Eat smart.
Live better.

AI calorie counting is typically accurate within ±15–20% per meal. That sounds loose, but it works because consistency matters more than precision: tracking the same meal the same way every day reveals real trends, even if every individual estimate is slightly off. CrosWeight uses portion calibration and food recognition to keep error margins predictable, and the weekly average smooths out daily noise. For weight management, a tracker that captures 85% of your intake consistently beats a perfect tracker you abandon after three days. Pair AI estimates with weekly weigh-ins and trust the trend line, not any single meal.

When CrosWeight can't identify a homemade dish — say, your grandmother's stew or a custom protein bowl — switch to manual macro entry. Tap the meal, hit edit, and enter the calories, protein, carbs, and fat directly. For repeat meals, save them as a custom template so next time it's one tap. If you only know one component (e.g. you weighed the chicken but eyeballed the rice), enter what you know and estimate the rest from a similar logged meal. Photo recognition covers ~80% of common foods; the manual fallback closes the gap without breaking your streak.

3–5 days of tracking per week beats daily perfection. Research on adherence shows that people who track most days but allow flexibility stick with it for months, while perfectionist daily-trackers tend to quit within 4–6 weeks. The behaviors that drive results are: weekly weight averaging, hitting your protein target most days, and noticing patterns (weekend overshoot, hidden liquid calories). You can miss a Sunday and still progress. What matters is that over a month, your logged days form a representative picture of how you actually eat.

Use AI tracking for daily logging, portion awareness, macro balance, and trend monitoring — the high-frequency, low-stakes stuff. See a registered dietitian or nutritionist for medical conditions (diabetes, PCOS, GI disorders), pregnancy, eating disorder recovery, athletic performance programming, or if you have ten or more weeks of consistent tracking and still can't move your trend. AI handles "what did I eat and how much" — humans handle "why isn't it working" and "is this safe for my body". Most people benefit from AI alone; some need both.

Log immediately after eating whenever possible. Recall accuracy drops sharply after about two hours: a 2018 study on dietary recall found people underestimated intake by 15–25% when logging end-of-day vs immediately. CrosWeight's photo capture takes ~10 seconds, which is faster than typing the meal later from memory. If you genuinely can't log mid-meal (a work lunch, a date), snap the photo discreetly and let the AI process it later — the image itself preserves the portion data your memory will lose. End-of-day batch logging is fine for repeat meals but unreliable for new ones.

Restaurant meals are the hardest case for any tracker because portions and oils vary by chef. Three strategies: (1) Photo log it and accept the estimate, knowing it may be ±25% off — one inaccurate meal a week barely moves a trend line. (2) For chains, search the brand in the database for posted nutrition info. (3) For unknown dishes, log the closest match and round up by 20% to account for restaurant butter, oil, and dressings. Don't skip logging because it's imperfect — a rough estimate is more useful than a missing meal. The pattern still emerges.

Photo recognition and AI macro estimation require an internet connection because the analysis runs in the cloud. However, you can still log manually offline: open the app, enter calories and macros directly, and the entries sync to your weekly trend the next time you connect. Your existing logs, weight history, and goal data are cached locally and viewable offline. For travel or low-signal environments, log manually in the moment and let CrosWeight back-fill AI analysis later when you reconnect. Your streak and progress data never depend on connectivity.