Voice logging vs. photo logging: when each one wins

We ran an internal test: 50 meals, two methods (voice + photo), measured time-to-logged. Below are the results, and the rules of thumb that came out of them.

The numbers

Across all 50 meals:

• Photo log average: 9.8 seconds (capture + AI run + confirm)
• Voice log average: 7.2 seconds (Siri trigger + speak + confirm)

Voice was ~25% faster on average. But the average hides the truth.

When voice wins

Single-item logs. "Log a banana." Voice in 4 seconds vs. photo in 11.

Repeat meals. "Log my usual breakfast." Voice in 3 seconds vs. photo in 12.

Hands busy / driving / cooking. Photo isn't even an option.

At the table with company. Voice (with AirPods) is invisible. Photo isn't.

At a coffee shop counter. Quick log of "log a tall oat latte" before leaving. No need to find a clean angle on the cup.

When photo wins

Composite meals. A bowl with 6 ingredients takes 30 seconds to describe verbally and 8 seconds to snap.

Restaurants you've never been to. A photo gives the AI vastly more information than a verbal description ever could.

Portion uncertainty. "How big is this burrito?" — the photo answers; voice doesn't.

Anything with a sauce. "Log a chicken thigh with chimichurri" works; "log a chicken thigh with a green sauce that has parsley, garlic, oil, vinegar, oregano, and chili flakes" is silly.

Food you don't know the name of. Photo identifies it; voice depends on you knowing.

The hybrid that beats both

Take the photo. Then say "no croutons, extra olive oil." Photo for the visual; voice for the edit. Total time: 8 seconds, accuracy comparable to manual entry.

What about typing?

Manual typing averaged 38 seconds per meal in our test. Roughly 4–5x slower than either AI method. We do not recommend typing as the primary log path. It exists for edge cases.

The personal calibration

Look at your last 30 logs. Count how many were:

• Single-item (voice candidate)
• Composite (photo candidate)
• Repeat meals (saved-meal candidate)

If 40%+ are repeats, build a Shortcut. If 30%+ are single-items, learn the Siri commands. If you're ~80% composite, photo is your default.

Friction is the only metric that matters

The most accurate log is the log you make. The cleverest log method that takes 30 seconds is worse than the dirty 5-second log you actually do. The internal speed test is not academic — it's why our retention numbers track logging time per meal more closely than any other product metric.

Optimize for "did I log it" before "did I log it perfectly."