Which of Your Keywords Actually Return Images — and How to Find Them
Most keyword research ignores whether a query produces image results at all, which means image effort routinely lands on pages that can never earn image visibility. A method for finding the queries where images are the opportunity.
Standard keyword research asks two questions: how many people search this, and how hard is it to rank. Neither asks whether the query produces images at all.
That omission has a specific cost. Image effort gets spread evenly across a site, including across pages serving queries where an image will never appear in any result, and the return looks disappointing because half the work had no available surface. The fix is not more effort. It is aiming it.
What makes a query visual
The pattern is consistent once you look for it. A query returns images when the answer is an appearance.
Reliably visual:
- Identification — what is this thing, what type of X is this
- Appearance — what does X look like, X colour options
- Comparison of physical objects — X versus Y where both are things you can see
- Ideas and styles — X ideas, X design, X layout
- Outcomes — before and after X, X results
- Diagnostic-by-sight — X damage, worn X, X failure signs
- Instructional stages — how to X, where the steps are physical
Reliably not visual:
- Price, cost, quotes
- Availability, hours, location as a fact
- Definitions and explanations
- Policy, warranty, legal
- Anything answered by a number or a yes
The middle ground is where judgement is needed, and it is worth being sceptical of your own optimism. A query like emergency locksmith near me is commercially valuable and not visual — the searcher wants a phone number, and no image is going to intercept that. The instinct to chase image visibility on your highest-value commercial terms is usually chasing the wrong surface.
Finding them without a tool that does not exist
There is no reliable public metric for image intent. Mainstream keyword tools report web volume and do not separate the surfaces, and where an image-specific figure is offered it is modelled rather than measured. So the method is manual, and twenty queries takes about half an hour.
- Take your existing keyword list — the one you already use for web search — and pull the top twenty by value, not by volume.
- Run each in ordinary web search. Note whether an image block appears, and where on the page.
- Run each in Google Images. Note whether the grid is dense and on-topic, or thin and scattered. A thin grid means Google does not have much and the query is not really visual, which is a genuine signal.
- Note whose images fill the first two rows and which pages they link to. Usually one or two sites dominate a visual niche, and it is often not the sites that dominate the web results.
- Mark each query visual, mixed or textual.
The output is a short list, typically three to eight of the twenty, and those are where image work has an available surface.
Two additions worth making to that list. Take the queries where you already earn image impressions — the Image search type in Search Console shows exactly which pages and it frequently surprises people, since the winners are rarely the pages anyone optimised. And add any query where a competitor's image is currently occupying space you could plausibly take, which the step-four observation gives you free.
Mapping intent to what the image must do
Once a query is confirmed visual, the useful question is what job the image performs, because that determines the shot.
| Query intent | What the searcher wants to see | Image that wins |
|---|---|---|
| Identification | The object, unambiguously | Plain background, single subject, sharp |
| Appearance or options | Variation | A clear set, consistent treatment |
| Comparison | Both things, comparably | Same framing, same lighting, side by side |
| Ideas and styles | Something aspirational | Composed scene, context, atmosphere |
| Outcome | Proof | Genuine before and after, same angle |
| Diagnostic | The problem state | Close, well-lit, defect clearly visible |
The identification row is the one most under-served and most commercially useful, because identification queries are close to action. Someone who does not know what a part is called and photographs it is very close to needing it replaced. That is also precisely the behaviour visual search was built for, and the optimisation criteria for Lens map onto the same plain, well-lit, single-subject shot.
The ideas row is the trap. Those queries have high volume, produce dense grids, and convert poorly, because the searcher is browsing and the large preview satisfies them on the results page. Volume on a visual query is not the same as value.
The constraint everyone forgets
An image result links to a page. Always.
So the ceiling on any image opportunity is the page it sits on. A strong image on a thin page cannot outperform the page, and building a page primarily to host images generally fails on both surfaces at once.
This reorders the work. Do not build for image search. Identify the pages you already need — the ones that already earn or should earn web visibility — and check which of them serve visual queries. Those pages get the image investment. Everything else gets a competent, accessible image and no further attention.
For a service business, the practical difficulty is usually not identifying the queries but having anything to photograph, since the work is often invisible or happens somewhere you cannot photograph. That is a real constraint rather than an excuse, and what to show when there is nothing photogenic is a longer answer than a keyword list can give.
Ecommerce, where this is easier and stranger
Product catalogues have visual queries by default, and the useful research question shifts from which queries are visual to which of my products are searched visually.
The distinguishing factor is whether the buyer knows the name of the thing. Products people can name are searched textually. Products people cannot name — a bracket, a connector, an obsolete part, a fitting — are searched visually and by photograph, and that is where image and visual search do disproportionate work.
Those items are usually the low-priority long tail with the worst photography in the catalogue, which is the arbitrage. The broader case on which product shots move listings applies, with the addition that for unnameable items the plain identification shot is not one image among several — it is the entire opportunity.
A worked example
An abstract method is easy to nod along to, so here is the shape of the output for a hypothetical mid-size business selling and servicing physical equipment.
Twenty existing commercial keywords go in. Running each through the two searches produces:
Textual, drop from image consideration (11 queries). Everything about price, hire rates, opening hours, service contracts, warranty terms and location. These are the highest-value commercial terms on the list, and none of them are image opportunities. Accepting that early is the point of the exercise.
Mixed, low priority (4 queries). Product-category terms where an image block appears intermittently below the fold. Worth having a competent image on the page; not worth a photography brief.
Visual, act on these (5 queries). Two identification queries about parts customers cannot name, one comparison between two similar models, one wear-and-damage diagnostic query, and one installation query where the steps are physical.
The five that survive get specific instructions rather than a general improvement: the identification queries need a plain, single-subject, well-lit shot of the actual part; the comparison query needs both objects photographed identically; the diagnostic query needs a close, honest picture of the failure state, which is the shot nobody wants to publish and the one that answers the question.
That is five briefs on pages that already exist, and it is perhaps a day of work. The contrast with spreading effort across all twenty is the entire argument.
Seasonality and the refresh cycle
Two things make this a periodic exercise rather than a one-off.
Image demand is seasonal in several categories, more sharply than web demand in some. Anything tied to weather, holidays, or a purchase cycle will show a visual query producing a dense grid in one month and a thin one in another, so a list built entirely in a trough will understate the opportunity and one built at a peak will overstate it. Where you know the cycle, check the queries at the point in the year they matter.
The results themselves also change. Google adjusts when and where image blocks appear, and a query that produced no images last year may produce them now, or the reverse. Re-running the twenty-query check annually costs half an hour and occasionally reveals that a whole category has become visual, or has stopped being.
Neither is a reason to monitor continuously. Once a year, plus a check before any significant photography spend, is proportionate.
Confirming before committing
Before spending real effort on a query, three cheap checks.
Is the page indexable and are the images crawlable? Any amount of research is wasted if the file cannot be fetched, and this is a two-minute check that resolves the most common cause of nothing happening. The ordered version is in the not-showing-in-Google-Images walkthrough.
Does the current image on that page actually answer the visual query? Frequently it does not, because it was chosen to decorate the page rather than to answer anything. This is the single most common gap between research and result.
Is there a textual claim tying the image to the query? Machine vision recognises objects; it does not know the object is yours, or that it is the thing this page is about. That connection is made by alt text, by the caption and by the sentence next to the image, which is why writing alt text that works for both readers and search is the last mile of this whole exercise rather than a separate chore.
Expectations
Image opportunity in most sectors is narrower than web opportunity, the click-through is structurally lower, and the reporting is thinner. That is the honest frame, and it argues for concentration: ten to twenty queries, mapped to pages that already matter, done properly.
Where it is worth more than the numbers suggest is that the same work feeds surfaces that do not report separately — visual search, and whatever images are being drawn into AI answers, which is a moving target nobody can currently measure. Being present in the index is the shared entry condition for all of it, and a query list is how you decide where being present is actually worth buying.
Frequently asked questions
How do you find keywords that return image results?+
Search the query yourself and look at what the results page contains, then check whether Google Images returns a dense grid or a thin scattering for the same term. There is no reliable public metric that flags image intent, so manual inspection of twenty representative queries is both the fastest and the most accurate approach available to most sites.
What kinds of queries show images in search results?+
Queries with a visual answer. Identification questions, appearance questions, comparison of physical objects, styles and ideas, before-and-after outcomes, and how-something-looks variants. Queries about price, availability, definitions, policies or opening hours rarely produce meaningful image placement because the answer is textual and images add nothing.
Is image search traffic worth pursuing?+
It is worth pursuing where the query is visual and the visitor is close to acting, and not worth pursuing where the image satisfies the search on the results page. A shopper photographing an object to find where to buy it converts well, while someone browsing inspiration images frequently never clicks anything, so the value is entirely dependent on which intent you are serving.
Do keyword tools show image search volume?+
Not usefully. Mainstream tools report web search volume and do not separate image demand, and where an image-specific figure is offered it is generally modelled rather than measured. Search Console remains the only source of real image-surface data for your own site, though it reports what you already earn rather than what is available.
How can you tell if a competitor is winning image traffic?+
Run the query in Google Images and note whose images fill the first rows and which pages they link to. Repeat across ten or fifteen queries in the category and the pattern is usually obvious, since one or two sites tend to dominate a visual niche. This is observational rather than quantified, and for image results observation is more reliable than any third-party estimate.
Should you build pages specifically to target image search?+
Rarely. An image result links to a page, so the page must stand on its own merits, and a thin page built to host images tends to fail on both surfaces at once. The productive version is identifying pages you already need and making sure their imagery is strong enough to earn the image placement as well.
Does image search intent differ from web search intent?+
Often, for the same words. Someone typing a product name into web search may want a price or a review, while the same words in image search usually indicate they want to see the thing, identify it or compare its appearance. That divergence is why the image on a page should answer the visual version of the question rather than merely decorating the textual answer.
How many image-intent keywords should a small site target?+
Fewer than you would target for web search, because the pool of genuinely visual queries in most sectors is small and the effort per page is higher. Ten to twenty well-chosen queries, mapped to pages that already exist and already matter, is a realistic scope and will outperform a long list treated superficially.
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.
