What's working on LinkedIn right now (and what isn't)

Welcome to Resonance’s blog series: The Bi-Weekly Frequency.

Through here, you’ll be able to access a weekly roundup on all things B2B tech, algorithmic changes on the most popular communication channels, and top news of the week - all analysed by members of our team.

 

If you write for B2B brands on LinkedIn, the last few weeks have made one thing very clear: conviction beats broadcast. Announcements don’t pull engagement, but a person naming a real anxiety, backing it with one concrete detail and closing with something useful, does.

You can see the formula in the numbers. The founders dominating the engagement counts run the same four beats every time: a blunt, slightly contrarian opening line, a short build, a personal proof point ("I built a $15M business"), then a low-friction question. It's a repeatable template of hook, belief, receipt, offer, and it pulls a hundreds-plus comments a post because the opening line is built to make people nod or argue. Executives win the same way when they do it right: a specific internal mechanism or number (an eight-week build cycle, a valuation figure) tied to a bigger claim about the industry, not "excited to announce.”

Another high performing type of post is stories about a burned-out employee at a big organisation, because it mirrors something people actually feel. That's the real lesson hiding in the data: lived experiences travel. Users resonate, interact and share their own experiences, thus boosting engagement.

Which tells you exactly what isn't working: leadership platitudes, launch posts with no stakes, claims that are "safe" precisely because it commits to nothing. On LinkedIn right now, that content is simply invisible.

So the takeaway for anyone running a B2B brand or exec profile: find the balance between noncommittal corporate content and bold assertions. Either-or can be damaging to the client’s image.

A helpful framing is to start asserting. Pick a belief your buyer already agrees with and that isn’t damaging, then name the fear underneath it. Give a proof point and something for the reader to take away. And if possible, say it as a person, not a page. The reward is going to those who can portray the sense that a real human with real skin in the game is telling you something true that you need to know.

News of the week: AI and geopolitics cluster

OpenAI admitted its own models powered the autonomous agents that compromised Hugging Face's infrastructure, and didn't catch it for several days. When Hugging Face's security team tried to investigate, the hosted frontier models they first reached for refused to help, their safety guardrails unable to tell defender from attacker. The team then switched to an open-weight Chinese model and contained the incident on their own infrastructure.

For B2B tech, the takeaway is that your guardrails can block your own incident response. Security vendors should expect buyers to start asking what is the alternative solution if their models refuse to run, and to treat fallback models as solid safeguards instead of simply exceptions.

The same week, Meta, Microsoft, OpenAI and others signed a letter defending open weights against "premature restrictions," timed to Jensen Huang's first-ever X post making the "more eyes catch more flaws" case. Underneath all of that is the larger fight: AI-industry super PACs pouring record sums into the 2026 midterms, chasing a single national framework that preempts state-by-state rules. If successful, the lobbying efforts would further solidify the US’ continuity as the Western tech innovation hub, and create an even starker contrast with the EU.

However, the awkwardness of a pro-openness letter posed days after a closed model breached the open ecosystem's flagship repository didn't go unnoticed, and it's a reminder that positioning on AI safety now gets stress-tested against real events, fast. Business buyers hate this kind of uncertainty most of all. Buying and getting legal sign-off on new software already takes them months, and it's hard to commit to anything when the rules that govern it might still change.

On the other side of the globe, two Chinese models arrived in roughly a week: Moonshot's Kimi K3 and Alibaba's Qwen 3.8, both scoring close to the top US systems. Demand for K3 was strong enough that Moonshot briefly suspended new subscriptions. One marketplace reported Chinese models already accounting for roughly 60% of token usage among US companies, though that share is led by cheaper, older models like DeepSeek, not these two launches.

For buyers, two different things are happening at once. At the cheap end, models like DeepSeek cost 60–90% less than American ones, and plenty of US companies have already switched to save money. At the top end, these new models show China closing the quality gap, but they aren't the bargain the headline makes them sound. K3 costs about as much to use as a mid-tier Claude model, and running it yourself takes a lot of expensive hardware.

The bottom line of all of this for tech PR is that a client's claims now get tested against reality almost immediately, and usually in public. In all these cases, the gap between what was said and what actually happened became the story. Claims that are specific and a little hedged (e.g. this model is close to the best at these particular tasks, this one is genuinely cheaper, this one you can run yourself if you've got the hardware) survive a journalist or a buyer poking at them. And because security, openness, and China are all touchy subjects at the moment, the useful thing you can do is get clients to decide what they actually want to say on each one now, calmly, rather than being forced into an answer by a reporter on a bad day.