How Dog Says Hi works
This page explains how a photo lookup is produced, what the readings are, how uncertainty is shown, and how the guides are made.
How a reading is produced
A lookup starts only when you choose a photo and press “Look at this dog.” The tool accepts JPG, PNG, and WebP photos (HEIC and HEIF files are not supported). Your browser resizes the photo to a JPEG no more than 1,600 pixels on its longest edge, then sends it — with your optional note of up to 600 characters — to this site’s Worker at /api/analyze. The Worker applies an anonymous rate limit — a best-effort limit configured for 10 lookups per 60 seconds per connecting IP in each Cloudflare location — and rejects requests larger than 12,000,000 bytes, then forwards the lookup to Qomza’s analysis service at /v1/analyze together with your Cloudflare connecting IP as request metadata. An AI model returns a structured reading, the Worker checks it against the expected shape, and the site shows it in the result panel: visible traits first, then activity in the photo, then possible breed cues, then everyday care ideas.
Qomza keeps each submitted photo, note, request metadata, and result in private operational storage, and those records are not deleted automatically. The Privacy page describes exactly what is sent and retained, and how to raise a privacy request.
Readings are generated by an AI model
Every reading is written by an AI model looking at a single image plus your optional note. It can be wrong or incomplete, and it describes only that frame — not the dog’s history, health, or temperament. Treat each result as a starting point for noticing your dog, not a verdict.
How uncertainty is shown
Each result opens with a status line: “Visible details in this photo,” “Some details are hard to confirm,” “Another angle may help,” or “The photo may show something else.” Breed suggestions sit under “Possible breed cues” beside the note “These are visual possibilities only. Mixed ancestry is common, and appearance alone cannot confirm breed.” When nothing visible supports a cue, the list stays empty and says so: “No breed cue is strong enough to include.”
What the tool does not do
A photo cannot confirm breed or ancestry. The tool does not assess health, emotion, or temperament; it does not diagnose conditions, evaluate symptoms, or triage concerns; and it does not certify a breed. Grooming and activity notes are general starting points, not a care plan for your dog. If you have a health concern, contact a veterinarian. The Terms of use describe the service scope.
How the guides are made
The guides are researched from named veterinary, animal-welfare, and peer-reviewed sources — including PDSA, Blue Cross, the American Kennel Club, the American Veterinary Medical Association, and shelter-dog research published in PLOS One — and each guide links its sources beside the claims they support. Guide hero images and a few inline figure illustrations are AI-generated, and their captions say so; the homepage portrait is a credited photograph. Each guide shows a “Last updated” date reflecting its most recent content change.
Corrections
We read every problem report, and when a guide error is confirmed the guide is corrected and its “Last updated” date changes to the day the correction is published. To report a wrong reading or a mistake in a guide, use Report a problem on the Help page.