What You Should Know
What’s Changed in AI Search
As the weeks wither away in 2025, we’re reflecting on one of the year’s biggest stories in communications: the shift toward AI search. A survey by Modus Digital and Semrush showed 85% of CMOs say AI search has become a top priority. While communicators and PR teams are still adjusting to this new direction, AI models are also tweaking what they read and cite.
Yesterday, Muck Rack published an update to its Generative Pulse report that was first released in July. The new data shows AI still leans heavily on earned media to generate its answers — nearly 25% of all citations come from journalism. While that’s a 2% dip since July, strong media coverage still matters more than ever, both for human readers and AI models.
Earning the most valuable media coverage for AI search, however, requires more precision. Muck Rack compared the most pitched journalists on its platform to the journalists most cited by AI engines and found that on average there is only a 2% overlap. Which reporters and outlets matter most for you depends on what you want to be found for. You’ll need to track what makes sense for your brand (sign up for your free AI search audit) as you refine your AI search strategy.
Freshness also matters. Half of all citations come from reporting published within the last 11 months, and the highest citation rate happens in the first week after a story runs. That pattern rewards recent coverage and shows the value of consistent media coverage in influential outlets for the topics that you want to be known for. It also reveals opportunities for fresh coverage when older articles are often cited in AI answers. Pursuing new coverage on that topic could boost your brand’s AI visibility.
One of the biggest changes is a 5x increase in the number of press release citations — but it’s still miniscule, growing to 1% of citations from only 0.2% in July. The report says cited press releases have twice as many statistics, 30% more action verbs, and 2.5 times as many bullet points compared to non-cited releases. But you can cram a press release full of action verbs and bullet points and it can still remain invisible. A relevant topic and strong storytelling and structure remain the biggest elements of a winning formula.
The throughline in all of this is that AI search is rewarding teams that can pivot their media relations efforts at the speed of the news cycle. The brands that show up most in AI answers will track which outlets actually move the needle for them, identify opportunities for new or fresh coverage, and produce content that gives reporters something solid to work with. None of that requires a reinvention of PR — it just requires doing the job with more precision than muscle memory.
Elsewhere …
- LISTEN: Why Legal AI Fails Without Attorneys Writing the Content
- OpenAI’s Altman Declares ‘Code Red’ to Improve ChatGPT as Google Threatens AI Lead
- Anthropic Reportedly Preparing for One of the Largest IPOs Ever in Race with OpenAI
- AI-Assisted Shopping Is the Talk of the Holiday Shopping Season
Tips and Tricks
Manage token usage and context windows
What’s happening: AI seems to keep growing and growing, but the space within an AI conversation is finite. The way these tools work is that they review the entirety of the thread before responding. Essentially, they have to re-learn the context every time they’re prompted.
So what: The issue is particularly acute with Claude, which will cut off a conversation if the context window is full or you’ve maxed out your model usage. You can see your usage in “Settings,” but it’s something you have to keep an eye on manually because Claude tends not to give much warning before it cuts off a thread. Claude recently implemented “automated context management,” which summarizes earlier messages to save space. If you’re really desperate to finish the string you’ve started, Claude now has a feature where you can use a wallet to buy extra capacity on paid accounts.
Also: Another frustration, and this goes for any AI tool, is that even if the conversation continues, too much context can confuse the AI and lead to bad or nonsensical outputs.
What to do about it: Be cognizant of what and how much you’re feeding the tool. A file upload for instructions? Probably a good idea. One example of a piece of content you’re trying to build? That’s probably not enough or else the AI tool will just copy the actual content when you just want it to loosely follow a structure or tone. Twelve examples? You won’t have much room left in your context window.
If you’re using AI to draft content, try to use as few file uploads as possible and once you have a draft, consider starting a new thread to iterate through because too many versions of the same document will also confuse AI.
Quote of the Week
“Good writing is complex. A tapestry is also complex, so A.I. tends to describe everything as a kind of highly elaborate textile. Everything that isn’t a ghost is usually woven. Good writing takes you on a journey, which is perhaps why I’ve found myself in coffee shops that appear to have replaced their menus with a travel brochure. ‘Step into the birthplace of coffee as we journey to the majestic highlands of Ethiopia.’ This might also explain why A.I. doesn’t just present you with a spreadsheet full of data but keeps inviting you, like an explorer standing on the threshold of some half-buried temple, to delve in.”
— Writer Sam Kriss, in a New York Times Magazine essay on AI writing
How Was This Newsletter?
