AI inbox summaries now decide which PR pitches journalists actually read
New data from 628 pitch tests and 39 million subject lines shows Gmail, Apple and Outlook AI summaries are filtering press pitches before journalists see them.
AI tools built into Gmail, Apple Mail and Outlook are now summarizing press pitches before most journalists ever open them, according to new research the PR software company BuzzStream published this month, aggregated this week in a guest analysis on Search Engine Journal. The findings matter for any software company that still measures PR success by how many reporters it emails rather than how many actually read what it sent.
What changed
BuzzStream, working with digital PR agency Aira, sent 628 pitch variants across 13 campaigns to Gmail, Apple Mail and Outlook addresses and logged what each inbox's built-in AI summary pulled out. Gemini's summaries ran short, averaging 28.8 words, and pulled 87.1% of their keyword matches out of each email's opening half. Apple Intelligence did much the same, at 82.5%, and Microsoft's Copilot, despite writing longer summaries that averaged 156.5 words, still skewed toward the opening half 65.4% of the time. Bulleted text shaped at least 64% of summaries across all three platforms, rising to 92% for Copilot. The personalized opening line that digital PR has leaned on for a decade barely registered: it showed up in just 11% of Google's summaries and none of Apple's. Apple and Microsoft summaries also misrepresented the original email, mostly data points and quotes, in roughly one of every three cases.
A separate BuzzStream analysis of 14.3 million emails tied domain age directly to whether a pitch gets opened at all: sending domains five to six years old averaged a 34.9% open rate, against 21.6% for domains one to two years old. Across 1,588 sending domains, full SPF, DKIM and DMARC authentication nearly quadrupled opens, 30.1% versus 7.8% for unauthenticated senders. A third study, covering 39 million subject lines, found that stating a concrete finding beat teasing one (43.1% opens versus 39.8%), and that putting the key number in a subject line's first three words beat burying it in words four through six, 46% versus 38.9%.
Behind the numbers is a crowded inbox: BuzzStream estimates two to three PR professionals for every working journalist in the US, and in Muck Rack's State of Journalism 2026 survey of nearly 1,100 journalists, 47% put themselves in the group that rarely or never gets a pitch lined up with their actual beat.
"A good story isn't enough anymore." — Vince Nero, BuzzStream, in Search Engine Journal
The squeeze runs in both directions. The same AI tools that now summarize inbound pitches also make it cheaper to send more of them: link builders and growth marketers have been migrating into PR as a way to get cited in AI answers, and generic AI-drafted pitches, including some built around fabricated expert sources, have added to what lands in a reporter's inbox. So the inbox gets fuller on the sending side at the same time it gets more automated on the receiving side, which is why a pitch now has to clear two filters instead of one: whatever the email client's AI decides is worth surfacing, and then the journalist.
Who this affects
Product launches, funding news and original research reports are a standard SaaS PR play, and most of that outreach still runs through a journalist's email client rather than a chatbot. If the first reader of a pitch is now an AI summarizer rather than the reporter, the elements built to charm a human — the flattering line about their last article, the softened lede — mostly never reach anyone. This article is itself an example of the pattern it describes: a software vendor publishing a data study to earn exactly this kind of trade-press pickup. BuzzStream sells outreach software and discloses that in its own byline on Search Engine Journal. Muck Rack's survey, run by a separate company with its own commercial interest in PR software, points at the same oversaturated-inbox problem independently, which is the main reason the two studies are worth reading together rather than on their own.
What a software marketer should do differently
- Confirm SPF, DKIM and DMARC are fully configured on the domain doing the pitching
- Vet journalists against their last 10-20 articles, not a beat label or job title
- Lead subject lines with the concrete finding, and put the key number in the first three words
- Put the single most important fact in the first two sentences of the email body
- Use bullets for supporting detail, since AI summarizers weight bulleted text heavily
- Spot-check coverage once it runs; summaries misstate facts in roughly one of three cases
Before your next outreach batch, run the sending domain through an SPF/DKIM/DMARC checker, and rewrite the subject line so the specific finding sits in the first three words rather than the last.
AIResearchers planted a fake concept online, and 8 of 10 AI search tools cited it within a week
Sienna Mcpherson · 5 min read
AIGoogle and OpenAI's API only match a quarter of what ChatGPT actually cites, a new study finds
Sienna Mcpherson · 4 min read
AI