third party embed slows survey
A related phrasing for the same underlying problem.

Compatibility: Compatibility: Playwright or Selenium against survey platforms (Delighted, Qualtrics, Typeform) respondent and results pages, current versions.

## Why it happens
The agent's real need is usually the data, not the page. Respondent pages are the least stable surface a survey platform offers, while results exports are built for machines. Separating the experience check (browser, occasional) from data collection (export, scheduled) puts each job on the surface designed for it.

## Edge cases / pitfalls
- Driving a respondent page can create test responses that pollute real NPS. Use preview links or a test survey for browser checks.
- Consent overlays differ by region; the agent's egress location decides which one appears. Test from the agent's actual network.
- Survey platforms rate-limit result exports too. Schedule collection with the same pacing discipline as any API.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_M-wWfEJsC--xZewzRBf_4Q