I Let AI Clean My 50,000 Photos for a Year. Here's What Happened.

It started in May 2025 with a number that made me feel sick: 52,847 photos in my library. Twelve years of swipes, screenshots, food pics, conference badges, three weddings, two breakups, one move overseas, a dog who's now 9 years old. I had 87 GB of memories I never looked at.

I decided to try something different. Instead of doing what I'd done for a decade, vow to "clean my photos this weekend" and then not, I gave AI full control. For one year I let an AI photo cleaner make the recommendations, and I made the final call. This is what happened.

Numbers, mistakes, surprises. The photos I almost deleted that turned out to matter. The ones I deleted and never missed. And the question I keep coming back to: if AI knew me better than I knew myself, did I lose something by trusting it?

Recommended reading: How Gamification Makes Photo Cleaning a Habit

📊 The Starting Point (May 2025)

Here's what AI told me about my library on day one:

• 52,847 photos + 3,201 videos
• 87 GB used on iPhone (out of 256 GB)
• 11,418 duplicates or near-duplicates (21.6%)
• 4,832 blurry shots (9.1%)
• 9,201 screenshots (17.4%)
• 2,847 receipts and documents photographed (5.4%)
• Oldest photo: March 2013
• Most photographed person: me (sigh)
• Most photographed object: my dog
• Highest single-day count: 412 photos (a wedding, June 2018)

The number that hit me hardest: 11,418 duplicates. Eleven thousand photos that were the same thing twice. Or four times. Or twelve times. I had taken 17 versions of one particular sunset in Greece. Seventeen. Of the same sunset.

📅 Month 1: The Easy Wins

The first month was almost too easy. AI surfaced obvious cleanup candidates and I burned through them while watching TV:

• Week 1: 6,200 duplicates deleted. 4.1 GB freed.
• Week 2: 3,800 blurry photos deleted. 1.8 GB freed.
• Week 3: 8,400 screenshots reviewed; deleted 7,100 of them.
• Week 4: 2,400 receipt/document photos. Kept ~200, deleted the rest.

End of Month 1: 33,447 photos remaining. 19,400 deletions. 14.7 GB freed. I had spent maybe 4 hours total, in 5-minute swipe sessions, mostly on the toilet or in line at the supermarket.

What surprised me: I couldn't remember a single photo I'd deleted. Not one. The "what if I miss it" anxiety I'd carried for a decade was, evidently, mostly theatrical.

🎯 Month 2-3: Things Got Weird

The easy wins were gone. Now AI started showing me the harder cases. This is where the experiment got interesting.

The 47 identical selfies problem. AI grouped 47 photos of me from the same morning, July 2017. Same lighting, same shirt, same bathroom mirror. It suggested keeping the "best" one. I clicked through all 47. Realized I had no memory of why I took them. They were from a phase where I was trying to learn to smile more naturally for photos. (It didn't work.) Kept 2. Deleted 45.

The food photography era. 2,800 photos of plates. AI grouped them by location, restaurant, even by dish color. I went through restaurant-by-restaurant. Realized 90% were forgettable. But there was one folder, my mom's kitchen, that I almost deleted. 312 photos. Looking at them, I noticed she had used the same tablecloth for three years. The light through her kitchen window changed with the seasons. I kept all 312.

The screenshot graveyard. Most screenshots were obvious deletes (memes, tweets, web pages). But AI found 89 screenshots that were addresses, phone numbers, recipe ingredients, things I had screenshot because I'd need them "later." Some of "later" was three years ago. Kept the 12 that still mattered.

⚠️ Month 6: I Almost Lost Something Important

In November 2025 I almost made a real mistake. AI flagged a folder it labeled "low-quality, possibly duplicates": 184 photos from 2014, mostly blurry, mostly poorly composed.

I almost batch-deleted them. Something stopped me. I scrolled through.

They were from my grandfather's last summer alive. We didn't know it then. There were 184 mediocre photos because I was 22 and didn't know how to compose a shot, and because phone cameras in 2014 were terrible, and because I was just taking photos to take them. None of them are "good." All of them are now irreplaceable.

I kept all 184.

This is the moment that changed how I used AI for the rest of the year. The AI was right that they were blurry. The AI was right that they were technically low quality. The AI was wrong that they were deletable, because the AI didn't know my grandfather.

From that day forward, I added a personal rule: never batch-delete by quality alone for photos older than 5 years. Old blurry photos are often the only photos of a moment. New blurry photos are usually just bad photos.

🔄 Month 9: AI Started Surprising Me (In Good Ways)

By February 2026, I'd been at this for 8 months. AI had learned my patterns. It stopped suggesting I delete photos of my dog. It noticed I always kept sunset photos and stopped flagging them. It started organizing photos into albums I didn't ask for: "Trips to Italy," "My nephew growing up," "Sunday morning runs."

One Sunday it showed me an album titled "Photos of you from when you laughed." Sixty-one photos across 9 years. I'd never have thought to look for those. Most were taken by friends or family. I noticed I'd laughed more in 2018 than any other year. I noticed I was laughing more again recently. It was a strange thing to learn from an algorithm.

🏁 Month 12: The Final Count (May 2026)

Today, exactly one year later:

• 14,221 photos + 1,003 videos
• 22 GB used (was 87 GB, saved 65 GB)
• 38,626 photos deleted over the year (73% of original library)
• Photos I miss: 0 that I can name
• Time spent: roughly 38 hours total across the year (about 6 minutes a day)
• Albums AI auto-created that I kept: 31
• Photos AI suggested I delete that I kept: ~2,400 (mostly old/sentimental)
• Photos I deleted that I later regretted: 0

🧠 What I Learned

1. The library you have is bigger than the library you use

I had 52,847 photos. I actually looked at maybe 1,000 of them per year. The other 51,847 were taking up space in my phone, in my iCloud bill, and in some weird way in my head. Cleaning them didn't erase memories. It made the photos I cared about easier to find.

2. AI is great at the obvious; humans are needed for the meaningful

For duplicates, blurs, screenshots, receipts? AI is a magic wand. For old, low-quality photos of dead loved ones? AI is dangerous. The split is somewhere around the 5-year mark. Newer photos: trust AI. Older photos: trust yourself.

3. The dread is bigger than the work

I spent 10 years dreading the cleanup. The actual cleanup, when broken into 6-minute daily sessions, was almost pleasant. The hard part wasn't the work, it was facing the work. AI made facing it easier because each session had a clear, finite scope.

4. AI taught me about myself

I would never have known I laughed less in 2021. I wouldn't have noticed the tablecloth pattern in my mom's kitchen. I wouldn't have seen that almost every photo I take of my dog is from the same angle. Patterns emerge when something else organizes the data for you.

5. The 11,418 duplicates were a sign of something

I took 11,418 duplicate photos because somewhere in my brain I thought one of them would be "the one." I never went back and chose the one. I just took more. AI surfacing this made me take fewer photos in real time, because I now know I won't go back and pick the best later. Pick it now or it's clutter forever.

🔁 Would I Do It Again?

Yes. With three changes:

1. Start with the old stuff first, not the new stuff. The deletions get harder as you go further back. Doing the easy modern duplicates first builds momentum, but the hard old decisions are where the value really lives.
2. Set the "5-year rule" from day one. Don't let AI batch-delete anything older than five years. Review each one yourself.
3. Don't aim for a number. I didn't set out to delete 73%. That number just emerged. Aiming for it would have made me cut things I shouldn't have.

🛠️ What I Used

I used Tidy for the on-device AI scanning (duplicates, blurs, screenshots, faces). It runs entirely on the phone, no cloud uploads, which mattered to me because some of these photos were sensitive. The album auto-generation in months 9-12 was a side effect of how Tidy categorizes content.

The methodology was:

• Daily 5-minute sessions during downtime
• One category at a time (duplicates, then blurs, then screenshots, then by year)
• Always swiped manually, never batch-deleted
• Anything I hesitated on for more than 2 seconds, I kept

📝 Closing Thought

A year ago I had 52,847 photos and felt nothing. Today I have 14,221 photos and feel like I have a photo library, an actual collection of moments that matter. The technology was the catalyst, but the truth is, the photos were there all along. AI just helped me see them.

If you've been putting it off for years like I had, here's my one piece of advice: start tonight. Five minutes. One category. You'll be surprised how good it feels.