Finally, Someone Likes My Photos
A tongue in cheek investigation where a fake Brooklyn baker helps me explore what makes Instagram tick
Last February, I published an article titled “No One Likes My Photos” that explored my hate/love relationship with Instagram. It sought to understand why only some of my images got attention. The article clearly struck a chord with readers who had similar experiences, and it became one of my most-read and most-commented articles.
Where my earlier article addressed my hypotheses on which visual qualities determined if an image would get attention on Instagram, I ignored the most important factor in IG-based image discovery — the algorithm. As a former e-commerce product manager who spent many years working on both traditional as well as AI-based search algorithms, I am embarrassed by the omission of this key aspect of image discovery. But not to worry, it’s time to delve into the belly of the Instagram beast, THE ALGORITHM…
The Experiment Background
Recently, I’ve posted a number of self-portraits that involve mirrors and double exposures. As an art student, it’s my job to explore different ways of seeing, and these self-portraits have been a fun exercise. The other day, I checked Instagram and noticed something statistically very interesting. With my nine most recent images, all the images that contained people in them had virtually the same number of likes. The same thing was true with the images without people in them; they all got virtually the same number of likes. However, images with people in them had gotten 3x as many likes.
The idea that images with people in them would get more attention is not surprising, especially since a number of the images were self-portraits. But then I noticed something that became the genesis of my little experiment here. One of the images with people in it was a run-of-the-mill shot of patrons sitting in a diner. What intrigued me about that image was that it had gotten the exact same number of likes as every one of my self-portraits.
This similarity in the number of likes set off the lightbulb in my head. It was then that I realized that I was seeing the algorithm in action:
{if image contains person, then boost the visibility of image}
So there we are, a basic confirmation that virtually any image with a person in it will get shown to more users and, therefore, garner more likes. But looking a bit deeper into my feed and the number of likes images with people in them received, I noticed something else interesting. Back in April, I posted an image that contained a person, and that image received twice the number of likes that my usual people-based images received. On the surface, I thought I understood why — in the image description, I mentioned that the photo had just won a best-in-show award at my university’s annual student art show. It made sense:
{if image description contains celebratory language, then boost the visibility of image}
Digging Deeper into the Algorithm
That second bit of algorithm logic made sense until I dug still deeper into my feed (bear with me here; we’re almost to the fun part where I invented a fake Brooklyn baker to prove a hypothesis)…
The last time I made a post with celebratory language was back in February when I announced that my first monograph had just shipped. Although there were no faces in the image, I expected the post to receive lots of likes, given the celebratory reason for the post. Surprisingly, the post didn’t get much attention. My experience with this post put a dent in my celebratory language hypothesis. It was then that I decided to conduct my experiment.
Enter The Baker From Brooklyn
With a pair of algorithmic findings on what did and didn’t get attention on Instagram, I decided to test a new hypothesis: that images with a person in them and a celebratory message in the description would be my winning combination. The hypothesis made, it was now time to test it:
With my wife gone for four days, I had extra time on my hands. Instead of spending my Saturday doing chores, I decided that the best use of my time would be to test my hypothesis by spending hours creating a fake person, a fake bakery, and a fake award and posting this fakery as an Instagram post.
Of course, as a native New Yorker who loves a good crusty bread, my fake person needed to be a Brooklyn baker. I then went to work:
- Use Dal-E to create a selfie image of a baker in Brooklyn
- Use image editing software to style the image and to remove any mention of OpenAI from the metadata
- Use ChatGPT to create a list of 10 hipster bakery names (I went with Flour & Fauna)
- Use ChatGPT to write a celebratory post announcing that my (fake) bakery had been voted the best new bakery in Brooklyn.
Having gone through the process of creating my fake baker and bakery, I posted it all to Instagram and waited to see what would happen.
The Results
Within 30 minutes of posting the image of the fake baker, the likes started to come in. They steadily increased throughout the day, and by evening, my fake baker post had received 25% more likes than my most recent self-portrait. Interestingly, only three people left comments, where they questioned when I had moved to Brooklyn and why I had opened a bakery.
With the numbers in, I confirmed my hypothesis that images with clearly discernible people in them would get boosted in the IG algorithm, and if the description contained celebratory language, it got boosted even further. Logically, this all makes sense, and I came away from my little experiment not with a sense of discovery but rather with the feeling that it was fun to have the opportunity to watch the IG algorithm in action.
Caveats and Conclusions
As someone who spent his career working with data, I would be remiss if I didn’t note that my test was flawed because I tested two hypotheses at once (a no-no), that the dataset in my experiment was extremely small, and that the findings are far from conclusive. That said, in my past life as a product manager, some of my best work analyzing data came from small data anomalies that piqued my curiosity and sent me down the rabbit hole.
However, the most important finding from the experiment was that using AI to create a fake baker, a fake Brooklyn bakery, and conducting a little experiment on Instagram was far more fun than spending a Saturday cleaning the house.
Unless otherwise noted, all images Copyright Lawrence Lazare
Live View features authors from a variety of backgrounds who share their passion for photography. You can meet them all, read their stories, and follow them by visiting the Live View About page.
