Here is the verdict up front: image SEO still matters, but not all of it matters the same way. Accessible alt text, good page context, and images that a crawler can actually find and load all have documented roles in how Google understands and shows your content. Descriptive filenames offer only light clues. Rights metadata embedded in the file mostly supports attribution and licensing features, not a proven boost to your rankings or your odds of getting cited in an AI answer. Grade this one confirmed for the mechanics Google documents, and weak or unproven for the idea that renaming files or stuffing EXIF fields will move your traffic.
That split matters because "image SEO" gets used as a catch-all for several different jobs: making an image discoverable, explaining what it shows, giving people who use screen readers a real alternative, making it load well, and declaring who owns it. These are not the same task, and treating them as one checklist is how teams end up spending an afternoon renaming a photo library instead of fixing the thing that would actually help a reader. Most of the confusion is really an alt text SEO confusion wearing a bigger label, since alt text is the one piece of this that people argue about most.
What the claim actually says, and why it keeps coming up
Google's own image guidance, reviewed as of September 2026, separates image discovery, page context, alt text, filenames, technical quality, and metadata into distinct topics. That's a signal in itself. If Google treated all of these as one ranking lever, it wouldn't bother explaining them one at a time.
The confusion is understandable, though. A photo can do well in Google Images while adding almost nothing to how a text-heavy article performs in ordinary search results. A chart can make an article genuinely more useful to the person reading it without creating any measurable image-search traffic at all. Neither of those outcomes tells you anything about whether changing a file's embedded metadata would make that article more likely to get cited by an AI system. Three different outcomes, three different sets of evidence, and a lot of advice online blurs them into one.
The evidence for: what actually helps
Alt text does real work, for two different reasons
Google Search Central calls alt text the most important attribute for giving it more information about an image. It says it uses that text alongside computer vision and the surrounding page content to understand what the image is showing. That's a real, documented contribution to how Google understands an image. It is not a claim that alt text by itself decides an image's rank, or that adding it raises the ranking of the page it sits on. Google also specifically warns against filling the attribute with keywords.
This is the heart of the alt text SEO question most teams actually mean when they ask if the work still matters.
The W3C's accessibility guidance draws a related but separate line. An informative image needs a text alternative that conveys what the image contributes to the page, not a literal inventory of everything in the frame. A genuinely decorative image, or one whose information is already covered by the surrounding text, can appropriately carry an empty alt attribute so a screen reader skips right past it. That's different from leaving the attribute off entirely, which some screen readers handle by reading the filename out loud instead. An empty alt in the right context is not a defect to fix by cramming in keywords. It's doing what it's supposed to do.
Page context usually matters more than the filename
Google says it pulls an image's subject matter from the page it sits on, including captions and image titles, and recommends placing images near text that is actually relevant to them. On filenames specifically, its language is direct: they provide "very light clues." The example it gives is the difference between my-new-black-kitten.jpg and IMG00023.JPG. That supports naming files sensibly when you create or publish them. It does not support setting aside a week to rename an existing image library in the hope of a real ranking gain, and Google's own wording never puts a number on what a filename actually contributes.
Worth keeping straight, because these get lumped together constantly: the image's filename, the HTML alt attribute, a visible caption or image title, the host page's own title and meta description, and metadata embedded inside the file itself are five separate things. Google does say the host page's content and metadata can have a real influence on how and where an image shows up. That influence flows from the page, not from a filename tweak.
Discovery and a page that actually loads matter more than most teams assume
Standard HTML image elements are what let Google find and process an image at all. It does not index images that are only referenced through CSS. When responsive image attributes are in use, the fallback URL in the img element's src still matters. Image sitemaps can surface images Google might otherwise miss, including some reached through JavaScript, but they're a discovery aid, not a ranking guarantee.
On loading, Google's lazy-loading guidance says relevant content needs to load when it comes into the viewport, without requiring an interaction Google doesn't perform, like a click. It also recommends sharp, high-resolution images where they're warranted, because they look better in result thumbnails, while noting in the same breath that images can dominate a page's weight and slow it down or make it more expensive to serve. The real decision is a balance: keep the image genuinely useful and the page fast, rather than maximizing resolution for its own sake. The same logic applies to stock photography. Whether an image actually adds something to the page is a more defensible question than whether its filename contains a keyword, and none of the sources here measure any general ranking advantage for original photography over stock.
The evidence against: what gets overstated
A few claims travel around content teams as settled fact when the underlying guidance doesn't support them at that strength.
"Every image needs keyword-rich alt text" overstates what's documented. Google does use relevant alt text for understanding an image, and the W3C explicitly calls for empty alt text on images that are appropriately decorative. What isn't supported is the idea that a keyword-filled alternative on every single image improves rankings.
"Renaming every image is a major SEO win" overstates the filename guidance. Google calls filenames very light clues and favors short, descriptive names as ordinary practice. What's missing is any evidence of a material traffic or ranking lift from a bulk-renaming project.
"EXIF keywords or geotags boost rankings" isn't supported either. The image-metadata guidance documents specific uses for rights, credit, and provenance data through IPTC, structured data, and C2PA. It does not establish that an EXIF keyword or a GPS coordinate improves rankings, and the fair way to state this is that the guidance doesn't establish the tactic, not that Google is incapable of reading EXIF data at all.
"A licensing badge increases rankings" conflates two different things. Qualifying metadata can make an image eligible for Google Images' Licensable badge. That's a display feature, not a guaranteed badge or a documented ranking boost.
"Image metadata gets an article cited in AI answers" is the newest version of this pattern, and it's the least supported of the group. Google recommends useful images where they fit and describes ordinary search eligibility for its AI features. Nothing in the guidance ties a separate AI-citation benefit to EXIF data, IPTC fields, a filename, or an alt attribute. This is the part of image SEO relevant to AI answers specifically, and it is the thinnest evidence in the whole set of claims.
The pattern across all five is the same: real documented mechanics get stretched into promised results. A vendor's before-and-after traffic story would need a defined sample, controls for everything else that changed on the page, a specified search surface, and a measurement period before it could actually establish cause and effect. Nothing in the reviewed evidence clears that bar.
What "file metadata" actually covers
This term gets used loosely enough that it's worth separating out what it actually contains.
EXIF is technical, capture-related information commonly attached to image files. It isn't a stand-in for every kind of image metadata, and there's no established EXIF SEO benefit to chase, whatever a keyword or a coordinate in that field might contain.
IPTC photo metadata lives inside the image file and can carry creator, credit, copyright, and licensing information. Google says these fields can travel with the file itself. For the Licensable badge specifically, Google looks for a Web Statement of Rights field through IPTC and recommends a Licensor URL where one is available.
Page-level image structured data describes an image through markup tied to the specific page it appears on, which means it has to be supplied separately for every relevant page instance, unlike embedded IPTC data that rides along with the file. For badge eligibility through structured data, Google wants the image's license property, with acquireLicensePage recommended when it applies. One eligible method is enough on its own. If both IPTC and structured data are present and they disagree, Google uses the structured data. None of this guarantees the metadata actually shows up in results.
Article structured data's image property is a different tool again, one that identifies a representative image for an article and can help Google show better images in search. That's a distinct job from embedded rights fields. Similarly, primaryImageOfPage and og:image are ways to suggest a preferred preview image, though Google says the actual preview selection is automated either way. None of these substitutes for having an informative image on a genuinely relevant page.
C2PA provenance metadata is its own category too. Google says it can extract this kind of detail and may surface it in an "About this image" panel, including whether AI tools were involved in creating or editing the image. That's a possible provenance display, not an announced ranking signal for AI search. Google does note that stripping file metadata can reduce file size, but it recommends keeping important rights and identification information in place, and it flags that removing that information can create legal complications in some places. Stripping everything for a faster page load isn't the free win it might sound like.
One small housekeeping note worth knowing: Google's image-sitemap documentation removed the older caption, geo_location, title, and license sitemap tags. If your process still references those, it's out of date, and they shouldn't be confused with a visible caption or a current licensing field.
Ordinary search, image search, and AI search are three different scoreboards
Google Images and visual discovery are the clearest, most direct use case for everything discussed above: alt text, page context, a usable image URL, and appropriate rights metadata all connect fairly directly to how an image performs there. Even so, actual visibility and clicks still depend heavily on the query and how the results get presented.
Ordinary web results are a different story. Relevant images can improve the reader's experience, and Google's documentation covers page context, representative images, and certain rich-result appearances. What it does not establish is a universal rule that adding alt text, a descriptive filename, or EXIF data makes the article itself rank higher for its main text query. Keep "this image makes the page more useful" and "this image is a confirmed ranking factor for the page" as two separate claims, because the evidence only backs the first one.
AI Overviews and AI Mode add a third scoreboard. Here is the part of image SEO relevant to that surface: Google's guidance says a page has to be indexed and eligible for a regular search snippet before it can be considered as a supporting link in an AI answer. It says ordinary SEO practices still apply, recommends supporting your text with high-quality images where they fit, and says no special AI text file or schema is required. Being eligible is not the same as being included, and nothing in the guidance suggests that image metadata earns you a citation on its own. It's also worth saying plainly that this is Google's stated behavior, not a rule that automatically carries over to every other AI answer engine.
What the numbers actually show
Pew Research Center looked at real browsing data from 900 U.S. adults over March 2025 and found that when a Google result page included an AI summary, people clicked through to a traditional result link in 8% of those visits, compared with 15% on result pages without a summary. In the same analysis, people clicked a link inside the AI summary itself in just 1% of visits where one appeared, based on search-result appearances Pew collected in April 2025. Pew is careful to note that AI summaries change over time, so these numbers describe a snapshot, not a permanent rate.
It's worth being precise about what these figures do and don't tell you. They describe overall click behavior on results pages, not image-search clicks specifically, and they say nothing about the causal effect of any image optimization work. There is no reliable, universal uplift figure anywhere in the reviewed guidance for what alt text, filenames, or metadata edits are worth in rankings, AI citations, or revenue. Nobody should convert the Pew numbers into an image-SEO return estimate; they weren't measuring that.
If you want to see how your own site is actually doing, Google Search Console's Performance report gives you clicks, impressions, click-through rate, and average position, and its search-type filter separates Image results from Web results, with AI-feature appearances folded into the Web type. Compare the pages and queries that matter to you over time, and resist attributing a change to a filename or metadata edit when the article, the query mix, or the search appearance also changed in the same window. Image-search referrals alone won't tell you whether a picture actually helped the reader, and a shift in your Web performance trend won't isolate what the alt text did.
Where to actually spend your time
Protect accessibility and editorial meaning first. Don't trade a truthful alt description, or an appropriately empty one, for keyword repetition. Give the images that matter a real, explained relationship to the surrounding text instead of dropping them in as decoration.
Avoid the preventable failures. An image that's genuinely useful but can't be found or loaded by a crawler never gets the chance to deliver any of the image-search benefits described above. Balance legibility and thumbnail quality against how much weight the image adds to the page.
Use descriptive filenames as routine practice, not a growth initiative. Google's "very light clues" language supports naming files sensibly as you go. It doesn't support a special project to rename a whole library on the promise of moving your core rankings.
Treat rights and provenance metadata as rights and provenance work. It deserves real attention when attribution, licensing, or accurate origin information genuinely matters to your business. It isn't a proven lever for general rankings or AI citations, and selling it internally as one will set the wrong expectation.
Measure the surface you're actually investing in. Keep Image-search performance, Web-search performance, and AI-feature appearances separate in how you look at your data. Put real effort where a site's visual queries, genuinely informative images, or real licensing needs make the payoff credible, rather than applying the same checklist to every blog illustration regardless of whether it's doing any work.
None of this is settled science in every direction. Google doesn't disclose how heavily it weighs alt text against computer vision and page context, none of these sources quantify what a filename is actually worth, none of them establish a traffic return from EXIF edits, and none of them offer a cross-platform rule for how image metadata factors into AI citations anywhere outside Google's own stated behavior. What would change this verdict is a new, explicit statement from a search engine about those specific mechanisms, or controlled, surface-specific evidence of an incremental result. Until then, neither should be assumed.
For a team trying to publish at real volume, the practical read is this: the writing and production work that already goes into a page, keeping keyword coverage, structure, and internal linking built in from the start rather than retrofitted, matters more to your overall search and AI-visibility position than any file-metadata project will. Image hygiene is worth doing well. It isn't the lever some checklists make it out to be. DeepSmith's own writing pipeline handles that groundwork as part of producing each article, so keyword coverage, structure, and internal linking are already in place by the time a draft reaches you. You can see it on your own content with a free trial.
