Talk to Franz

Workshop · Half a day to 2 days

Google Discover & News Feed Workshop.

A woman with curly hair in a yellow linen shirt at a sunny café table, frozen mid-sip with her espresso cup raised, mouth open in delighted surprise as she looks at her phone
AI-generated image (OpenAI, via Codex) for Full Stack Optimization

Discover is the social-media-ization of news: spiky, viral, driven by clicks. Together with your editors I build your Discover process: the tasks, the owners, the feedback loop.

For

  • Editors-in-chief
  • Editors and desk heads
  • SEO editors
  • Audience teams

Is this you?

  • Discover traffic swings nobody can explain
  • Strong stories that never reach the feed
  • No view of Discover next to Search and News

Discover is the social-media-ization of news

Discover is the feed under the Google search box on phones. Nobody types a query. The feed decides what each reader sees, the way a social network does.

I call it “The social-mediasation of news content.” On my slides it says: “Its a social media algo for news.” And: “Google Discover Optimizes not for Search Queries but for Interaktion.”

This is how I explain the feed to every newsroom:

  • It is all about interaction. “The more clicks it gets, the more impressions it gets.” Readers click, stay, scroll on, click again, and the feed shows the story to more people.
  • It spreads like a social post. “As soon as you interact with an item, as soon as you scroll slower … you get more of the same. Additionally you get the same stuff as other users that have a similar usage and behavior profile as you.” My slide puts it shorter: “google discover is a popularity contest”. A reader clicks and interacts, the story is shown to more readers like them, and it starts again.
  • It is spiky. “Sometimes one article which gets one day more traffic than the rest of the whole website for the rest of the week or even month.” In one newsroom’s 16-month data, the best single Discover article brought as many clicks as about two and a half weeks of the whole site’s Search traffic, and the best Discover day had about eight times the clicks of a normal day. “Discover is a power distribution.”
  • It is big. In six newsroom accounts I worked with from 2023 to 2026, Discover brought about 60 to 80 percent of all clicks from Google. Across three news sites over the year to October 2026 it was about 74 percent, while all AI sources together sent 0.02 to 0.04 percent of sessions, depending on the window. “The traffic you get from AI is minimal!”
  • It is a news channel. On one site with mostly evergreen content, Discover brought well under 1 percent of what web search brought. The feed wants what is new.

That is why I say: “For News Site SEO is not about Google Search, but Google Discover Optimization.”

The real challenge: headline, image and story aligned

In the feed, the reader gets one headline and one image. Nothing else. “Die User sehen die Headline ohne Context”: they see the headline without the context of your home page or section.

So three things have to work together, and in most newsrooms three different people own them:

  1. The headline must be clickable. “Headline (which is the H1, not the SEO title), Image and Timing. The strongest leverage you have … is the headline and the image. Because that is what shows up in the Google Discover feed.”
  2. The image must make people stop scrolling. It is the biggest part of the card.
  3. The story must deliver what both promise. If the targeting promises something and the page does not deliver, the reader bounces back and does not return soon. You lose the brand, not only the click.

Getting all three aligned, every day, under deadline, is the work. One good story is luck. A newsroom that does it every day has a process.

Discover optimisation is a process, not a checklist

Discover is a channel, and a channel can be run. It is a machine: the same steps, every day, measured and improved.

Best practices are where we start. The real work in every workshop happens with your editors: we work together on your tasks and your processes, because no two newsrooms work the same way. You do not leave with a stack of checklists. You leave with tasks your team can do and a process your team can run.

One-time tasks

What we decide together, sorted into critical, good idea and nice to have:

  • Technical: what your CMS and article template must ship (images, markup, feeds, the 3 SEO tests), and what your developers can really build.
  • Editorial: your headline rules, your image process, read-more links, who adds polls and videos.
  • Measurement: your Discover baseline, your “did work / did not work” list, the numbers each desk is measured on.

Every task gets an owner and a date.

The Discover cycle

This part never goes away. It runs every week, for as long as you publish.

  1. Topic setting: which topics travel in the feed, read from your own Discover data, not from gut feeling.
  2. Publishing aligned: H1, image and story checked together before the story goes out.
  3. Measuring: impressions, clicks and click rate per story, per desk and per market in Search Console.
  4. Feedback loop: what was done, what worked, what did not, back to the desks.
  5. Testing: one change, one desk, a start date and a stop rule.
  6. Repeating: what worked becomes the rule, what did not gets dropped, evergreen stories get updated.

Who owns what

A process without owners stops after two weeks. In the workshop we name them:

RoleOwns
Desk headsTopic setting, headline and image per story
Photo deskThe image process
SEO editorThe feedback loop: what worked, what did not, what we test next
DeveloperThe article template and the 3 SEO tests after every release
Data ownerThe Discover baseline and the reports per desk

The names differ in every newsroom. The roles stay.

My Discover best practices

Google’s documentation is the floor, not the playbook. These are the practices the tasks start from, editorial first, then technical.

PracticeWhat it means for your desk
They need to click youThe click trigger sits in the H1 and in the image
They need to click somethingA reason to stay after the first paragraph
Image languagePeople, action, own footage, colour
Large, linked images1200 px, three crops, linked to the large version
NewsArticle, feeds, sitemapThe markup and feeds the article template ships
The 3 SEO testsPageSpeed, JS turned off, Search Console live test
Evergreen, updatedChange everything but the URL
Each channel apartDiscover, Search and News are three different reports

They need to click you

“To do 1: They need to click you!” Click-through is the first signal.

  • Put the click trigger in the H1 and in the image, not in the third paragraph.
  • Write the H1 for a reader who knows nothing yet: who, what, where, plus the reason to click it in Discover.
  • Keep the click trigger, drop the false promise. Clickbait is what lost the most in my data, see below.

They need to click something

“To do 2: They need to click something!” The second signal happens on your page.

  • A read-more link after the first paragraph, at the latest after the second. On my slides it is “the most important editorial process” for more traffic. Its link text may tease, but it must stay on topic.
  • Polls, videos and galleries added by a second editor, not left to chance.
  • A block of links to other stories, so every reader has a next click. Staying on the page and on the site counts too.

Image language

“Discover is a visual medium.” Hard to quantify, so I give the desk an image process, then we test it:

  • A person beats an object.
  • Action beats stillness.
  • Your own footage beats the agency photo.
  • Colour beats brown and grey.
  • Authentic beats generic. Fewer stock photos of things.
  • Use what your social media team knows about images.
  • Resolution matters less than the motif, as long as the image meets the size rule below.

Large, linked images

My rule: large images “need to be at least 1200 px wide and enabled by the max-image-preview:large setting”. But they must be fast and SEO-optimised too:

  1. max-image-preview:large in the robots meta tag.
  2. One source image, cropped server-side to 1200 px in 16:9, 4:3 and 1:1, all three in the NewsArticle image array.
  3. An og:image at 1200 x 675 with width and height, showing the story, never the logo.
  4. The image linked to its large version, at least 1200 px. The click may be cancelled with JavaScript.
  5. Responsive srcset and sizes, compressed files, a file name and alt text that say what is in the picture, width and height on every image.
  6. The hero image is the only one not lazy-loaded, with fetchpriority="high".
<meta name="robots" content="index, follow, max-image-preview:large">
<a href="/images/mayor-opens-new-bridge-1200.jpg">
  <img src="/images/mayor-opens-new-bridge-640.jpg"
       srcset="/images/mayor-opens-new-bridge-640.jpg 640w,
               /images/mayor-opens-new-bridge-1200.jpg 1200w"
       sizes="(max-width: 640px) 100vw, 640px"
       width="640" height="360" fetchpriority="high"
       alt="The mayor cutting the ribbon on the new bridge">
</a>

NewsArticle, feeds, sitemap

  • NewsArticle markup is critical for Discover; breadcrumbs are good to have; everything else is nice to have. Headline, the three images, datePublished, dateModified, author and publisher with logo. Check it in the Rich Results Test.
  • Keep the publisher stable. One name, the same on every page.
  • An RSS feed per section, linked in the <head> of the start page and of every article. In one case below, the section feeds had disappeared, and restoring them was on the fix list.
  • A Google News sitemap, submitted in Search Console and in Google’s Publisher Center. Google does not require it for Discover. I do: Discover and Google News are closely linked, and without a news sitemap and managed visibility in Google News, Discover stays luck, not a channel.
  • AMP is optional. If you run it, it must validate and carry the same NewsArticle markup as the regular page.
{
  "@context": "https://schema.org",
  "@type": "NewsArticle",
  "headline": "The same words as the H1",
  "image": [
    "https://example.com/images/story-16x9-1200.jpg",
    "https://example.com/images/story-4x3-1200.jpg",
    "https://example.com/images/story-1x1-1200.jpg"
  ],
  "datePublished": "2026-11-05T08:00:00+01:00",
  "dateModified": "2026-11-05T11:30:00+01:00",
  "author": [{ "@type": "Person", "name": "Author Name", "url": "https://example.com/authors/author-name" }],
  "publisher": { "@type": "Organization", "name": "Your Publisher", "logo": { "@type": "ImageObject", "url": "https://example.com/logo.png" } }
}

The 3 SEO tests

Every article template passes these three, before and after every release:

  1. PageSpeed Insights: at least 80 (still orange), better 90 (green), with screenshots for mobile and desktop. If you cannot pass it, AMP is the fallback.
  2. The “JS turned off” test: above the fold and the main content are visible with JavaScript turned off. Interlinking works with JavaScript turned off: visible links must be links.
  3. The Search Console live test: Inspect URL, Test Live URL, View Tested Page. The screenshot must show the rendered page.

Evergreen, updated

  • Update evergreen articles at least once a season or once a year.
  • Change everything but the URL. Never change the URL after publishing, and never the targeting. Changing a slug after publication must be an explicit editorial decision, because it can cost that story its Search and Discover traffic.
  • Update the date and move the story back to the start page. The latest update date is the date you show readers and Google.

Each channel apart

Search Console reports Discover, Search and Google News separately. Read them separately.

  • Nothing to install, ad blockers and consent do not matter, but you get only clicks and impressions, no session data.
  • Google News results that appear inside Google Search are reported as Search, not as News.
  • Sort your own stories into “did work” and “did not work” in the Discover report. That list teaches the desk more than any slide.

What I found in Discover data, 2024 to 2026

The best practices come from cases like these, from my own audits, workshops and second opinions, most of them for newsrooms.

What happensWhat I measuredWhat I specify
Same URLs, opposite outcome by countryAbout 75 to 80 percent of Discover back in every market but one; under 2 percent thereSplit by market and URL pattern before blaming the site change
Impressions fall, the click rate staysImpressions down to under 2 percent, click rate flat at about 10 percentRead impressions, clicks and click rate apart
A windfall is not a baselineClicks quadrupled, then fell to about 5 percent of the old floorSet the target at the old floor, not the peak
Clickbait lost the mostAbout 30 percent of one loss came from sensational headlinesKeep the click trigger, drop the false promise
You cannot fix Google, but you can actWeekly clicks about halved for three weeks, then backWait for the end, take the new baseline, act on data
Images too small for the feedArticle images about 600 px wide; no large image linked on another siteThe image rules above
Impressions, not clicks, separate winnersTop and weak pages had almost the same click rateTopic first, then headline and image
Done means passed on the live pageA speed score that went down, not up, while every task was marked doneA pass rule per task, a re-test after every release

Same URLs, opposite outcome by country

After one site change, a publisher lost a large part of its traffic. The new system got the blame.

Old and new articles sat on different URL patterns. So I split the publisher’s own Search Console data by URL pattern and by market.

  • In every market except one, the new URLs got about 75 to 80 percent of the old Discover clicks back.
  • In one large market, they got under 2 percent.
  • “Same URLs, same code, opposite outcome by country.” A broken URL pattern would have hit every market.

It matched Google’s February 2026 Discover update, which favours “locally relevant content from websites based in their country” (source).

What I specify: a control market, a split by URL pattern, one KPI (new-URL Discover impressions per day in the lost market) and a stop rule: if it stays flat for three to four weeks, stop spending on that market and defend the home market.

Impressions fall, the click rate stays

In the same case, Discover impressions in that market fell to under 2 percent of their level within about six days. The click rate stayed at about 10 percent.

  • Readers who still saw the stories clicked them as often as before.
  • So the readers did not reject the content. The feed stopped showing it.
  • A fall over six days is not a penalty, which would be an instant zero.

What I specify: every Discover drop gets read in three lines (impressions, clicks, click rate), per market and per device, before anyone touches the CMS.

A windfall is not a baseline

Before the loss, monthly Discover clicks from that market had quadrupled within a few months. Afterwards they fell to about 5 percent of the old floor.

The loss everyone measured was against the peak. The honest target was the old floor. Spiky traffic needs a baseline that ignores the spikes.

What I specify: the full 16 months of Search Console history before any target, not the last 28 days.

Clickbait lost the most

In one Discover loss, I tagged URLs whose headlines carried sensational markers.

  • Their Discover clicks fell by almost 99 percent.
  • They made up about 30 percent of the whole loss.

Google’s February 2026 update lists “Reducing sensational content and clickbait in Discover” (source).

What I specify: a headline test on one desk for a few days, without clickbait, measured against the other desks.

You cannot fix Google, but you can act

In one case, a newsroom’s weekly Discover clicks about halved for three weeks, during and right after a Google core update. The week after, they were back at over 80 percent of the old level.

My slide for the team at the time: “It is very likely that the market (Google) changed. So there is no way to ‘fix’ this, as - if it was the market change - you did not break it. And you can not ‘fix’ Google. But there are as always optimization strategies in the area of development, content and marketing that you can strive for for better performance.”

Google is a marketplace. An update changes the market, not your site. So after every update:

  1. Wait until it has finished rolling out. A relaunch, a rebuild or a new headline policy waits too.
  2. Look at the data once it is over: impressions, clicks and click rate per desk and per market.
  3. Take it as the new baseline. Not the old peak, not the low point in the middle of the update.
  4. Act on it: back to the Discover cycle. What works now, what does not, which new approach the desks test next.
  5. Measure the impact of every action against the new baseline.

Action and reaction. That is the Discover game, again and again, based on data.

What I specify: Google’s update dates next to every Discover chart, a new baseline after every update, and the next round of tests set from it.

Images too small for the feed

  • In one audit the article image was about 600 px wide: half the size my image rule asks for.
  • In another audit, the srcset promised images up to about 3,800 px wide, but the files served were at most about 1,000 px. The og:image was a 1200 px copy blown up from that smaller file, max-image-preview:large was missing, and the image was not linked to a large version.
  • On another site, no page type linked an image to a large version. In 90 days that site got under 1,000 Discover impressions, against millions in web search. Ranking in Search does not put you in the feed.

What I specify: the image rules, built into the image pipeline, so no editor has to think about pixels.

Impressions, not clicks, separate winners

I compared the top and the weakest of the 1,000 strongest pages of several publishers over 16 months of Discover exports.

  • The click rate of strong and weak pages was almost the same.
  • The difference was impressions: how many feeds a story got into.
  • Only about a fifth to half of the top Discover URLs were also top Search URLs.

So a Discover story and a Search story are different jobs. Topic choice decides more than technical optimisation.

What I specify: topic first, then headline and image. The workshop sorts your own top and weak Discover stories side by side, so the team sees which topics travel.

Done means passed on the live page

One editorial team implemented almost every task from its workshop. I re-measured the core page types on three dates over three months.

  • The speed score never reached the goal. It went down, not up.
  • The main content loaded faster, but the page spent longer blocked by scripts, so the score fell anyway.
  • The team’s own report a year later: everything implemented, no improvement seen.

The tasks were done. Nobody owned the question whether the result arrived on the live page and stayed there. So in every workshop we set up that check together with the team:

  1. Every task gets a pass rule you can test on the live page. Not “optimise images”, but “the article image is linked to a 1200 px version, and max-image-preview:large is set”.
  2. Done means the test passes on the live page, not that the ticket is closed.
  3. The 3 SEO tests run after every release, owned by the developer, on one article, one section page and the start page.
  4. A score that drops gets traced to the release that caused it, then fixed or rolled back, then tested again.
  5. Three measurement dates go into the calendar before the workshop ends, for example two weeks, six weeks and three months later.
  6. Technical result and traffic result are read apart. First: did the fix arrive? Then: did Discover impressions move against the baseline?

What I specify: one scorecard per newsroom: task, pass rule, owner, date, result on the live page. It is the agenda of every follow-up call.

How it runs

  1. Before the day: I pull 16 months of Search Console data for Discover, Google News and Search, side by side. You get the baseline on the first slide.
  2. Half a day to two days, in-house or remote. Editors, desk heads, the photo desk, the SEO editor, developers, marketing and the person who owns the data in one room.
  3. A workshop, not a presentation. We work on your tasks together: your best and weakest Discover stories, their headlines, images and markup, and the 3 SEO tests run live on your article page.
  4. Your Discover process, written down: the cycle, one owner per step, one experiment per fix with a start date and a stop rule.
  5. Re-measured on fixed dates afterwards, with the same tests, so you see what worked.
  6. You leave with task lists and processes for every team, see below, plus the slides as PDF and my book.

What you leave with: tasks and processes for every team

This is the most important part of the workshop. You do not leave with my checklist. You leave with your own task lists and your own processes, one set for every team that touches Discover:

Your editorial team

Your task list:

  • Headline rules for the H1: who, what, where, plus the reason to click.
  • The image process for the desk and the photo desk.
  • Read-more links, polls and videos, and who adds them.
  • The evergreen update list.

Your process:

  • Topic setting from your own Discover data.
  • Headline and image checked together before a story goes out.
  • A weekly “did work / did not work” review, desk by desk.
  • Headline tests, one desk at a time.

Your development team

Your task list:

  • The image pipeline: 1200 px, three crops, linked large version, max-image-preview:large.
  • NewsArticle markup, section feeds and the Google News sitemap.
  • The article template passing the 3 SEO tests.

Your process:

  • The 3 SEO tests after every release.
  • A dropped score traced to its release, fixed or rolled back, tested again.
  • The scorecard kept up to date.

Your marketing team

Your task list:

  • Where Discover sits next to social, newsletter and your other channels.
  • The image and headline language shared with the social team.
  • The numbers the channel is reported on.

Your process:

  • A regular channel review: what Discover brings, what the other channels bring.
  • Where the channels help each other and where they collide, and who decides.

Every task has an owner, a date and a pass rule. Every process has an owner and a fixed rhythm.

Specified together, not top down

None of this is a specification handed down from management or from me. We write it together with the people who will do it, because they are the ones who have to do it.

  • Tasks need to be done, so the people doing them must be invested in them.
  • Processes never go away. They only survive if the team owns them.
  • Every newsroom works differently. The same rule needs a different task in your CMS, your desk routine and your release cycle.

It is change management

If Discover is a market you want to go for, your processes have to change to fit this channel, and your platform has to change too: the article template, the image pipeline, the release checks.

And Discover is not your only channel. Search, Google News, social and newsletter pull on the same stories, the same editors and the same developers. So the workshop aligns them: what each channel needs, who decides when they collide, and how you measure all of them side by side.

For Search and Google News in depth, there is the Newsroom SEO Workshop.

Questions newsrooms ask me about Discover

Do we need special markup to get into Discover?

Google says no special tags are required (source). Eligible is not the same as visible, though. I still treat NewsArticle markup as critical for Discover: it carries the headline, the dates, the author and the three image crops.

Do we need a Google News sitemap for Discover?

From Google’s side, no. Content is eligible if it is indexed and meets Discover’s content policies (source).

From my side, yes, if you want Discover as a channel. Discover and Google News are closely linked. Only with a Google News sitemap and visibility in Google News that you actively manage can you treat Discover as a channel you work on every day. Without it, it is “it worked here, it worked there”, and nothing you can repeat.

Sites outside Google News can still appear in Discover. For them it stays an occasional hit, not a channel you can plan.

Our Discover traffic dropped. What first?

Date the drop and check Google’s update dates. If an update is rolling out, wait until it is over, then read impressions, clicks and click rate per market and per URL pattern, take that as the new baseline and act on it. The big drops I analysed in 2024 and 2026 were on Google’s side, not in the code. You cannot fix Google, but you can act on the data.

Can a site win Discover on topics outside its core?

Less and less. Since February 2026, Google says it identifies expertise “on a topic-by-topic basis” (source). Build the sections you want to win in Discover as real sections, with their own feed and enough stories.

Should we build the newsroom around Discover?

Around Discover and Search, never Discover alone. Spikes come and go, and a windfall can vanish in a week. In the workshop we look at what your Search and Google News traffic would carry if Discover halved.

Do sensational headlines still work?

They lost the most in my data. Keep the click trigger, drop the false promise.

Is this about AI search too?

Only where the data says so. In my 2026 news-site data, AI sources sent 0.02 to 0.04 percent of sessions. Discover brought about three quarters of all Google clicks.

Sources: my own audits, workshops, slides and second opinions from 2023 to 2026; the client numbers are rounded and anonymised. Google pages read on 2026-11-05: Discover and your website (last updated 2026-03-09) and Google’s February 2026 Discover Core Update (published 2026-02-05).

Some of our customers

Examples, not the full list

  • Kurier
  • oe24
  • Heute
  • Wiener Zeitung
  • Speedweek

From real workshops

  • A Discover loss after a site change, split by market. The cause was on Google's side, not in the code.
  • 2026: a newsroom content workshop with before/after rewrites of the team's own articles.
  • 2024: a one-day Discover checkup with editors, developers and data in one room, plus a benchmark against other publishers.
  • 2024: Discover and News training for an editorial team.
  • News agencies, 2018 to 2026: content SEO workshops and remote trainings.