back to thedrum.com
Does having relevant images on your page impact your chance of ranking? Reboot conducted an SEO experiment to find out if image relevance on landing pages has any impact on organic search performance.
Previously, we found unique images correlated with better rankings with our long-term duplicate image experiment. However, this raised a new question within our SEO agency: does image relevancy itself influence rankings?
Helpful content has long been critical for ranking highly and, as images are integral to most pages, we hypothesised image relevancy signals may impact perceived page relevance.
Search engines aim to rank pages highly that are relevant to user intent and satisfy their query. So if page-level relevance matters, perhaps image relevance does too.
In summary, our core hypothesis was:
The relevance of images on a landing page influences its organic search rankings. Pages with images highly relevant to the topic may outrank pages with less relevant images.
We devised an experiment to isolate image relevancy as a ranking factor and test this hypothesis. By controlling other variables, we hoped to determine if image relevance alone impacts organic search performance.
The results could reveal new insights into how images factor into search rankings and page relevance signals. Our goal was to inform SEO best practices around optimising images and landing page content.
1. Found and targeted a keyword that our experiment sites could rank for without any historical or external signals, such as links and little existing content.
2. Made sure Google could identify that the images on one half of the experiment sites were more relevant to the target keyword than the images on the other half.
3. Minimised outside factors that could impact the rankings of our experiment sites and make our results less reliable.
Choosing a Target Keyword
For our experiment, we needed to select a specific target keyword to build and optimise pages around.
We focused on "cats vs dogs" to leverage obvious visual differences between the two. Specifically, we chose the Dogue de Bordeaux breed since we had unique dog images readily available.
The keyword had to be low competition, so new pages with no external signals could rank. But it still needed enough search volume for rankings to be tracked.
After research, we landed on "Dogue de Bordeaux characteristics" as an ideal target keyword.
Why We Chose This Keyword:
- By optimising identical test pages around this precise keyword, we could better isolate image relevance as a ranking factor.
We needed to classify images as more or less relevant to our target keyword. This way, we could test if Google uses image relevance in rankings.
We utilised Google's Vision API to analyse 500+ unique dog and cat images. The API uses machine learning to understand image contents.
Our criteria:
We ran the images through the API using a custom Python script and extracted the results. The API provided extensive data on each image's contents and subjects.
An example of the results from Google’s Vision API after feeding it Dogue de Bordeaux images.
We shortlisted images where Google seemed highly confident they contained either Dogue de Bordeaux dogs or cats. This gave us a collection of both clearly relevant and irrelevant images.
In total, we published 20 near-identical sites:
By controlling other factors, we aimed to isolate image relevance as the variable to test. If Google uses it in rankings, the dog sites should theoretically outperform the cat sites.
The Vision API gave us confidence Google could detect our tailored image relevancy if utilised. Now we needed to test if it impacted search results.
Isolating image relevance was crucial. Now that was done, we took steps to control other ranking factors. The goal was to neutralise other ranking variables outside image relevance. While no experiment is perfect, this methodology allowed us to better isolate the impact of image relevance alone.
By controlling external factors, we could attribute any meaningful ranking differences to the image relevance variable we intended to test.
Identical Site Templates
All experiment sites matched in terms of site speed, UX, and performance. This was confirmed via Page Speed tests.
Uptime Monitoring
Uptime was consistently tracked to ensure no downtime differences by running frequent checks every day with StatusCake.
Matched On-Page Optimisation
When the content was written by our in-house content marketing team, care was taken to ensure that there was identical keyword placement and density. Content for all sites was similar but not duplicated.
New Domains
We only used brand new domains with no prior links and archive.org history for each of the 20 experiment sites.
Private Domains
Domains were kept private beyond the research team at Reboot to prevent accidental organic clicks.
Unique Hosting
The experiment sites were hosted on separate AWS servers with different IPs to avoid any negative impact on the page performance and speed.
Controlled Indexing
We ran a gradual indexing schedule, alternating dog and cat sites, so we could be sure that the pages were only indexed when we were ready. New IP addresses were used each time.
We found that ranking the websites was challenging, even for this low-competition keyword. This was likely because the websites had little unique, high-quality content outside of the experiment pages and no historical ranking signals.
To encourage Google to rank the pages within the top 250 search results, we took a couple of steps to increase the trustworthiness of the sites. This included adding ‘About Us’ and ‘Contact’ pages to each site. We also created links from a few identical directories to each site, publishing the nearly identical referring content simultaneously.
Because of this, we used Rank Tracker to monitor the experiment site rankings, as this allowed tracking beyond the top 100 results that SEMrush would cover.
Rank Tracker monitored all 20 sites in a single campaign across the top 1,000 rankings. This captured the full performance picture across all of the experiment sites.
After several months of tracking, the data showed no clear correlation between image relevance and rankings.
Our data analysis found no statistical significance between the dog vs. cat site rankings over time. Image relevance did not directly influence rankings based on the target keyword.
This aligns with comments by Google's John Mueller that they do not analyse landing page images for web search relevancy.
However, our findings suggest image relevance should not be an SEO priority over other optimisation factors, and we still recommend using relevant images to improve user experience. Relevant, engaging and high-quality will only help your site visitors.
Additional Findings
Although image relevancy may not have directly impacted search engine rankings, site command searches did surface more dog sites than cat sites. This indicates image relevance may be a factor specifically for image search, though not standard web search.
We also observed the dog images dominating the first page for related image searches.
In summary, while image relevance did not impact web search rankings in our experiment, the additional findings suggest some relevance influence may exist for image-related searches.
This experiment provided tangible data on the impact of image relevance on organic search rankings. While more research is needed, the results suggest image relevance itself should not be an SEO priority over other on-page factors.
However, using relevant, high-quality images likely improves user experience. As search algorithms evolve, image relevance may become a bigger ranking consideration as well.
