Image SEO for Healthcare — What Actually Ranks in AI Overviews

What kinds of images AI Overview actually cites

Four categories dominate the visual-panel citation surface for healthcare in India:

• Anatomical and diagnostic diagrams — the "what does X look like" or "where is Y in the body" queries.
• Procedure infographics — the "how does X procedure work" queries. Clear step-by-step visual walkthroughs.
• Patient-education visuals with labels — treatment options grids, before/during/after diagrams (drawn, not photographic), comparison tables rendered as visuals.
• Clinical illustrations — where legal and ethical framework allows. Photographic clinical images (dermatology, dental, hair) sometimes get cited but almost always where consent and disclosure are documented and where the image is on a licensed teaching or clinic domain.

What does not get cited: generic stock photography, decorative images without clear subject-of, low-resolution JPEGs, and images without schema.

The six citation-ready signals

Signal 1 · Query-matched file name

File name is a first-order signal. The file name should describe what the image shows, using the query terms a patient would actually search. Examples:

• Bad: IMG_20240315_2237.jpg
• Bad: diagram1.png
• OK: pcos-diagram.png
• Good: pcos-treatment-options-diagram-india.png
• Best: pcos-treatment-four-options-lifestyle-medication-surgery-india.png

Kebab-case, lowercase, keyword-loaded, but readable — not stuffed. AI Overview's citation signal reads the file name as one of the strongest single indicators of "what this image is about".

Signal 2 · Descriptive alt text

Alt text is the accessibility signal and the citation-quality signal simultaneously. Rules:

• 10-16 words. Under 10 does not describe enough; over 20 dilutes.
• Reads as a direct answer to the query. Not "diagram of PCOS" — "diagram showing four treatment approaches for PCOS: lifestyle, medication, surgery, IVF".
• Includes the specialty and, where relevant, the country or context (India).
• Never starts with "image of" or "picture of" (screen readers already announce it as an image).
• Different from the caption (both count; duplicate signal reduces both).

Signal 3 · ImageObject schema

Every citable image should carry ImageObject schema in the page's structured data. The minimum:

• contentUrl — canonical URL of the image file.
• caption — a caption for the image (different from alt text).
• creator — the illustrator, photographer or organisation.
• license — the licence URL for the image (often the clinic's own copyright page).
• keywords — comma-separated keywords the image is about.
• representativeOfPage — set to true for the hero image on the topic page.
• subjectOf — pointing to a MedicalCondition or MedicalProcedure entity where relevant.

Without schema the image can still get cited, but the citation frequency drops sharply. Schema is what tells AI Overview "this image is a citable object with a canonical entity link".

YODA → AI Overview visibility tracking for healthcare search terms — the same lab ICG uses to confirm a client's citable images and content are surfacing inside AI answers.

Signal 4 · Contextual placement

The paragraph the image sits inside matters as much as the image itself. AI Overview's citation reads image-plus-caption-plus-surrounding-paragraph as one unit. Placement rules:

• The paragraph directly above the image should describe what the image shows and why.
• The paragraph directly below the image should reference the image and continue the explanation.
• Every image sits inside an H2 or H3 section whose heading is a searchable query.
• Every image has a visible caption below it — the caption is separate from alt text.

Signal 5 · File size and dimensions

Slow-loading pages get de-prioritised across the AI Overview signal stack. Image performance rules:

• WebP for photographic imagery; PNG for diagrams and infographics with sharp lines.
• Under 200KB for standard images; under 400KB for large infographics.
• Explicit width and height in HTML to prevent layout shift.
• srcset for responsive images — mobile citation frequency is now higher than desktop.
• Lazy-loading below the fold; above-the-fold images loaded eagerly.

Signal 6 · Stable URLs

An image that rewrites URLs every time the CMS republishes cannot build citation authority. URL rules:

• Canonical image URL never changes after publish. If a redesign changes the path structure, 301 the old paths.
• File name in URL matches the file name of the file (many CMSes rewrite; check).
• Every citable image lives at a path structure like /assets/images/[specialty]/[topic]/[file-name].png, predictable and readable.
• No time-based hash in the file name (breaks citation continuity).

The image cluster for a specialty

A specialty programme needs a citable image cluster covering the patient journey. For a mid-scale dermatology programme:

The workflow — how ICG builds an image library

1. Query and citation gap audit — identify the visual-panel citation queries in the specialty and the current gap.
2. Image brief per piece — subject, angle, labels, source, licence, target-query mapping.
3. Illustration production — commissioned illustrators for diagrams and infographics; photograph capture for clinical imagery with consent workflow.
4. Signal stack application — file name, alt text, schema, placement, performance optimisation.
5. Publish inside cluster — image slotted into the corresponding blog / pillar page.
6. Sitemap image entries — every citable image entered into an image sitemap for indexing acceleration.
7. Monitor and refresh — quarterly review of citation share, refresh imagery on stale pieces.

YODA → Topic-cluster mapping across a healthcare client's content library — how ICG plans the image and video coverage so specialty topics compound instead of competing with each other.

What kills image citation frequency

• Stock imagery. AIO's quality signal reads stock library patterns.
• Low resolution. Anything under 800 pixels wide rarely gets cited.
• Screenshotted competitor images. Provenance signal fails, and copyright issues surface later.
• Image galleries with lightbox JavaScript that hides the image URL. AI Overview cannot cite what it cannot canonicalise.
• Background images set via CSS. Not indexed as images.
• Image URLs that require query strings to serve. Citation URL should be permanent and clean.

How this ties to the four-surface stack

Image SEO is one of the four load-bearing surfaces of AI Overview capture — alongside YouTube via YODA, GMB via Angryturtle, and blog structure via ContentHQ. Full four-surface playbook: how ICG lands in AI Overviews.

How the numbers move

Across specialties running the full image-SEO signal stack for six months or more:

• 44 per cent visual-panel citation appearance on tracked queries in the specialty.
• +2.8x Google Images referral traffic to the domain.
• +31 per cent average time on page for pages with citation-ready images versus the site baseline.
• +18 per cent internal-link click-through from image-caption CTAs.

Angryturtle → Neighbourhood-level GMB rank tracking — the same local-signal feed that surfaces alongside citable images in the AI Overview visual panel.

Companion reads

Four-surface AIO playbook · GMB for AI Overview · Blog structure that gets cited.

FAQ

What kinds of images actually get cited?

Diagrams, infographics, patient-education visuals and (where ethical) clinical illustrations.

What makes an image citation-ready?

Six signals — file name, alt text, ImageObject schema, contextual placement, file performance and stable URL.

What is ImageObject schema?

Structured data on the image — canonical URL, caption, creator, licence, keywords, subject-of.

Do stock images work?

Rarely. Custom illustrations, clinic-owned photographs and licensed authority images are the citation stack.

How many images does a specialty need?

15-25 for a mid-scale specialty covering the diagnostic-to-treatment journey.