Welcome back to the Bi-weekly Frequency
Through here, you’ll be able to access a bi-weekly roundup on all things B2B tech, algorithmic changes on the most popular communication channels, and top news of the week - analysed by members of our team.
Last time, we looked at what’s working on LinkedIn and unpacked the latest AI and geopolitics headlines. This week, we’re following where the algorithms are going: why LinkedIn is gaining ground as a source for AI search, and what a tougher safety bar for frontier models means for tech companies and their comms.
Algorithmic changes in AI search
In September, the AI search tools stopped treating "social" as one single thing and started picking its favourites.
OtterlyAI's September index, which runs 1,600 prompts daily against seven engines, found LinkedIn was cited 39% more often by the end of the month than at the start. It gained on five of those seven engines, led by ChatGPT (up 66%) and Google AI Mode (up 62%). X barely registers as a source at all.
A model decides what to quote, so give it something quotable.
A post with one concrete detail, like a number or a named customer, can be lifted straight into an answer, and a vague opinion can't. This also tells you what isn't working: betting on one platform. Reddit is the largest social source overall, but that's mostly Google's doing; ChatGPT's use of it is down 94% since July, and Claude has never cited it, according to industry discussion.
A caveat, though, is that social is a small prize. It took roughly 4.3% of AI citations in September, down 16% from August, while company-owned pages took about 58%. LinkedIn is rising but inside a shrinking slice.
For comms teams: write the LinkedIn post so a machine could quote it. That means one clear claim, backed by a number or a name, in plain words that still make sense when lifted out of the post. Then put the fuller version on a page you own, because company sites still get far more AI citations than any social platform.
News of the week: Frontier AI and the safety bar
The biggest AI launch of the week was the one that didn't happen. OpenAI scrapped GPT-6.1 Astra, due in ChatGPT and Codex in October, after internal testing raised safety concerns. The model was more capable but less honest about what it had done, and it pushed ahead on tasks without asking permission. For B2B tech, the reflection is that there is a rising concern over safety, and failing to meet the bar can cost you a launch.
The same week, another big player restricted a release under the same claims. Google published Gemini 4 Argon, which can find, validate and patch critical software vulnerabilities on its own. It was limited to a vetted group of cybersecurity experts, though, with wider access coming later.
For buyers and their comms teams, the lesson is that "most capable yet" now comes with conditions. Stick to what you can prove today: which model is available, who can use it, and what it does well. Avoid promises that depend on the next model, because its release date can slip.
On the policy side, President Trump and the heads of Google, Anthropic, Meta, OpenAI, Nvidia and SpaceX signed a voluntary accord on controls and audits for their most advanced models. It sounds like Big Tech agreeing to safety rules, but it carries little penalties. A day later, the FTC confirmed it is investigating OpenAI, Anthropic and other AI companies over the risks their products pose. The industry promised to police itself, and within 24 hours a regulator showed it wasn't relying simply on that. It is likely we will see more voluntary pledges to be turned compulsory.
The bottom line for tech PR and comms teams:
- AI search tools are picking favourites: LinkedIn citations rose 39% in September, while X barely registers and ChatGPT's use of Reddit is down 94% since July.
- Social is still a small prize (about 4.3% of AI citations against roughly 58% for company-owned pages), so write the LinkedIn post so a machine could quote it, then put the fuller version on a page you own.
- Calling something "safe" doesn't count for much anymore. Someone has to decide it is, and they do it in public. "Our agents act within these permissions" holds up under questioning, and "our AI is safe" won't.
- Clients should pick the claims they can stand behind now, before a regulator, a safety evaluation or a vendor's release schedule picks for them.
We’ll be back in two weeks to cut through what’s changing across B2B tech – and what’s actually worth paying attention to - stay tuned
