Media Briefing: The 2024 media glossary, pt. 2

Here’s the latest guide to what people are actually saying when they talk about the ever-changing media industry.  

Authenticated audiences

What it is: Online users who have self-identified themselves when on a website or platform by making or logging into an account, sharing their email address or using another one-click sign-on service like Google’s One Tap. 

Why it’s important: Some publishers reported that they’re beginning to see revenue lift on non-third-party cookie supported ad inventory thanks to deterministic IDs, like RampID, Unified ID 2.0, ConnectedID and ID5, which are based on authenticated audiences. With third-party cookies being phased out – if no longer entirely vanishing – revenue benefits from cookieless alternatives gives publishers better directional guidance for which solutions to be testing and implementing. 

Further reading: 

• How publishers are trying to get people to log in to their sites
• Publishers’ audience authentication strategies are put to the test
• Publishers search for new ways to grow (and authenticate) audiences, overheard at the Digiday Publishing Summit

LLM

What it is: Stands for “large language model,” which is the “brain” for generative artificial intelligence technology

Why it’s important: LLMs are how AI tech companies improve the accuracy and effectiveness of their chatbots and other AI technologies. When publishers enter into a content licensing deal with AI companies, their content is being used to make the LLM “smarter” or more up-to-date. Publishers’ content is also used to inform LLMs without their permission via web crawlers (see Robots.txt). Publishers developing their own generative AI tech can also license LLMs from companies like Google or OpenAI, or use open-source models that are free, like Meta’s Llama 2. 

Further reading: 

• The case for and against open-source large language models for use in newsrooms

Prompt engineering

What it is: An advanced method to communicate with LLMs and generative AI products like chatbots using written language to yield specific responses from the AI. Rather than saying ‘write me a cover letter,’ the prompt to the AI chatbot would include specific inputs like resume information and the job description; Also referred to as prompt design or prompt construction

Why it’s important: For publishers experimenting with AI technology, such as creating a custom chatbot, prompt engineering is an important skill set that’s needed to get the AI to operate the way they want. For example, if the publisher wants their audience-facing chatbot to respond to users in the same voice or style as their editorial content, they would need to train the chatbot to learn that through prompt engineering. 

Further reading: 

• WTF is prompt engineering

RAG

What it is: Short for “retrieval augmented generation,” it’s a method for a company, like a publisher, to make its content and proprietary data available to an LLM

Why it’s important: As publishers enter into content licensing agreements with artificial intelligence companies, their product and technology teams will have to develop RAG processes in order to fulfill their end of the bargain (sharing their content with the AI company). While it’s meant to streamline the content sharing process, RAG is also a technique to improve the accuracy and relevancy of publishers’ AI chatbots.

Further reading: 

• Amazon adds tools for scaling generative AI applications — and improving accuracy issues
• With a new ‘answer engine,’ Brave browser adds another generative AI tool for search

Robots.txt

What it is: A protocol publishers can deploy on their websites to attempt to block specific crawlers, like AI bots, from scraping and ultimately using content on the site without permission by the site owner. Robots.txt signals to the crawler not to scrape the site, but the crawler can ignore it. 

Why it’s important: Publishers and generative AI tech companies have been locked in a back-and-forth about AI’s right to use publishers’ content to train their LLMs under the fair use clause of copyright law. While it’s resulted in many lawsuits and cease and desist letters issued by publishers to the tech companies, other pubs that can’t afford legal pursuit – or just want to take more swift action – have relied on their robots.txt files to block crawlers, such as OpenAI’s GPTBot. 

Further reading: 

• How 3 publishers are making their sites more/less habitable to AI crawlers 
• Why protecting paywalled content from AI bots is difficult business
• WTF is OpenAI’s GPTBot?

SERP ranking

What it is: SERP stands for “search engine results page,” and high SERP rankings on search platforms, like Google, is what commerce publishers relied on for getting traffic to their websites

Why it’s important: Google released a Core Update to its algorithm in March, which several commerce publishers said decreased their SERP rankings, leading to a significant decrease in traffic to their websites. As a result, they’ve started relying on newsletters, social media platforms and on-site promotion to try and increase the viewability of commerce content. By and large, fewer and fewer publishers say they’re able to comfortably rely on search traffic as a primary way to connect with audiences. 

Further reading: 

• Publishers say Amazon Prime Day sales were up but commerce revenue remains ‘hard to predict’
• Three strategies publishers are adopting to drive affiliate commerce revenue for Amazon Prime Day 2024

What we’ve heard

“Our ad revenue has grown … We’re up 20% and we’re looking at a very strong second half.”

– Jessica Sibley, CEO of Time, on the latest episode of the Digiday Podcast, discussing the results of the company’s shift to a B2B revenue model.

3 Qs with Apartment Therapy Media’s president Riva Syrop 

Apartment Therapy is the latest publisher to roll out an AI-powered chatbot, using the shiny new toy as an opportunity to attract advertisers for custom, high-touch ad campaigns. 

Called the 5-Minute Makeover, the chatbot creates an interior design mood board for users based on their answers to questions about their preferred design style. With The Home Depot as the chatbot sponsor, users are presented with furniture and decor items from the retailer’s catalog of 400,000 products that the chatbot thinks they would most like based on their responses. 

Even though the products used to create the mood boards will be shoppable, Apartment Therapy Media’s president Riva Syrop said the deal with Home Depot, which runs through early 2025, does not include an affiliate commerce component. She declined to share how much money the deal is worth, but added a “significant” brand deal is needed to afford a lot of people using the chatbot.

Was this chatbot built for the Home Depot deal? Or was this a product ATM was planning to create regardless of a brand deal? 

It’s been our objective anyway, so I think we would have done it [without an advertiser] … Developing the tool doesn’t have a lot of cost associated with it … like tons and tons of internal resources across our product team, our content team, our design team [went into building it] so there’s a huge resource lift there. But [the cost comes in when] returning the results to people as they [use the chatbot]. 

What are the metrics that Home Depot is using to assess the chatbot campaign performance? 

How many people are gravitating towards it, how many people are engaging with it, the nature of their inquiries. But what we’re really measuring is how many people are creating mood boards from it … It’s going to be hard for us to sort of gauge transactions, because I think a lot of the purchasing is going to happen off the [chatbot], but certainly clicking over to the Home Depot will be one of the measurements that we’re looking at.

Which AI technology did you use to build this chatbot? 

Google … [but] we used a couple different large language models because some of them are really geared towards different operations. So operations that require really robust thinking and high accuracy, [those LLMs] tend to be a little bit slower because it’s doing so much work. And some of the more nimble large language models [are less accurate but] can return results really, really quickly. 

Because [our chatbot has the] five-minute makeover promise, we had to try and make sure that everything was coming back in an average of 10 seconds. So we actually ended up merging a few different large language models to make sure the different things were responding to different types of queries.

Numbers to know

~100: The number of employees that Apple eliminated, impacting staffers on the Apple News, Apple Books and Apple Bookstore teams.

$50 million: The amount of revenue Gallery Media Group expects to earn in 2024, marking its 13th straight year of profitability. 

$100 million: The amount that Amazon’s podcast network Wondery paid for the rights to distribute and sell ads for sibling NFL stars Jason and Travis Kelce’s podcast “New Heights.” 

$250 million: The amount of money that Google and California legislators agreed to pay out to news organizations over the next five years so they can test artificial intelligence technology. This essentially delays California’s proposed bill that would force big tech companies, like Google, to share revenue with news publishers in the state for profiting off of their content.

What we’ve covered

AI Briefing: The AI search race heats up with more audit tools, ads and pending antitrust outcomes:

• Large language models powering AI search platforms like ChatGPT and Perplexity are still largely a black box, 
• Some CRM and SEO providers, however, are looking to give marketers more info about how brands appear on the inside — even from the outside.

Read more about the latest AI developments here.

Perplexity’s pitch deck offers advertisers a new vision for AI search:

• Perplexity’s plans for competing in the ad space are becoming a bit less puzzling. 
• Although the pitch deck didn’t talk about pricing, a source within Perplexity confirmed the goal is to target CPMs “north of $50.”

Learn more about Perplexity’s go-to-market ad sales strategy here.

Five key signs that Reddit is getting ready to launch its own search ads business: 

• When it comes to building out a search ads business, Reddit is staying tight-lipped. 
• But all the recent moves the platform has made suggest that it’s a matter of when, not if, it’ll happen.

See why industry execs believe Reddit is planning its launch here.

What we’re reading

More ads are showing up in podcasts, worrying hosts about audience’s tolerance for commercials:

The Wall Street Journal reported that the number of ads running in podcast episodes are increasing compared to previous years, representing on average almost 11% of the run time of episodes published in the second quarter. While more ad spots should mean more revenue, podcast creators are increasingly concerned that too many ads will ruin the listening experience.

Gannett is shutting down its product review site, Reviewed, following AI controversy:

Last October, Reviewed was suspected of publishing consumer product reviews written by generative AI, causing its staffers to publicly criticize their employer. The Verge reported that nearly a year later, the site will shut down, effective Nov. 1. 

Hearst taps new editor-in-chief for Cosmopolitan and Seventeen:

Hearst Magazines is injecting its legacy women’s and teen magazine brands, Cosmopolitan and Seventeen, with new blood, Business of Fashion reported. Willa Bennett is taking over the top editor role of both brands, succeeding Jessica Giles, who served as editor-in-chief of Cosmo for the past six years. Bennett’s background at streetwear and youth culture brand Highsnobiety suggests Hearst’s ambitions to expand its cultural scope of coverage and prospective audiences. 

Sarah Palin’s libel trial against The New York Times is revived: 

After the former governor of Alaska and Republican vice-presidential nominee lost her original libel trial against the Times in 2022, a federal appeals court ordered a new trial for Palin’s lawsuit, the Times reported. The federal appeals court cited that the previous judge wrongly excluded evidence and possibly swayed jurors during deliberation.