We pulled 5,302 Google autocompletes about people's own faces. Here is what they are really asking

It is after midnight. You type a question into Google that you would not say out loud: am i ugly, do I look normal, why do I look fine in the mirror and wrong in a photo? The suggestion list drops down before you finish, and suddenly the private question looks public.

Here is what the data actually says: nobody is asking what they look like. They are asking who to believe. A single AI face score feels tempting, but can fail as a tiebreaker. Here is what our census found, what it cannot prove, and what the questions say about first impressions.

Key numbers

• 5,302 distinct completions from 26 seed phrases in our collection method.
• 2,334 self-directed appearance completions after the filter in our method.
• 205 completions, or 9%, concerned mirror versus camera (family breakdown).
• 30 completions named AI, ChatGPT, or GPT (machine section).
• 100 ms was enough for impressions in one face-perception study to correlate highly with impressions formed with unlimited time. (Willis and Todorov)

Table of contents

• How we collected this
• The seven things people ask about their own face
• The mirror-versus-camera problem is the whole story
• People are now asking a machine
• The "or just" queries
• What we did not find
• The bottom line
• Sources

How we collected this

We queried Google's public suggestion endpoint on 2026-09-04 with 26 seed phrases, including "am i ", "am i ugly ", "am i attractive ", "why do i look ", "do i look ", "is my face ", and "is my jawline ". Each was queried bare and with every letter from a to z appended. The endpoint returned 5,302 distinct completions; we kept completions containing a first-person marker and an appearance word, excluding medical, post-operative, software, and media noise. That left 2,334.

This is a census, not a count of searches. Autocomplete is ranked, filtered, and personalizable: it signals what Google suggests, not how many people typed a phrase. Do not read 443 as 443 people.

Caveat: our filter makes the categories legible, but it also makes judgment calls. Another researcher could set slightly different inclusion rules and get a different map.

The seven things people ask about their own face

The 2,334 completions fell into seven families. Verdict-seeking was largest at 443 completions, while one named feature followed at 412.

One named feature. "is my chin recessed" makes uncertainty feel solvable, though a face is perceived as a whole. See how to know if you're attractive.

Mirror versus camera. "why do i look ugly in the camera but pretty in the mirror" shows a searcher with images but no trusted referee.

Photo: Atahan Demir on Pexels

“Or just…” self-doubt. "am i actually ugly or just insecure" is more honest than a binary verdict.

Numbers and rankings. Only 9 completions asked for a scale, including "rate my face out of 10", "rate my looks 1-10", and "am i ugly scale".

Will I ever. Three completions—"will i ever be pretty", "will i ever know what i actually look like", and "will i ever see your face again"—sound like reassurance that the future is not already decided.

Caveat: these families overlap in real life. We assigned each completion to one primary family so the census could be read, not because a person has only one concern.

The mirror-versus-camera problem is the whole story

The mirror-versus-camera split creates the need for a tiebreaker. It appears in "why do i look different in pictures than i do in the mirror", "why do i look normal in the mirror but fat in pictures", "why do i look in the mirror and not recognize myself", and "is my face inverted".

A mirror gives you a reversed, familiar, moving, three-dimensional view; a photo gives you a frozen, unreversed, single-perspective view at a camera-chosen distance. A close lens can change proportions: a mathematical model found that a photo taken at 12 inches increased perceived nasal size by 30% in males and 29% in females compared with an orthographic projection; at 5 feet, a standard portrait distance, the difference was essentially absent. (Ward et al.)

Perspective distortion caused by interpersonal distance changed social judgments of the same face. (Bryan, Perona, and Adolphs) That makes front-camera versus back-camera, the best focal length for portraits, and why you are not photogenic practical questions, not excuses. Neither image is “the real you”: use several ordinary images and real-world feedback, not one distorted close-up.

Photo: cottonbro studio on Pexels

Caveat: perspective explains meaningful distortion, not every bad photograph. Expression, lighting, timing, posture, and familiarity still matter, and we did not measure any individual image.

People are now asking a machine

Thirty of the 2,334 self-directed appearance completions named AI, ChatGPT, or GPT. The repeated request is for candor: "am i ugly brutally honest ai", "am i ugly honest ai", "chatgpt how good looking am i", and "am i attractive or not ai". People keep adding “honest.” An AI face score looks like a tiebreaker, but a confident output is only a model's interpretation of one image; it cannot turn one projection into a universal truth. The missing question is can ChatGPT rate your face: what kind of evidence would make that number worth believing?

Caveat: our 30-count captures explicit wording only. A person can use an AI tool without naming it in the search phrase.

The “or just” queries

The “or just” family is the smallest, with 23 completions, but it may be the most revealing. "am i actually ugly or just insecure", "am i ugly or is it just the camera", and "am i ugly or just not photogenic" ask whether the interpretation deserves trust.

A number cannot answer it. The humane response is to separate the photograph, the first impression, and the person's worth, then look for better evidence without pretending appearance anxiety is solved by a slogan. You do not need to love every image of yourself to be socially attractive, and you do not need to dismiss the worry as silly to examine it carefully. Do face-rating apps cause insecurity? is the right question when measurement starts making the observer less steady.

Caveat: we are not diagnosing anyone from a search phrase. We are reading the wording as a clue about uncertainty, not a clinical symptom.

What we did not find

The negative result matters: only 3 completions were in the hopelessness family, and only 9 were in numbers and rankings, against 443 verdict-seeking completions. Our reading is that ranking vocabulary is much louder in forums than in this search box, but this is an interpretation of a filtered dataset, not a measurement of how people feel. We did not find a clean, single question called “what do other people see?”—that concern was scattered across mirror, camera, feature, and machine language.

Caveat: absence from autocomplete is not absence from human experience. Google decides what to show, and our inclusion rules decide what to count.

The bottom line

People are not short of images of their faces. They are short of a trusted way to reconcile the mirror, camera, reactions, and anxious reading. Our 2,334-completion census asks not “am I ugly?” but “who should I believe?”

Real World Appeal is our attempt to provide that missing axis: a free first-impression report on a 70-155 perception axis, with no paywall on your result. You see the result before deciding anything, and you can start at /test. It is still a model, not a judgment of your worth and not a replacement for varied real-world feedback.

Sources

Ward, B., Ward, M., Fried, O., & Paskhover, B. (2018). “Nasal distortion in short-distance photographs: The selfie effect.” JAMA Facial Plastic Surgery, 20(4), 333-335. https://pubmed.ncbi.nlm.nih.gov/29494735/

Bryan, R., Perona, P., & Adolphs, R. (2012). “Perspective distortion from interpersonal distance is an implicit visual cue for social judgments of faces.” PLoS ONE, 7(9), e45301. https://pubmed.ncbi.nlm.nih.gov/23028918/

Willis, J., & Todorov, A. (2006). “First impressions: Making up your mind after a 100-ms exposure.” Psychological Science, 17(7), 592-598. https://pubmed.ncbi.nlm.nih.gov/16866745/

Our original census was collected on 2026-09-04 from Google's public suggestion endpoint with client=firefox, hl=en, and gl=us. The endpoint is not treated as a linkable source here; the method and its limits are reported above.