Can AI evaluate glance readability of text within images?

Artificial intelligence can now recognise objects, detect text, analyse images and even generate visual content. But can it determine whether shoppers can actually read the text within an image at a glance?

The answer depends on what the AI is measuring.

Many AI tools like Microsoft Azure “computer vision” can identify text in an image using Optical Character Recognition (OCR). Some can estimate whether text appears readable based on visual characteristics.

However, evaluating glance readability requires something fundamentally different. It requires predicting whether people can successfully read critical information within a 300 millisecond glance on a mobile screen under realistic viewing conditions.

This distinction is particularly important in digital commerce.

Today’s shoppers rarely pause to study product images in detail. Instead, they scroll, scan and compare products rapidly, often viewing dozens of listings within a matter of seconds. Hero images, promotional banners and product thumbnails are frequently seen on mobile devices where screen space is limited and every pixel matters.

In this environment, success is no longer determined simply by whether an image attracts attention. The more important question is whether the shopper can immediately understand the information that matters most.

• Can they identify the product?
• Can they distinguish between variants?
• Can they recognise the key benefit?
• Can they understand the message before they continue scrolling?

These are the questions that glance readability seeks to answer.

As digital commerce becomes increasingly mobile-first, brands are beginning to move beyond subjective design reviews towards objective methods of evaluating whether critical information can actually be understood at a glance. AI has an important role to play in this transition, but only when it is grounded in validated image analysis and scientific measurement rather than opinion alone.

This article explores how AI evaluates glance readability, why traditional readability metrics are unsuitable for ecommerce images, and how RHINO AI from Neem combines computer vision, OCR, APCA contrast analysis and the Cambridge Contrast and Size Tests (CamCAST) to provide an evidence-based assessment of glance readability.

Why Traditional Readability Measures Do Not Apply to Digital Commerce Images

The word readability has traditionally been associated with written language.

Conventional readability assessments evaluate continuous passages of text, measuring factors such as sentence length, vocabulary complexity and reading difficulty. Their purpose is to estimate the level of education or reading ability required to understand a document.

This approach works well for books, reports and long-form content.

It does not work for digital commerce images.

Product imagery rarely contains paragraphs of text. Instead, it typically includes only a handful of carefully selected words designed to communicate essential information as quickly as possible.

These may include:

• Product type
  
• Variant
  
• Size or pack count
  
• Quantity
  
• Key benefit
  
• Usage occasion
  
• Promotional claim

Rather than reading continuously, shoppers are expected to understand these messages almost instantly while scanning product listings.

This changes the problem completely.

The challenge is no longer determining whether someone can read a paragraph.

The challenge is determining whether someone can identify a few critical words before moving on to the next product.

This is why glance readability has emerged as an important area of measurement within digital commerce.

Instead of asking:

“Is this text easy to read?”

Brands are increasingly asking:

“Can shoppers read this information quickly enough for it to influence their purchasing decision?”

That subtle shift changes how readability should be measured.

It requires consideration of factors that conventional readability formulas were never designed to assess, including:

• Character size
  
• Font weight
  
• Foreground and background colour contrast
  
• Uppercase, lowercase and numerical characters
  
• Display size on mobile devices
  
• Viewing conditions at thumbnail scale
  

These characteristics determine whether text remains readable when a product image is reduced to the small dimensions commonly used on digital commerce websites, marketplaces and retailer apps.

As Oliver Bradley’s work with ‘Rhino AI by Neem’ highlights, many assets are designed and approved on large desktop monitors before being viewed by shoppers on smartphone screens. Text that appears perfectly readable during the design process can become significantly harder to read when displayed as a mobile thumbnail.

Without objective measurement, these issues often remain unnoticed until after publication.

What Is Glance Readability?

Glance readability refers to the ability of a visual asset to communicate critical information within a brief visual glance.

Unlike traditional readability, which evaluates continuous text, glance readability focuses on whether essential words within an image can be understood rapidly under realistic viewing conditions.

In digital commerce, this is particularly important because product images often perform the role that packaging performs in physical retail.

A shopper browsing an online marketplace may only see a small thumbnail before deciding whether to investigate a product further. If critical information cannot be read at that moment, the opportunity to influence the purchasing decision may already have been lost.

For this reason, glance readability has become an increasingly important consideration for:

• Hero images
  
• PDP Carousel Secondary image
  
• A+ Enhanced Content banners
  
• Product listing images
  
• Media Ad banners
  
• Digital shelf assets
  
• Marketplace imagery

The concept also highlights an important distinction between attention and understanding.

An image may successfully attract a shopper’s attention.

That does not necessarily mean the shopper can understand the information it contains.

For example, an eye-catching product image may encourage someone to look towards it, but if the product variant, size or key benefit is presented using text that is too small or lacks sufficient contrast, the shopper may still be unable to extract the information needed to make a confident purchasing decision.

Attention is necessary, but not sufficient.

Understanding what you are being shown is mission critical for success.

This distinction aligns with the broader shopper journey described through Neem’s collaboration with Cambridge University’s Inclusive Design team.

The first challenge is ensuring that shoppers notice the image.

The second is ensuring they can read the information presented within it.

If either stage fails, the effectiveness of the image is reduced.

Glance readability provides a way of objectively evaluating that second stage.

Why Consumer Behaviour Has Changed

The growing importance of glance readability is closely linked to changes in how people consume digital content.

Consumers today are exposed to an extraordinary volume of information across digital commerce platforms, retailer websites, marketplaces and social media.

Faced with so many competing messages, people naturally rely on rapid visual assessment rather than detailed reading.

Instead of carefully analysing every product, shoppers often ask themselves a series of simple questions.

• What is this product?
• Is it relevant to me?
• What benefit does it offer?
• Can I understand it immediately?

If those answers are not obvious, many consumers simply continue scrolling.

This behaviour is particularly noticeable on mobile devices.

Limited screen space means product images are frequently displayed at thumbnail size, while users often browse quickly, compare multiple products simultaneously and switch between applications throughout the shopping journey.

Under these conditions, reading every word becomes impractical.

Scanning becomes the default behaviour.

This is why glance readability has become increasingly relevant within digital commerce.

Rather than expecting shoppers to invest time decoding product imagery, brands must communicate essential information immediately.

Reducing cognitive effort makes it easier for shoppers to understand products quickly.

When information is easy to recognise, shoppers can compare products more efficiently and make decisions with greater confidence.

This shift represents an important change in how visual content should be evaluated.

Historically, creative reviews often focused on branding, aesthetics and visual appeal.

Increasingly, they must also consider whether the information shoppers need can actually be understood during the brief moment they spend looking at the image.

Glance readability provides a measurable way of assessing that capability.

Why Measuring Glance Readability Matters

For many years, brands measured digital content using metrics such as impressions, click-through rates, engagement and visual attention.

These measures remain valuable.

However, they do not answer a critical question.

Can shoppers actually read the information that influences purchasing decisions?

An image may generate attention while still failing to communicate the product benefit, size, flavour, variant or usage information that differentiates one product from another.

This is particularly important for digital commerce hero images, where a relatively small amount of text often carries a disproportionately large amount of commercial value.

Measuring glance readability allows brands to move beyond subjective opinions and evaluate visual assets using objective criteria.

Instead of asking whether an image looks readable (on a laptop screen), organisations can begin assessing whether it is likely to perform under real-world viewing conditions (on a mobile screen in fast scroll).

As AI continues to evolve, this transition from subjective design review to evidence-based image evaluation is becoming an increasingly important part of digital commerce optimisation.

The next step is understanding how AI performs that assessment and why scientifically validated image analysis differs from simply asking a generative AI model whether text appears easy to read.

How AI Evaluates Glance Readability

Not all AI evaluates images in the same way.

Many generative AI tools can identify the text within an image and offer an opinion about whether it appears easy to read. While this may provide a useful starting point, it is not the same as objectively evaluating glance readability.

Evaluating glance readability requires AI to move beyond recognising text. It must assess measurable characteristics that influence whether shoppers are likely to read that text within a 300 millisecond glance.

RHINO AI approaches this challenge by combining computer vision, Optical Character Recognition (OCR), APCA contrast analysis and multidimensional interpolation of experimental data from the Cambridge Contrast and Size Tests (CamCAST). Rather than relying on subjective judgement, the platform evaluates measurable visual characteristics that influence glance readability.

The assessment begins by analysing the image itself.

Step 1: Computer Vision and Optical Character Recognition

Before glance readability can be evaluated, the AI must first understand what is contained within the image.

RHINO uses computer vision together with Optical Character Recognition (OCR) to identify and analyse text elements automatically.

OCR enables the system to detect the presence of text, while computer vision examines the visual characteristics of that text within the wider image.

This automated analysis allows RHINO to identify information that would otherwise require manual inspection across potentially thousands of assets.

During this process, the platform evaluates characteristics including:

• Character size
• Font weight
• Foreground colour
• Background colour
• Whether the text is uppercase, lowercase or numerical

These measurements form the foundation of the glance readability assessment.

Importantly, the system is not simply determining whether text exists within an image. It is measuring the characteristics that influence how easily that text can be read when viewed under realistic ecommerce conditions.

Because this process is automated, brands can assess large image libraries consistently without relying on subjective manual reviews.

Step 2: Measuring Contrast Using APCA

Once the text has been identified, the next stage is evaluating its visual contrast.

Contrast plays a significant role in determining whether text can be distinguished quickly from its background.

If the contrast difference is insufficient, even relatively large text may become difficult to recognise at thumbnail size.

RHINO evaluates contrast using the Advanced Perceptual Contrast Algorithm (APCA).

Rather than relying solely on traditional accessibility measurements, APCA provides a perceptual approach to evaluating contrast based on how people actually perceive differences between foreground and background colours.

Within RHINO, contrast is considered alongside the other visual characteristics measured during image analysis.

This is important because contrast alone does not determine glance readability.

For example, increasing contrast cannot fully compensate for text that is too small, while increasing character size alone cannot overcome poor contrast.

Instead, these characteristics work together.

Evaluating them collectively provides a more realistic indication of likely performance than considering any single factor in isolation.

Step 3: Applying the Cambridge Contrast and Size Tests (CamCAST)

Once the image characteristics have been measured, RHINO evaluates them using the Cambridge Contrast and Size Tests (CamCAST).

Developed through research led by Dr Sam Waller at the University of Cambridge’s Inclusive Design team, CamCAST provides experimental data describing how common words perform under different visual conditions.

The experiments measured the glance readability performance of common three and four-letter words presented for a 300 millisecond glance.

Rather than varying only one characteristic, the experiments examined multiple visual variables, including:

• Character size
  
• Contrast level
  
• Font weight
  
• Uppercase text
  
• Lowercase text
  
• Numerals

By measuring performance across combinations of these characteristics, CamCAST provides experimental evidence describing how changes in visual presentation influence glance readability.

This experimental dataset forms an important part of RHINO’s evaluation process.

Rather than making subjective assumptions about whether text appears readable, RHINO compares the measured characteristics of an image with the experimental evidence contained within CamCAST.

This provides an objective foundation for predicting likely glance readability performance.

Step 4: Multidimensional Interpolation

No two digital commerce images are identical.

Character sizes vary.

Contrast varies.

Font weights vary.

Some text is uppercase.

Some is lowercase.

Some contains numbers.

The exact combination present within an image is unlikely to correspond precisely with one of the individual experimental conditions measured during CamCAST.

This is where multidimensional interpolation becomes important.

RHINO uses multidimensional interpolation of the CamCAST experimental data to estimate glance readability for the specific combination of measured characteristics present within an image.

Rather than relying on opinion, the system predicts performance using the closest available experimental evidence.

This approach enables RHINO to assess a wide range of digital commerce assets while remaining grounded in experimentally measured glance readability data.

The result is an evidence-based prediction rather than a subjective assessment.

Calculating the Potential Glance Inclusion (PGI)

Once the image has been analysed using computer vision, OCR, APCA and CamCAST, RHINO calculates the Potential Glance Inclusion (PGI).

PGI is defined by Dr Sam Waller as an approximate prediction of the percentage of the population who are potentially able to read the text at a glance while wearing the glasses or contact lenses they normally use during everyday life.

The prediction is based on the measured characteristics of the text together with population-based vision data from the Better Design Survey, a UK household postcode-sampled survey that assessed people’s real-world vision ability within their own homes.

Rather than representing ideal laboratory eyesight, the survey reflects the visual capabilities of people in everyday conditions.

This makes PGI particularly useful when evaluating digital content intended for broad consumer audiences.

The score is expressed as a percentage between 0 and 100.

For digital commerce hero images displayed as mobile thumbnails, however, approximately 80% represents a realistic upper limit because of the physical size constraints associated with thumbnail imagery.

Instead of viewing the score simply as a percentage, brands can use it as an objective measurement when comparing different versions of the same asset or identifying images that may require optimisation.

Understanding RHINO’s Performance Levels

To make interpretation simpler, PGIP scores are grouped into performance bands.

For hero images displayed at thumbnail size on mobile devices, RHINO currently uses the following thresholds:

• Level A: around 25% can glance read it
  
• Level AA: around 50% can glance read it
  
• Level AAA: around 75% can glance read it

Level AA represents the target that is typically achievable for most products.

Images achieving Level AA comply with the GS1 definition of a mobile-ready hero image.

Level AAA represents a more ambitious target that is often achievable for full-square images that make use of the available image area.

Secondary images benefit from considerably more available screen space than hero image thumbnails.

As a result, higher glance readability performance is generally achievable, and RHINO therefore applies different performance thresholds to secondary images.

• Level A: 77% to 88%
  
• Level AA: 88% to 94%
  
• Level AAA: over 94%

Images achieving Level AAA for secondary images satisfy the European Accessibility Act’s requirements relating to adequate text size and sufficient contrast for images of text, according to the technical definition contained within EN301549.

These performance levels provide brands with a consistent framework for evaluating image quality and prioritising improvements across large digital asset libraries.

Why Objective Measurement Matters

For many organisations, image approval still relies heavily on design reviews and stakeholder opinion.

Creative teams review assets on large, high-resolution monitors.

Marketing teams approve campaigns based on branding, aesthetics and messaging.

Yet the shopper’s experience is often very different.

Images that appear perfectly readable during desktop review may become difficult to interpret once reduced to the small thumbnail sizes commonly used across digital commerce platforms.

Without objective measurement, these issues can remain hidden until products reach the digital shelf.

By combining computer vision, OCR, APCA, CamCAST and PGIP into a single evaluation process, RHINO enables brands to assess glance readability using consistent, repeatable criteria.

Rather than asking whether text looks readable, organisations can evaluate whether it is likely to be readable under the conditions in which shoppers will actually encounter it.

This shift from subjective opinion to evidence-based evaluation represents one of the most significant developments in digital commerce content optimisation.

The next question, however, is why manual review is often unreliable in the first place, and why even experienced designers can unintentionally overestimate the glance readability of their own work.

The influence of prior knowledge

There is another challenge that is less obvious.

Everyone involved in creating an image already knows what it says.

Designers know the headline.

Brand managers know the product claims.

Marketing teams know the promotional message.

Because the information is already familiar, it becomes difficult to judge how easily someone encountering the image for the first time will be able to read it.

As highlighted in Neem’s research, unconscious visual processes help people recognise familiar words more quickly than unfamiliar ones. This can create the impression that text is easier to read than it actually is for first-time viewers.

The result is an unintended bias during creative reviews.

Teams believe an image communicates clearly because they already know the answer.

The shopper does not.

Why objective evaluation matters

AI does not bring prior knowledge or personal preference to the evaluation process.

It assesses measurable characteristics consistently across every asset using the same criteria.

This objectivity allows organisations to evaluate thousands of product images without variations in judgement between reviewers, agencies or internal teams.

Rather than replacing creative expertise, AI provides an evidence-based layer of quality assurance that supports better design decisions.

Why AI Is Better Than Manual Evaluation

Manual reviews remain an important part of the creative process.

They help ensure that imagery reflects brand identity, communicates the intended message and supports wider marketing objectives.

However, manual review becomes increasingly difficult when organisations manage hundreds or thousands of digital commerce assets across multiple retailers, product categories and international markets.

This is where AI offers significant advantages.

Consistency

Human judgement naturally varies.

Different reviewers may reach different conclusions about the same image depending on their experience, preferences or familiarity with the product.

AI applies the same evaluation process every time, ensuring consistency across large image libraries.

Scalability

Reviewing thousands of product images manually is both time-consuming and resource intensive.

Automated image analysis enables organisations to evaluate large numbers of assets efficiently while maintaining the same assessment criteria throughout.

This makes glance readability evaluation practical at enterprise scale.

Repeatability

Objective measurement makes it possible to compare assets using consistent performance metrics.

Instead of relying on subjective feedback such as “the text looks a little small” or “this version feels clearer”, brands can compare measurable scores generated using the same methodology.

This supports more informed decision-making throughout the content optimisation process.

Evaluation under realistic conditions

One of the most valuable characteristics of RHINO is that it evaluates whether text is likely to perform at the size at which shoppers will actually encounter it.

This allows organisations to assess hero images intended for mobile thumbnails without relying solely on visual inspection.

As digital shelves continue to become more competitive, this capability provides brands with a practical way of identifying readability issues before assets are published.

The Commercial Value of Measuring Glance Readability

Glance readability is not simply a technical measurement.

It has practical implications throughout the ecommerce content lifecycle.

Product images are often responsible for communicating information that cannot be conveyed through packaging photography alone.

A hero image may need to explain:

• Brand
  
• Product type
  
• Variant
  
• Size or pack count
  
• Key product benefit
  
• Usage occasion
  
• Important differentiators
  

If this information cannot be understood quickly, shoppers may struggle to distinguish between similar products or identify the option that best meets their needs.

Objective evaluation helps organisations identify assets that may require improvement before they appear on retailer websites or digital marketplaces.

This creates opportunities to:

• Improve the consistency of product imagery.
  
• Prioritise optimisation across large asset libraries.
  
• Support mobile-first content development.
  
• Reduce subjective debate during creative approval.
  
• Establish measurable quality standards across internal teams and external agencies.
  

As brands increasingly manage digital shelves across multiple channels, objective image evaluation becomes an increasingly valuable component of content governance.

AI and the Future of Digital Commerce Content Optimisation

Artificial intelligence is changing how brands create, review and optimise digital content.

Early applications focused largely on automation and content generation.

Today, AI is increasingly being used to evaluate content quality.

Within digital commerce, this represents an important shift.

Historically, image approval relied heavily on visual judgement and creative opinion.

Increasingly, organisations are seeking measurable evidence that product imagery performs effectively under real-world viewing conditions.

This reflects a broader move from subjective approval towards objective validation.

Rather than asking whether an image looks attractive, brands are beginning to ask more commercially relevant questions.

Can shoppers understand the product quickly?

Can they distinguish between variants?

Can they identify the information that matters before they continue scrolling?

These questions sit at the heart of glance readability.

As digital commerce continues to evolve, AI-powered evaluation is likely to become an increasingly important part of content optimisation workflows, helping organisations complement creative expertise with measurable evidence.

The Bottom Line

AI can identify text within images.

It can analyse visual characteristics.

It can measure contrast.

It can evaluate character size.

But evaluating glance readability requires more than recognising what appears on the screen.

It requires predicting whether shoppers are likely to read critical information during the brief moment they spend looking at an image.

By combining computer vision, Optical Character Recognition (OCR), APCA contrast analysis and multidimensional interpolation of experimental data from the Cambridge Contrast and Size Tests (CamCAST), RHINO provides an objective approach to evaluating glance readability.

This enables organisations to move beyond subjective design reviews and assess product imagery using consistent, evidence-based measurements.

As digital commerce becomes increasingly mobile-first and competition for consumer attention continues to grow, the ability to communicate essential information quickly is becoming an increasingly important aspect of digital content performance.

Glance readability reflects this changing reality.

It recognises that attracting attention is only the first step.

What ultimately matters is whether shoppers can understand the information they need before they move on.

For brands seeking to optimise hero images, product listings and digital shelf content, measuring glance readability offers a practical way to evaluate whether visual assets are likely to communicate effectively under real-world conditions.

Continue exploring the science of glance readability

Understanding how AI evaluates glance readability is only one part of creating high-performing ecommerce content. Explore our related resources to learn how brands measure glance readability, why consumers scan instead of read, how mobile-first design influences product image performance, and how RHINO helps teams optimise hero images and secondary images, using objective, evidence-based evaluation.