Google does not have a direct “satisfaction” meter, but it almost certainly infers satisfaction indirectly from aggregate behavior and quality proxies. The honest answer to both halves of the question is this: there is no single score that reads how happy a searcher was, yet the system does appear to learn, in aggregate, which results searchers ultimately settle on and which ones send them back to the results page looking for something better. Satisfaction is inferred, not measured. The specifics of how this works are not confirmed by Google in any detail, so treat the mechanics as informed inference worth verifying rather than fact.
It would be wrong to claim Google keeps a literal satisfaction number, and equally wrong to insist it ignores behavior entirely. The credible middle is that no per-visit emotion is captured, but patterns across many searches carry information. If searchers consistently click a result and stop searching, that aggregate pattern suggests the result did its job. If they click, return quickly, and click a different result, that pattern suggests the first one disappointed. None of this measures one person’s satisfaction; it reads tendencies across large volumes of searches, which is a very different and noisier thing than a direct meter.
The reason this distinction matters is practical. Because satisfaction is inferred rather than scored directly, you cannot optimize a phantom metric, and gaming a “dwell time” or “bounce” number treats an inferred signal as if it were a dial. What the system is ultimately approximating is whether your page genuinely answered the searcher well enough that they did not need to keep looking, which is a real thing you can influence directly by making the page actually satisfy the intent.
For your next page, optimize for genuine satisfaction rather than a number you cannot see: match the searcher’s intent fully, make the answer easy to find, and give people no reason to bounce back to the results in search of something better. Judge success by whether real users get what they came for, and treat any behavioral metric you can measure as a rough proxy, not the target itself.