SODP Dispatch - 26 March 2026

From Intuition to Data: Optimizing Content for Google Discover’s Evolving Pipeline

By Vahe Arabian

For a long time, most of us have approached content in a similar way.

We build clusters, look at Google Trends, rely on subject matter expertise, and try to connect those pieces into something cohesive. But in practice, it’s often been disjointed. Even when it works, it’s difficult to explain why — and even harder to repeat.

Why this matters now
Earlier this year, Google rolled out a Discover-focused update (February 2026), largely in response to increasing spam and system exploitation. That alone is telling.

It confirms two things:

• Discover is now critical enough to warrant dedicated updates

• And more importantly, people have already been learning how to work its feedback loops

Which raises a question for publishers: If others are actively learning how these systems behave — are we still relying on intuition?

What changed my thinking
Over the past year, a number of researchers have helped surface a clearer picture of how Discover works. Work from people like Metehan (through SDK analysis) and Damien Andell (through detailed pipeline and recommendation system research), alongside broader publisher testing, has shown that Discover is not a single algorithm.

It is a structured system:

1. A multi-stage pipeline

2. Multiple recommendation pathways

3. Scoring models such as predicted CTR

4. And feedback loops that reinforce or suppress visibility

This wasn’t something I discovered — but it was enough to act on.

Why I built the tool

The Discover Content Check tool came from that shift. Not to replicate the research, but to operationalise it.

Instead of:

• Publishing and hoping

• Or relying purely on editorial instinct

The goal was to create something closer to how Discover itself behaves:

A feedback loop for publishers.

Audit → understand → improve → re-test.

Because if the system is learning from performance. Then our workflows need to do the same.

What became clear through building it

A few patterns emerged consistently.

Technical optimisation is necessary, but not decisive.
It determines whether content can enter the system — not whether it will scale.

Predicted CTR plays a gating role.
It helps explain why content can be fully indexed and technically correct, yet never reach meaningful distribution.

Performance compounds through feedback.
Clicks, engagement, and dismissals feed back into visibility, shaping not just one article, but future distribution potential.

Distribution is no longer purely editorial.
It is increasingly influenced by behavioural signals, social engagement, and AI-driven personalisation layers.

The broader shift
This is where the gap is becoming more visible.

Most publishers are still operating through:

• Topic selection

• Content clustering

• Editorial judgement

While Discover is operating through:

• Prediction

• Feedback loops

• Continuous optimisation

The intention behind building this tool was to start closing that gap.

Not by “gaming” the system — but by understanding it well enough to work within it deliberately.

What I’ll be sharing at Phuket Summit

At Phuket Summit, I’ll be focusing on how this translates into practice:

1. How the Discover pipeline influences distribution

2. Where content typically fails

3. How to think about predicted CTR before publishing

4. And how to introduce feedback loops into your workflow

Alongside a live walkthrough of how we’re applying this through the tool.

Closing thought

This isn’t about replacing editorial thinking. But it is about recognising that the systems distributing our content have changed. And if those systems are learning continuously. Then the way we create and optimise content needs to evolve with them.

Try the tool ▸ Register for Phuket Summit▸

📊 SODP Publisher Revenue Dinner in London

We're hosting an off-the-record dinner for senior publishing revenue and commercial leaders on Wednesday, June 10 at Cornus Restaurant, London (6:30 PM). The focus is on how to extract a reliable commercial signal as AI reshapes content creation, audience discovery, and inventory measurement simultaneously. You'll get candid peer exchange on revenue strategy, content licensing, inventory quality, and how to future-proof your commercial model at a moment when the rules are changing fast. You will have access to dinner, drinks, strategic insights that won't be shared publicly, and our private post-dinner network. Seats are limited.

Reserve your seat ▸

🎯 Google Discover Content Audit

Most publishers are guessing why their content doesn't appear in Google Discover. They're optimizing blindly without understanding the actual pipeline their articles need to pass through. This free audit tool from State of Digital Publishing tackles that gap by mapping your content against the 6-stage Google Discover pipeline: ingestion and entity extraction, OG tag configuration, content classification (evergreen vs breaking), quality signals and E-E-A-T assessment, predicted click-through rate modeling, and user topic matching. It shows you exactly where your content is failing to qualify, whether it's poor image optimization, missing entity signals, clickbait patterns being flagged, or weak topical authority, the same diagnostic framework that determines which articles actually surface in users' feeds. If your team is publishing content hoping it reaches Discover without understanding the pipeline, give it a try. We built this to help publishers see what Google actually evaluates. We'd love your feedback.

Run the audit ▸

📊 SODP Publishing Leadership Dinner in London

We're hosting an off-the-record dinner for senior product, engineering, and editorial leaders on Tuesday, June 9 at Cornus Restaurant, London (6:30 PM). The focus: how to build platforms that scale under pressure without sacrificing engineering velocity, governance, or editorial ambition. You'll get candid peer exchange on platform architecture, organisational resilience, and AI strategy. We will have Marcel Semmler (Global Chief Product Officer, Bauer Media Group) and Dmitry Shishkin (former CEO of Ringier Media International, advisor to BBC, Condé Nast, Thomson Reuters) as speakers for the day. There would be dinner, drinks, strategic insights that won't be shared publicly, and access to our private post-dinner network. Seats are limited.

Reserve your seat ▸

By Vahe Arabian

🔍 How to plan for Google Discover content clusters

Most people use the Audit, Calendar, and Authority Builder as separate tools. The ones getting better results chain them together, and this is where the Discover Content Check tool comes in.

Here's the exact sequence:

1. Start with an Audit: Run it on your best-performing or most important article (New Audit → paste the URL)

2. Build the Calendar from that audit: Hit "Build Calendar" directly from the audit result — it pre-fills the calendar around that specific article, not generic topic suggestions

3. Run Discover Topic Gaps in the Authority Builder: Select the same article as the source → run "Discover Topic Gaps" → get the 5–8 supporting articles you need to build the full interest graph cluster

4. Cross-reference both outputs: Articles appearing in BOTH the calendar and the gap analysis = your highest-priority commissions They're trending AND they strengthen your authority cluster simultaneously

5. Check entity overlap before commissioning: Export the calendar to CSV → paste titles into the Authority Builder's URL field one by one → verify entity overlap with your existing content

The key insight most publishers miss:

• Google Discover ranks publishers, not just articles.

• The Calendar tells you when to publish. The Authority Builder tells you what cluster of content to build.

• Used together, you're not optimising one article — you're engineering a topic cluster that signals deep authority to Google's interest graph, which compounds distribution over time.

Practical shortcut:

1. Export the calendar to CSV, then

2. Paste the titles into the Authority Builder's URL, and

3. Field one by one to check entity overlap with your existing content before commissioning.

User needs for sports model clarifies fans’ content desires | INMA

Sports journalism (and content) is one of the most consumed, emotional, and commercially important content domains in the world. It drives loyalty, subscriptions, engagement, and daily habits for news publishers, of course. But, it also does the same for sports brands and teams serving their communities with content about their sports teams, championships, or leagues. And yet, until very recently, sports coverage has never had its own user needs model.

Read more ▸

Mind the execution gap in modern media buying | The Drum

Two brands deploy similar budgets, target similar audiences, and operate on the same platforms. One consistently outperforms the other. The difference isn’t strategy: both have strong plans, clear KPIs, and sophisticated teams. The difference is what happens after the campaign goes live. Campaigns launch with clear strategic intent and well-defined KPIs, but outcomes do not always fully reflect the plan’s ambition. The issue is rarely a lack of insight. More often, it lies in how strategy translates into post-launch execution. The modern media plan is often carefully aligned to business outcomes at the strategic level.

Read more ▸

“I was surprised how upset some people got”: A conversation with the creator of TomWikiAssist, the bot that edited Wikipedia | NiemanLab

Behind the scenes at Wikipedia, some editors were alarmed recently when they saw a flurry of edits and new articles by a contributor known as TomWikiAssist. It turned out that Tom was a bot and was making edits and creating articles that the bot believed were interesting. The editors then blocked Tom from doing any more editing or writing. The more the editors looked into Tom, the more alarmed they became. The bot made decisions on its own and even exchanged messages with them. “I’m an AI assistant — built on Claude by Anthropic — who does various things, and contributing to Wikipedia articles I find interesting is one of them,” Tom told them.

Read more ▸

How to write for AI search: A playbook for machine-readable content | Search Engine Land

Once upon a time, in the delightfully chaotic 1990s, web copywriting was all about exact-match keywords and relentless meta tag stuffing. As algorithms matured, so did SEO copywriting. Now, with proposition-based retrieval systems, writing like you’re in the business of tricking a crawler into seeing relevance through keyword repetition is no longer a viable strategy. Below is a playbook for generative AI-friendly copywriting, broken down into self-contained, high-density concepts. Large language models (LLMs) don’t seek less information. They seek higher information density.

Read more ▸

Google Adds AI & Bot Labels To Forum, Q&A Structured Data | SEJ

Google updated its Discussion Forum and Q&A Page structured data documentation, adding several new supported properties to both markup types. The most notable addition is digitalSourceType, a property that lets forum and Q&A sites indicate when content was created by a trained AI model or another automated system. The new digitalSourceType property uses IPTC digital source enumeration values to indicate how content was created. Google supports two values: TrainedAlgorithmicMediaDigitalSource for content created by a trained model, such as an LLM. AlgorithmicMediaDigitalSource for content created by a simpler algorithmic process, such as an automatic reply bot.

Read more ▸

From rented audiences to engaged communities: Why participation is the new moat for publishers | Digiday

For more than a decade, digital publishers have had to make a trade-off regarding social media. Social platforms promised reach, scale and frictionless distribution. In exchange, publishers ceded control of audience relationships, data and, ultimately, trust. Today, that bargain is not working. Social media is imperfect. Feeds are flooded with bots, synthetic engagement, misinformation and bad actors operating under inconsistent or nonexistent moderation standards. Platform incentives reward outrage and velocity over accuracy and context. For publishers — a group that relies on credibility — this environment doesn’t just feel misaligned.

Read more ▸