image-to-SEO: I Built an AI Tool to Decode Google Discover. Here's What It Found

Ever scrolled through Google Discover and wondered, “Why this article?”

As a content enthusiast, I’ve asked that too. Discover is a powerful, personalized feed—especially on mobile—and a potential goldmine for traffic. But how articles land there remains a mystery.

I had the chance to discuss this topic with many people during the BrightonSEO event.

So, I built a simple AI-powered tool: Google Discover Analyzer. It’s an image-to-SEO AI tool. It’s experimental. The Gemini AI model can detect the screen, image or images, language, headline, published/modified date(if it’s visible on the page) and entities!

The best way to use this tool is to upload a screenshot. It analyzes your article based on the page layout and the visuals you’ve included. It works with Gemini’s multimodal AI model.

I had previously released a Discover analysis tool—this one complements it with additional capabilities.

It’s free, educational, and hosted on Hugging Face Spaces—designed to analyze the characteristics that may align with Discover’s ranking signals.

Access tool here: https://huggingface.co/spaces/metehan777/image-to-google-discover

What Is Google Discover?

Unlike Google Search, Discover is a queryless feed. It proactively surfaces content based on:

Search history
Location
App usage
User interests

Getting featured here means organic reach to a highly engaged audience—especially on Android and Google apps.

⚠️ The Challenge: No Clear Rules

There’s no checklist for Discover. But based on research and Google’s guidelines and SEO experts’ experiences, strong performers often show:

High-quality visuals
Clear, compelling headlines (no clickbait!)
E-E-A-T signals (Experience, Expertise, Authoritativeness, Trust)
Topical relevance or freshness
Good mobile usability

This led me to a question: Can AI reverse-engineer patterns from content that feels Discover-worthy?

Meet Google Discover Analyzer

A lightweight tool using Gemini AI to analyze:

️ Screenshots of article pages (compressed in-browser)
Directly pasted article text

The AI is prompted to assess patterns relevant to Discover performance.

I picked a news appeared on my Discover feed. During that time, I was in London. Based on my search history, I researched a UK visa, created a route to Cambridge. It’s all before seeing this post.

⚙️ How It Works

Input: Upload a screenshot or paste article content
Analyze: AI (Gemini) processes the content using a structured prompt
Output: You get a detailed breakdown and an estimated “Discover score”

What Does the Google Discover Analyzer Do?

This tool evaluates how well an article might perform in Google Discover by simulating how AI (specifically, Gemini’s multimodal model) might interpret and score your content.

You can either upload a screenshot or paste the article text. The tool then generates a Discover score (from 0.00 to 1.00) and a detailed, structured breakdown across multiple dimensions.

Example Breakdown From the Screenshots

Overall Score: `0.75 / 1.00`

This article scored well because:

✅ It’s from a high-authority source (DW – Deutsche Welle)
✅ It covers a timely, internationally relevant topic
✅ It includes a clear and informative main image

Key weaknesses:

⚠️ The article uses a standard news format, which might reduce uniqueness
⚠️ E-E-A-T signals for the author are not visible (no visible byline, credentials, etc.)

What The Tool Analyzes:

1. Content Summary

Headline: "Vize ücretleri bilim insanlarını İngiltere'den soğutuyor"
Publisher: DW (Deutsche Welle)
Core Topic (in English):An analysis of how increasing visa fees in the UK are deterring international scientists and engineers from choosing the country.

2. Topic Identification

Primary category: World News
Subtopics identified:

UK immigration policy

Visa costs
Science and technology policy
Brain drain
International student/worker mobility
UK economy

This structured tagging mimics how Discover may categorize and match content to user interests.

3. Content Structure Analysis

(Found in another tab not shown here, but typically includes):

Headline format (question, list, statement)
Presence and type of visuals
Content length and layout

4. Discover Visibility Patterns

This part assesses:

Timeliness of the topic
Region and language targeting
Alignment with current trends
Topical authority signals (e.g., does the site often write about this subject?)

5. Optimization Recommendations

Based on Gemini’s pattern recognition, the tool suggests:

Making the headline more unique or emotional
Adding author credentials visibly
Improving visual layout or image ALT descriptions

️ Why Screenshot Uploads Work Best

Because Discover is visual-first, uploading a screenshot allows the tool to:

Evaluate layout and design
Detect the main image
Consider visual hierarchy and mobile readability

This is where multimodal AI shines—processing both image and text context together.

⚠️ Important Limitations

It’s AI, Not Google: The analysis is based on Gemini’s pattern recognition—not Google’s algorithm.
Correlation ≠ Causation: Just because a trait is common doesn’t mean it’s causal.
Score Is Experimental: It’s a Gemini-generated number based on one internal prompt, not an official ranking factor.

Best Use? Exploration & Learning

Use it to:

Compare content pieces
Brainstorm optimization ideas
Deepen your understanding of content structure

Don’t treat the score as a prediction—use it as a thinking tool.

Try It Out