What is time to first byte?

The web made the term familiar. A browser asking for a page cares when content starts arriving, because that is when the screen stops being blank.

Voice borrowed it for the same reason. The moment audio starts playing is the moment the caller stops hearing silence.

It is not the same as total response time. A system can start speaking quickly and still take four seconds to say everything.

For a phone call the first measure is what the caller judges. Silence is uncomfortable in a way that a slightly long answer is not.

Picture two voice assistants asked the same thing: what time do you close today? The timings below are invented for illustration.

Assistant A starts speaking after 0.7 seconds. It thanks the caller for asking, lists the weekday hours, then Saturday's, then mentions that Sundays are closed, and finally says today's closing time is 6:00. The caller gets the answer nine seconds after asking.

Assistant B starts after 1.1 seconds and says: We close at 6:00 today. The answer has arrived by the two-second mark.

Measured by first byte, A wins by 400 milliseconds. On the measure the caller cares about, B wins by seven seconds. You could call it time to first useful word, and no vendor publishes it, though it's easy to notice on a demo. The reverse trap exists too. A system that says let me check that for you on every single turn has a superb first-byte figure. It hasn't told anyone anything yet.

The original web version of this metric is one you can check in a minute. It matters to a phone-driven business because many calls start with a tap on your site.

In Chrome, open your homepage, right-click, choose Inspect, and open the Network tab. Reload the page and click the first row. Under Timing you'll see a line called waiting for server response. That's your time to first byte.

Google's guidance treats 0.8 seconds or less as good. On cheap shared hosting, two or three seconds isn't unusual, and that's before anything appears on screen. A visitor on a cell connection stares at a white page and hits back, and the call button never even loads.

Test from your phone on cellular too, and not only from the office wifi. If the number is poor, the usual fixes are better hosting, a caching plugin, or a content delivery network.

We publish no timing figures at all, and this is one of them. Nothing here has been benchmarked to a standard worth printing.

What you can do is listen. The start of the reply is easy to judge by ear on a demo call.

Where a reply feels long rather than slow, shortening what the assistant says is the fix that actually works.