What Science Says the "Perfect" Instagram Image Looks Like

Hi there!

A different kind of issue today — we’re forgoing the usual daily news brief to present fascinating research on the scientific connection between Instagram image composition and the amount of engagement that image will get.

 After all, if there's one thing digital marketers are looking for, it's the perfect formula: A cheat sheet that we can rely on to generate better results.

Take, for instance, engagement. We know there are some things social media managers can do in a post to juice engagement — asking for it, for one, in the form of "Tell us what you think below."

Engagement's important — not only for the social proof, but also that as social platforms like Facebook continue to remove interests, we rely on Likes and Comments and Shares as indicators of interest: proxies and data points that we can use to move people from a prospecting ad set into mid-funnel.

At most organizations, the one thing content managers sweat over are the images — especially on Instagram.

So, is there a formula for the composition of the Instagram images we create? One that, if followed, would get us more Likes?

That's what Gijs Overgoor, Assistant Professor of Marketing at the Rochester Institute of Technology, set out to discover. He and his colleagues analyzed more than 150,000 images from 633 brands across 27 different industries.

I spoke with him yesterday. The transcript of our interview is below but if you’d like to listen to it, it’s up now at https://todayindigital.com

In the research space there're really two opposing views on image creation. One faction thinks that simple images work better. The other says more complex images. Can you walk us briefly through that debate?

This debate traditionally comes from advertising where we have this one stream of research that say, "Remove all the clutter, make it as simple as possible, straight to the point so that the consumer knows what you're talking about. And that way, you can capture the attention and capture the interest and engagement of a consumer."

And then, the other stream of research says, "Well, what if we clutter our advertising? What if we may fill it up with a lot of complexity and a lot of different types of content, and that really makes them stop and look a little bit closer and draws the consumer in that way."

We picked up on these two streams of research and we were like what does that mean? It can't be that they're both right or maybe they are both right. And then, what does this look like on social media?

I want to get to the results in a moment, but can you talk me through your process? How was this done? Was it done manually or an automated tool?

We opted for algorithms. So the two streams of research that we just discussed, they went for the manual route. So they had a couple of grad students code, a bunch of advertisements, or they had specific trained judges go through these type of content and judge them manually on different types of aspects related to complexity. And we said, "Well, that is subjective. So what if we take algorithms and construct a more objective view of this matter?" So we set out to construct these automatically. And once we have algorithms, we can basically study as many as we want all at once.

I know some of the things you were looking for were things like how symmetrical an image was, whether there was a human face. Can you walk us through what else you looked for?

We started with this concept that is called visual complexity. And visual complexity basically captures any kind of intricacies and detail within an image related to several different aspects. And what we found is that we can split this up into two categories in the terms of images. So one of these is what we call the feature complexity.

And the feature complexity encapsulates all the inherent variation within an image. So really on like a pixel level, pixel to pixel basis, how much variation is there between these pixels. So you can think of color, you can think of how much detail and contrast there is within an image or how much variation there is between bright and not bright. So that is feature complexity on the one hand, so that you analyze pixels specifically and variation between pixels.

Then on the other hand, we have complexity in terms of design. So, the complexity in terms of design has everything to do with objects in an image. So the story of an image. Where are these objects located? And are they, for example, very scattered across this image? Are they asymmetrically structured, and how many objects are there? And then, the more objects there are and the more irregular they are, the higher the design complexity.

Okay. Well, let's start with the first one. What did you find was the optimum level of feature complexity, that being the pixel elements?

We found that the optimal level is somewhere in the mid regions. And specifically we find that we can find this optimum for each of the three features. So the color complexity, the luminance or brightness complexity, and then the edge density, or the amount of detail in an image, and those were in the mid region. So, you need a little bit, and this is what we found also in the opposing research is that you need enough to engage the senses to capture the eye. But it can't be too much that overwhelms the brain in terms of processing and the way that we want to. The brain decides whether or not we're going to look at it or not.

See, that's interesting because when you say mid, that surprised me because in my own brain, having been in the ad business for almost 30 years now, the things that I resonate with from an advertiser creative type point of view are arresting images. Almost jarring. These days sometimes we refer to it as thumb stopping. And those tend to be images, not in the middle. They tend to be visually either quite stark or lots of black through lots of high contrast. But you're saying that those don't perform as well.

You are certainly right. First of all, there's personal preferences like there always is. And we try to capture across the whole space. But when you look at these stark differences, that does mean that there might be a lot of detail in middle of the picture, but then a little bit less detail around it. Or there might be a lot of color in the focus area and a little bit less around it. And we actually also controlled for these photography elements. And even after we added all those elements, these color complexity, or the feature complexity in general, the optimal state, still in the middle. So I think we're still on the same page here in terms of expectations. Yeah.

Okay. So that's feature complexity. Looking at the individual pixel something that like a Photoshop nerd would be interested in. Let's talk about the art director's brain here. What did you find was the optimum level of design complexity?

There, we actually find the exact opposite where when we talk about the feature complexity, we say in inverted U-shape, so there's an optimal point in the middle. Now we have an area that's like a canyon, right? Where the tops are on both ends of spectrum. And there we see that it's either a single object or a very symmetrical and regular arrangement of objects within an image. So there is a clear focus if you will, and a clear story that there is to tell, or we go to the other end of the spectrum, where there's a lot of variation in a lot different objects and perhaps a lot of creativity, right? And those are the two optimals there. Where the middle is kind of misses the Mark a little bit where it's like, it's not necessarily clear what the story is about, but it also doesn't have the engaging or the gluing qualities of something very creative.

Did you find any ideal positioning or representation of a human face in these high performing images?

We did not look at where the face was located specifically. We did find... We added an indicator of whether a face or not was in there. And we know from previous research that faces are actually engaging. But across all the brands that we had and all the images that we looked at, we actually found a slight negative impact on the engagement, but this was very minor. So, that could have to do a little bit with the distribution across brands. So I would be cautious with interpreting that result. And it was not the main focus of our study either.

Right. And those brands that you mentioned, there were 633, I think of them, and they were big brands, right? Do you think that there would be a difference in how images from small businesses might perform?

Yes. I think it will be wise for any brand using these kind of metrics and using this framework to figure out what the optimum is for their brands and for their audience, right? And you know that very well is that you need to study your audience and you need to figure out what works best for you. And that can be a certain color palette, that can be a very coherent story. So you need to find the optimum there. We focused on very broadly identifying a general kind of statistic that works well. But obviously, there are nuances for big brands versus small brands that we have not looked into specifically, but that we definitely added into the recommendation section of our paper to look at specifically.

The dataset that you got from Instagram or images from 2015 and 2016, do you think there's a chance that your findings would be different if you'd have studied more recent images?

I would like to say no, but I'm obviously not sure. And it's something that as a field in academics of marketing we struggle with a little bit. Sometimes these review processes for the papers are a little bit lengthy. So then ones a paper comes out. You're like, "Well, the type of content got a lot more dynamic over time, whereas when we studied it, it was one single image and that was it." Now we have Carousels, we have Reels, we have stories, we have videos. So, it would be interesting to see what kind of effects hold there. And because we have all those type of content mixed in with our images, it would be interesting to see what that does with the visual complexity. But we would like to hope, or at least we hope that it stays like that throughout, right?

What made you want to study this?

That's a great question, actually. So, when I started, it started out as a project for my master's thesis. And really what it came from was me coming from a fairly technical background, hence the algorithms, right? To see if we could somehow predict the number of likes on Instagram. And we found that we could actually use these computer vision type models, some deep learning to make predictions about whether or not picture A would do better than picture B. And I've explored that into kind of almost like a product or an app for a B testing that we can just say, "You should go for picture A or picture B."

But that from a marketing perspective is not necessarily super interesting because it doesn't tell us why. And as a creative or in general as marketers, we want to know, well, why is picture A better than picture B? And that's when we started looking for these frameworks and we came across these opposing views of visual complexity which we thought we could tap into. And then we found a stream of computer science literature that already studied, how can we measure visual complexity? So those two combined formed a perfect pathway to an interesting study.

With all of these findings in mind, if you were to design the perfect Instagram image for a brand to use, what would that image look like?

Could you specify the brand or what kind of setting we're looking at? That gives me a little bit more to work with.

Sure. Let's imagine that it's a small business, maybe two or three employees and they sell candles.

In this case, the feature complexity is in that sense pretty simple because a candle and, and the lighted candle or the fire, those are not too high in terms of detail or in terms of colors, right? So it wouldn't be too difficult to make that, make that candle very focused and very bright. And the rest of the surroundings a little bit less bright so that we have this optimal region of feature complexity. And then, because we're looking at a specific candle, I think it makes most sense to be on the simple end of the spectrum for design complexity, where we really take this angle or the candle, and as the main focus in a very symmetrical display so that our customers know that this is our product and this is what we're talking about.

What about a B2B type business? So, not an organization that sells a product, but maybe a large company that is trying to attract new employees.

That gets a lot more difficult. For the feature complexity, the recommendation stays the same, right? It stays where you would take a picture and the feature complexity arises, and then you can determine on what end of the spectrum that is. And then you can manipulate it with Photoshop or with a simple filter to get it more towards the optimum. Okay. But then for the design complexity, that's a tough question because we will have to think about what we want to display on the image. And then, I can imagine that we want to do something with an experience of perhaps the service or a happy customer, right? Something along those slides. But this is me perhaps getting a little too creative or not creative enough.

That's okay. We only have another 712 more industry examples to get through (laughs). With all of that said, is the image the biggest driver of likes, or are there other factors like the account size or something that is more responsible for driving engagement?

The biggest factor by far is the number of followers. And the number of followers drives the size of how many people can view it at the first instance. And then obviously, there are a lot of different types of content that go viral for some reason, right? But very often that is like an odd bird out, or not like a weird reason. But generally, if I post the same picture as Kim Kardashian, right? Chances are that Kim Kardashian will get a couple more likes than I do.

She'll get more net likes, but will she have a higher engagement rate like a per capita kind of measurement?

Probably, she will have less of an engagement rate because what we see with people with less followers where the followers are more closely related to the person, you see higher engagement rates, right? And then there's a lot of other factors, like when was it posted? I think sprout social is working or did a piece on when to post recently. Yeah, there's a lot of other factors. And then obviously, the image is an element and because visual content is the social object that drives Instagram, it is important, but there is other contextual variables that matter more in that sense, yeah.

Well, it's very interesting research. I think that social media content managers, especially people who are getting into the business for the first time are often just need a guiding light to start them off. And I think this did a great job of helping people find that. So, thank you so much for your time.

Thank you for having me. That was a great discussion.