I’ve been running a WordPress blog in the personal finance niche for three years. Like most publishers, I used a mix of Shutterstock and Unsplash images for every post. It worked fine. Nothing special, but it worked.
Then in February I decided to run an experiment: replace every stock photo on my site with AI-generated images and track what happened over 30 days. New posts would only use AI images going forward, and I had go back and swap out the featured images on my top 30 posts by traffic.
Here is what I found.
Why I Tried This
The motivation wasn’t creative — it was competitive. I noticed that several sites outranking me for key terms had visually distinct imagery. Their featured images looked custom. Mine looked like… well, like stock photos. Because they were.
I’d been reading about Google’s helpful content updates emphasizing originality, and I started wondering whether my recycled Shutterstock images were silently working against me. Not as a ranking penalty, but as a user experience signal — bounce rate, time on page, click-through rate from search results.
I also noticed my Shutterstock subscription had gone up to $49/month for the plan I was on. That’s $588/year for photos I share with half the internet. Worth questioning.
The Setup
I used AI Photo Generator, a realistic ai image generator, to create all the replacement images. I chose it because I needed realistic, photographic-style outputs — not illustrations or digital art. My niche is personal finance, so the visuals need to feel grounded and trustworthy.
What I replaced:
- Featured images on my top 30 posts (by organic traffic from the prior 90 days)
- All images in 8 new posts published during the test month
- Social sharing (og:image) thumbnails for those same posts
What I kept the same:
- Post content, titles, and meta descriptions (no text changes)
- Internal linking structure
- Publishing frequency (2 posts/week, same as before)
- All other plugins, theme, and site speed optimizations
I wanted to isolate the image variable as much as possible.
The Process
Replacing 30 featured images plus creating images for 8 new posts meant generating roughly 50 final images over the month. Here’s what the workflow looked like:
For existing posts: I’d read the post, write a prompt that matched the topic, generate 3-4 variations, pick the best one, optimize it with ShortPixel, and upload it as the new featured image. Average time per post: about 5 minutes.
For new posts: I’d generate the featured image as part of my writing process, right after outlining the post. Having a specific image in mind actually helped me write better introductions — the visual set the tone.
Prompt examples from my actual posts:
- Post about emergency funds: “A glass jar filled with coins and small bills on a kitchen counter, soft natural window light, shallow depth of field, warm tones”
- Post about side hustles: “Person working on a laptop at a small desk in a bright apartment, casual clothes, morning light, coffee cup nearby”
- Post about retirement planning: “Older couple walking on a beach at sunset, seen from behind, relaxed posture, golden hour light”
Every prompt was specific to the post’s content. No more “person pointing at chart” generic stock imagery.
The Results: 30 Days of Data
Here are the numbers, comparing the 30-day test period against the prior 30 days. All data from Google Search Console and Google Analytics (GA4).
Organic Click-Through Rate (Search Console)
This was the most interesting metric.
| Metric | Before | After | Change |
|---|---|---|---|
| Average CTR (top 30 posts) | 3.2% | 3.8% | +18.7% |
| Total clicks (top 30 posts) | 8,420 | 9,890 | +17.5% |
| Impressions (top 30 posts) | 263,100 | 260,200 | -1.1% |
Impressions stayed roughly flat (slight seasonal dip), but clicks went up meaningfully. The CTR improvement was consistent across most of the 30 posts — not driven by one outlier.
My theory: unique, relevant featured images make a better first impression in Google’s image results and in the visual snippets that increasingly appear in search. When your image looks like it was made for the content (because it was), people are more likely to click.
Engagement Metrics (GA4)
| Metric | Before | After | Change |
|---|---|---|---|
| Avg. time on page (top 30) | 2:41 | 2:58 | +10.6% |
| Bounce rate (top 30) | 64.2% | 61.8% | -3.7% |
| Pages per session (site-wide) | 1.6 | 1.7 | +6.3% |
Time on page went up, bounce rate went down. Small changes individually, but moving in the right direction across the board. My guess is that custom images feel more intentional — they signal quality, so readers give the content more of a chance.
Image Search Traffic (Search Console)
This one surprised me.
| Metric | Before | After | Change |
|---|---|---|---|
| Image search clicks | 340 | 580 | +70.6% |
| Image search impressions | 18,400 | 31,200 | +69.6% |
Image search traffic nearly doubled. This makes sense in hindsight: stock photos already exist in Google’s index thousands of times. My AI-generated images were completely new to Google — unique URLs, unique visuals, unique alt text. They got indexed and started appearing in image results for my target queries.
This was traffic I was essentially leaving on the table with stock photos.
Revenue Impact
I monetize with display ads (Mediavine) and affiliate links. More traffic and better engagement meant:
| Metric | Before | After | Change |
|---|---|---|---|
| Ad revenue | $1,840 | $2,120 | +15.2% |
| Affiliate clicks | 1,240 | 1,410 | +13.7% |
Not life-changing numbers, but a meaningful bump — especially considering the only variable I changed was the images.
What Didn’t Work
Not everything was smooth:
Some prompts needed multiple attempts. My post about tax deductions needed an image of documents and a calculator. The first few generations got the calculator wrong — weird button layouts, slightly off proportions. Took me four tries to get something I was happy with. For some topics, stock photos are just easier because the exact thing you need already exists.
Batch replacing was tedious. Going through 30 posts, generating images, optimizing, and re-uploading took me about three hours total over a weekend. Not terrible, but not nothing. If you have hundreds of posts, you’d want to prioritize by traffic and do it in batches.
One image got flagged by a reader. Someone in the comments said the featured image on my retirement planning post “looked AI-generated.” They weren’t hostile about it — just observant. I replied honestly and nobody cared. But it’s worth knowing that some readers will notice, especially as AI literacy increases.
The Cost Comparison
Over the 30-day period:
| Stock (Shutterstock) | AI Generation | |
|---|---|---|
| Monthly cost | $49/month | ~$15 in credits |
| Images created | Limited to plan (25/month) | ~50 images |
| Uniqueness | Shared with millions | 100% unique |
| Relevance to content | Approximate match | Exact match |
I cancelled my Shutterstock subscription at the end of the experiment. That’s $588/year I’m not spending anymore.
What I’d Recommend for Other WordPress Publishers
If you’re considering making the same switch, here’s what I learned:
Start with your highest-traffic posts. Don’t try to replace everything at once. Identify your top 20-30 pages by organic traffic and start there. That’s where better images will have the most impact.
Write specific prompts. “Business meeting” gets you generic results. “Three people reviewing documents at a round table in a modern office, overhead fluorescent lighting, laptops open” gets you something useful. Treat the prompt like you’re briefing a photographer.
Optimize before uploading. AI-generated images are typically 2-5MB. Run them through your existing WordPress image optimization plugin before publishing. Your Core Web Vitals will thank you.
Keep your alt text descriptive. Write alt text that describes what the image shows, not how it was made. “Couple reviewing financial documents at kitchen table” — not “AI-generated image of couple with documents.”
Track your metrics. Set up a simple before/after comparison in Search Console and GA4. Give it at least 30 days. The data will tell you whether it’s working for your specific niche and audience.
Replacing stock photos with images from a realistic ai image generator was one of the highest-ROI changes I’ve made to my WordPress site this year. Better CTR, better engagement, a significant bump in image search traffic, and lower costs.
The stock photo model made sense when creating custom images was expensive and time-consuming. It’s neither of those things anymore. If you’re a WordPress publisher still paying for stock subscriptions, run this experiment yourself. The data might convince you the same way it convinced me.
