How to Actually Measure Image SEO in Search Console
Image work is the easiest SEO to do and the hardest to prove. Where the image numbers hide in Search Console, what they do and do not include, and a reporting format that survives a sceptical client.
Image SEO has an awkward property: the work is cheap and well understood, and proving it worked is neither. Filenames, alt text and metadata are easy to ship. Six weeks later someone asks what it did, and the answer is usually a shrug plus a screenshot of a checklist.
The data exists. It is one dropdown away, and almost nobody has clicked it.
The control everyone misses
In Search Console, open Performance, and look at the row of filter chips at the top. The first is Search type: Web. It is a control, not a label. Change it to Image.
That is the entire feature. Everything else in this article is what to do with it.
Two immediate moves are worth making before anything else:
Set the date range to 16 months and look at the shape. This is your baseline, and it exists retroactively whether or not anyone has been watching.
Use Compare rather than Filter to put Image alongside Web on one chart. The ratio of image impressions to web impressions is the single most useful orientation number you can get in thirty seconds, and it varies by an order of magnitude between sectors. A cabinetmaker and an insurance broker should expect very different answers, and knowing which you are prevents both false alarm and false comfort.
What the numbers mean, and where they lie
The four metrics are the familiar ones, but three of them behave differently under the Image search type.
| Metric | Behaviour in image results | How to read it |
|---|---|---|
| Impressions | The reliable one | Primary signal — did the work reach the index |
| Clicks | Structurally suppressed | Secondary; low is normal, not a fault |
| CTR | Misleadingly low | Do not compare against web CTR |
| Position | Grid position, not a list rank | Directionally useful, weakly precise |
The suppressed click-through rate is not a failure of your images. The image grid shows a large preview in place, so a substantial share of searches are answered without a visit. Low single-digit CTR is the normal state of a healthy image presence, and a report that frames it as underperformance is measuring against the wrong benchmark.
Position needs the same caution. Image results are a grid, not a ranked list, so an average position of 12 does not mean twelfth from the top in the way it would in web results. Treat it as a direction of travel rather than a coordinate.
The dimension that is missing
Under the Image search type, the Pages dimension reports the page URL. Never the image file URL.
This is the constraint that shapes all image reporting. If a page carries eight images, Search Console will tell you that page earned impressions in image search and will not tell you which of the eight was shown. There is no image-level dimension, no image-level inspection tool, and no request-indexing button for a file.
Practical consequences:
- Attributing a result to one specific image requires either a page with a single dominant image, or a manual search to see what is actually displayed.
- Testing a change to one image on a page full of images produces uninterpretable data.
- Any tool or report claiming per-file image ranking data is inferring it from scraping, not reading it from Google.
The workable response is to measure at the page level and treat the page as the unit of image work, which is also how the results surface behaves, since an image result is a link to a page.
A measurement routine that survives scrutiny
Four numbers, recorded monthly, take about ten minutes.
- Image impressions, 28 days, versus the previous 28. Your headline.
- Image impressions as a share of total impressions. Detects whether image growth is real or merely riding overall site growth.
- Top ten pages by image impressions. The Pages dimension with the Image search type applied. This is where you find surprises, and they are usually pleasant ones on pages nobody was optimising.
- Count of pages with any image impressions. Coverage. A rising count means new material is entering the index; a flat count with rising impressions means your existing winners are compounding.
Record these somewhere permanent, because Search Console's own comparison view will not go back far enough to matter once the account is a couple of years old.
The moment this routine becomes indispensable is when the same reporting has to be produced for a dozen clients on the same day. At that point the difference between a documented four-number pull and an improvised dig through the interface is several hours a month, which is exactly the calculation behind standardising the agency image workflow rather than improvising per account.
Segmenting, which is where it gets useful
Site-wide image impressions are a vanity number on any site with more than one kind of page. The interesting question is which section of the site earns image visibility, and that requires a page filter.
The Pages filter accepts a URL-contains match and, on most properties, a regex, so a handful of saved segments turns one chart into a real diagnostic:
- Product or service pages versus editorial pages
- One category or service line against another
- Pages published before an image project against pages published after
That last one is the closest thing to a controlled comparison available here. If images produced under a new process are outperforming the old ones on pages of otherwise similar type and age, that is a considerably stronger claim than a site-wide trend line, and it is the argument most likely to survive a sceptical reader.
Two cautions. Regex support and behaviour differ between the interface and the API, so verify a filter returns what you expect before building a report on it. And Search Console's sampling means very small segments produce noisy numbers, so a segment with a handful of impressions is not evidence of anything in either direction.
Discover, which is the surface people forget
Sites that publish editorial content frequently have a Discover report, and it is worth checking whether yours does, because it appears in the left-hand navigation only when there is data.
Discover placement is heavily influenced by having a large, high-quality image available, which makes it the surface where image quality most directly drives traffic volume. It is also volatile in a way that ordinary search is not: a single article can produce more traffic in two days than a section produces in a quarter, then stop entirely.
The right treatment is to report it separately and never to blend it into an image-SEO trend line, because a Discover spike will swamp everything around it and make a stable underlying picture look chaotic. Where it belongs in an image report is as a note that the same image work feeds it, with its own number alongside rather than inside the main figure.
Reading a drop correctly
A fall in image impressions has a small number of realistic causes, and they are separable.
Web impressions fell too. Then this is not an image story at all. Check the site-wide picture first, always.
Only image impressions fell, sharply, on one date. Something changed at the fetch layer. A migration, a CDN change, a plugin regenerating files at new URLs, a firewall rule. Compare a live image URL against what your sitemap lists. Sharp single-day drops are almost never algorithmic and almost always infrastructural.
Only image impressions fell, gradually. More likely demand or competition. Compare against the same period last year before concluding anything, because image demand is seasonal in several categories.
Coverage count fell. Pages are dropping out of image eligibility, which usually means pages are dropping out of the index. Diagnose the page problem, not the image problem.
When the cause is genuinely at the image layer, the ordered checks are the same ones in the not-showing-in-Google-Images walkthrough, and the fastest of them — fetching the public image URL and reading the response headers — resolves the majority of sharp drops in a minute.
The two surfaces you cannot measure
Honesty here is worth more than a padded report.
Google Lens. There is no Lens filter and no Lens search type. Visual-search visits land inside the Image type indistinguishable from grid clicks. You can reason about whether your imagery suits visual search, and the criteria are reasonably well understood, but you cannot count it.
AI answers and overviews. Image impressions from AI surfaces are folded into existing reports without a label. Nothing isolates them. This is a genuine measurement gap rather than a setting you have missed, and what is known versus guessed about images in AI answers is a short list on the known side.
The correct treatment for both is to report them as unmeasured surfaces that the same underlying work feeds, not to invent a number. A client who is told plainly that a surface cannot be measured trusts the numbers that follow.
Complementary evidence
Search Console is the source of truth for visibility, but two other measurements strengthen an image report and both are free.
Coverage of the basics, from your own crawl. Percentage of images with descriptive filenames, percentage with non-empty alt text, count of oversized files. These are inputs rather than outcomes, but they move immediately, they are entirely under your control, and they demonstrate work delivered in a month where impressions have not yet responded. A single-afternoon site-wide image inventory produces all three.
Largest Contentful Paint on your top templates. Image work and page speed overlap heavily, and LCP responds far faster than image impressions do. In the first six weeks after an image project, it is frequently the only metric that has visibly moved.
Pairing the two makes the reporting honest in both directions: here is what changed immediately, here is what is expected to change slowly, here is the surface we cannot see. If the concern is whether the images being shipped are good enough to be worth measuring at all, the pre-publication checkpoints to run before committing an image are the upstream half of the same problem.
The reporting line that works
For a client or a stakeholder, one paragraph:
Image impressions are up X percent over 28 days, on Y pages with image visibility against Z last quarter. Image search now contributes N percent of total impressions. Clicks from image search are structurally low, in the low single digits, which is normal for this surface. The same work feeds Lens and AI answers, neither of which reports separately, so those are not included in the numbers above.
It is defensible, it names its own limits, and it does not claim a surface it cannot see. That last part is what makes the rest of it believable.
Frequently asked questions
Where do you see image search traffic in Google Search Console?+
Open the Performance report, click the Search type control at the top which defaults to Web, and change it to Image. That single control is the whole feature, and because it silently defaults to Web, most people looking at Search Console have never seen their image numbers at all. Compare mode against Web is the fastest way to size the opportunity.
Why are image search clicks so much lower than impressions?+
Because the image results grid shows a large preview in place, so a great many searches are satisfied without anyone visiting the source page. Click-through rates in the low single digits are normal and are not evidence of a problem. Judge image performance by impressions and by the trend, and treat clicks as a secondary and structurally suppressed number.
Does Search Console show Google Lens traffic separately?+
No. There is no Lens filter, and visual search activity is not broken out as its own search type. Lens-driven visits generally land in the Image search type where they are indistinguishable from grid clicks, so any Lens-specific claim in a report is an inference rather than a measurement and should be labelled that way.
How long before image SEO changes show up in reports?+
Expect six to twelve weeks before a trend is readable, and longer on small sites. Images are crawled by a separate crawler with its own slower cadence, and the data itself lags by two to three days. The most common reporting mistake is declaring failure at four weeks, which is earlier than the pipeline can realistically respond.
Can you see which specific image ranked in Search Console?+
No. The Pages dimension under the Image search type reports the containing page URL, never the image file URL. If a page holds several images you cannot tell from Search Console which one was shown, so attributing a result to a particular file requires either a page with one image or a manual search to confirm.
What is a good number of image impressions?+
There is no absolute benchmark, because image demand varies enormously by sector. A visual category such as food, interiors or physical products can see image impressions rival web impressions, while a professional services site may sit in the low single-digit percentages and be performing perfectly well. Your own baseline over time is the only meaningful comparison.
Should image traffic be a client KPI?+
As a supporting metric, yes. As a primary one, rarely, because the clicks are structurally low and the surface is not where most commercial intent resolves. The defensible framing is that image work protects page performance, feeds visual and AI surfaces, and reduces accessibility and legal risk, with image impressions as the evidence that the work reached the index.
Does Search Console show images that appear in AI overviews?+
Not as a distinct category. Impressions from AI surfaces are folded into the existing reports without a separate label, so there is no way to isolate them. Any breakdown that claims to quantify AI-answer image visibility is estimating, and the honest reporting position is that the surface exists and cannot currently be measured directly.
Let SEOpix handle the metadata
Filenames, alt text, EXIF fields and GPS coordinates written automatically as each image is generated. Start with 10 free images a month — no credit card required.
