Hi SODP community!

Let's recap on what's been happening, the new content, industry updates, tips, and more.

.SODP POSTS.

.TIP OF THE WEEK.

The AI Citation Controls Publishers Have and Aren't Using

One number frames the year: People Inc. lost 800 million Google sessions between Q1 2024 and Q1 2026. That is the environment in which publishers finally won the controls they had been demanding for three years, and in which almost none of them are being used.

What Landed in Twelve Months

• December: RSL 1.0 shipped with 1,500-plus organisations behind it, putting machine-readable licensing terms on top of robots.txt.

• February: Bing launched first-party AI citation reporting, then rewrote its Webmaster Guidelines with directive-by-directive AI guidance and named Generative Engine Optimisation in formal policy.

• June: The UK's CMA ordered Google to give publishers an opt-out from AI Overviews, AI Mode, and AI summaries in Discover, plus an opt-out from model training. Google is rolling that toggle out globally.

That is more granular control over AI use of publisher archives than has existed at any point since ChatGPT launched.

Almost nobody is going to flip any of it.

Why Publishers Aren't Acting

• Paul Bannister (Raptive) called Google's version "a light switch while keeping the power plant running."

• Stuart Forrest (formerly of Bauer) points to the core reason: Google has broken out AI impressions but not AI clicks, so publishers cannot model what leaving costs them.

• Phil Andraos (Reuters) put it more simply: "There's not enough data."

• Thomas Lue Lytzen (Ekstra Bladet) expects most publishers to stay in because they fear the unknown.

So the industry now holds a set of levers it cannot price. That is the real story of 2026.

The Part That Should Bother You More

One of these controls is nearly three years old, fully documented, free, and reversible — and most publishing teams still cannot tell you their setting.

Bing gave publishers nocache and noarchive on 22 September 2023.

1. nocache — Your page stays referenceable in Copilot answers, but Bing displays only the URL, title, and snippet. Microsoft also states that for nocache content, only URLs, titles, and snippets may be used in training its generative AI foundation models.

2. noarchive — Your content is excluded from Copilot answers, is not linked to in them, and is not used for foundation model training.

The catch: Publish both on the same page and Bing ignores the stricter one. Microsoft's documentation is unambiguous: "If content has both NOCACHE and NOARCHIVE tags, we will treat it as NOCACHE."

The permissive rule wins. Any team that layered noarchive over nocache believing it had locked its archive out of Microsoft's training corpus has not.

Two Things Follow

1. Paywalled sites: Microsoft actively recommends adding nocache alongside noarchive so paywalled articles stay discoverable at URL, title, and snippet level. Its own note is that most paywall sites use noarchive on its own and lose that discoverability by accident.

2. Rich citation: As of February, Bing recommends against nocache on anything you want richly cited in Copilot, flags that nosnippet and data-nosnippet can degrade citation quality, and has added a data-snippet attribute so publishers can specify exactly what it may cite.

None of this surfaces in a rankings report. Content carrying either tag still appears in standard Bing search results — Microsoft is explicit on that — so the only place the loss registers is in AI citation data.

Why This Is a Leverage Question, Not a Technical One

Your directive stack is an IP position. When you sit down opposite a platform to negotiate a licensing deal:

• "We have been excluded from your training corpus since 2023" — is a materially different opening hand than —

• "We have been feeding you for three years and only just noticed."

Most publishers cannot currently tell you which of those two sentences is true about their own site.

Nobody can force AI clickthrough data out of Google this quarter. But you can know your own permission map, and almost nobody does. That is the cheapest leverage available right now — and it is sitting in your page templates.

What you should Do This Week

1. Audit your templates. Pull your indexed URLs and check the robots and bingbot meta on every template — article, paywall, evergreen, syndicated. Record who set each one and when. If nobody can answer, that is the finding.

2. Check your data. Open the Bing AI Performance report. It is the only first-party AI citation data any search platform currently publishes, and it will tell you what your current settings are actually costing or earning.

3. Decide deliberately. Go template by template, rather than inheriting a decision someone made in 2023 for a reason nobody wrote down.

One Scope Note Before You Act

This is Bing and Copilot. Google's controls are a separate system with separate vocabulary in a separate console. ChatGPT is no longer a straight pass-through of Bing's index either — OpenAI crawls with OAI-SearchBot and has been building its own.

One decision does not carry across all three. That is exactly the problem.

Happy to walk you through any of it.

.NEWS OF THE WEEK.

➡️ The supply-demand curve is repricing publisher inventory. A few weeks ago, a Digiday reporter working on a story about new Ozone benchmarking data reached out to me. The numbers she shared were quite interesting. In Q2 2026, publisher ad request volumes had fallen 32% to 37% year over year in the U.S. and 39% to 41% in the U.K. But most interestingly, at least in the U.K., rising eCPMs made up most of the difference. In June, average eCPMs were up about 30% year over year in the U.K. and 7% in the U.S.

READ MORE

➡️ The Keyword Universe Was Always Smaller Than We Thought. Citation tools are fundamentally different from rank trackers, and that difference is almost always seen or stated as a limitation. One respondent to my recent survey of digital marketing practitioners put the case plainly. You cannot reverse-engineer what is working when the answer changes every time you ask, so what you are left with is closer to a brand awareness signal than a diagnostic. I have heard some version of that from enough people now that it functions as the default reading in this space, and it is a fair one.

READ MORE

➡️ Can public media build its own social network? Chicago Public Media teams up with New_ Public to try. If a news outlet announces a website launch, you might think it’s safe to assume it’s a news website that’s launching. But that’s not what Chicago Public Media has in mind for chicago.com. Chicago Public Media is already the parent nonprofit of two existing news sites, WBEZ and, since 2022, the Chicago Sun-Times. For chicago.com, though, the news org is teaming up with New_ Public, a nonprofit working to reimagine online public spaces, to launch a network of online communities for Chicago neighborhoods.

READ MORE

➡️ YouTube doubles monetization thresholds and adds a rolling Shorts test. Platforms rarely change the price of anything. They change the definition of what qualifies, which produces the same result while remaining technically a clarification. Over seven days beginning August 10, YouTube published a sequence of adjustments that, taken individually, read as housekeeping: an eligibility update, a subscription tier expanding into new markets, a layout tweak, a parental control explainer. Taken together they describe something considerably larger.

READ MORE

➡️ Google: Subject/Object Entity Order Affects AI Answers. Google published a new research paper that found that frontier LLMs encode 95–98% of the tested facts but are unable to directly recall 26–34% in answers to queries. Part of the problem is that recall becomes more difficult when questions reverse the subject/object entity order in which a fact was encountered in training. Parametric information is, essentially, the information that LLMs have encoded during training. That information comes from the web pages, song lyrics, books, instructions, code, and everything else that the LLM was trained on.

READ MORE 

➡️ Media plans are becoming one part of a bigger growth system. The media plan is still important, but clients are increasingly asking agencies to solve a much bigger problem: how to connect audiences, data, technology, creativity, commerce and measurement to drive business growth. Publicis Groupe ANZ chief media officer Imogen Hewitt has seen the change build over the past three to four years, with the shift accelerating over the last 18 months as data, technology and AI have moved closer to the centre of client growth conversations.

READ MORE 

.UPCOMING EVENT.

.JOB BOARD.

➡️ Newsweek (USA) is seeking an Associate Editor to join their US digital newsroom, focused on AI assisted news production. This role is ideal for a journalist who is passionate about the intersection of emergent technology and editorial excellence.

SEE MORE

➡️ News Corp Australia is looking for an Social and Influencer Manager to join Medium Rare's Social Centre of Excellence and help shape the future of creator and social marketing across some of Australia's most recognised brands.

SEE MORE

➡️ Dmg Media (London) is seeking a YouTube Growth Lead with a passion for building new brands and franchises on the platform, resposible for developing and executing YouTube strategy for their new channel launches, as well as supporting the growth and optimisation of their existing YouTube channels across their media portfolio.

SEE MORE

.SOCIAL MEDIA.

➡️ Marco Giordano on LinkedIn:

Small websites don't need much effort on data architecture.

After 8 years of work with all types of websites, I can say this works for micro to medium websites.

With a small investment, you can use the data you need.

🤔 What's a small website?

The definition changes depending on your niche/country but in this case...

traffic/volume is a good indicator.

❗ It’s not always related to how many pages you have.

E.g. a website with just 1K pages can have massive traffic and have a lot of data.

Your personal website is often a good example.

👉 I removed orchestration tools or even data governance.

If you are just starting, you don’t need that level of complexity.

👉 Storage is still the #1 priority for web data.

Google forces you to store your data in BigQuery if you want to do it properly.

Despite stereotypes, storing your data gives even more benefits to small players.

Due to anonymization and thresholding, most data will NOT be available in the UIs.

👉 Even so, you can even survive with standard GA4.

I still recommend you store your data for future use because it’s super cheap…

but you can rely on GA4 for most things.

BigQuery looks hard for non-technical people, but with Google documentation and LLMs... it's fine.

👉 GSC and BWT aren't just for SEO.

It's also audience data, what people look for, use it to your advantage.

GSC UI is heavily affected by several limits, so once again, use the native connector to store data to BigQuery!

👉 GTM is just good.

Create your business-relevant custom events and you are done.

Many overcomplicate it...

in practice you need those events/variables that are strictly related to your business.

Server-side implementations could be beneficial for some.

👉 DBT/Dataform are optional and for "bigger" small players.

Chances are you have a super low level of complexity in terms of needs.

Once you get bigger, you may consider moving toward processing your data and creating pipelines.

CONTINUE READING

These are the highlights for the last week.

Until next week!

Vahe Arabian and the editorial team at SODP