What it means for us
A weekly intelligence brief for leadership communicators, speechwriters and executive advisors.
A significant number of readers work inside closed or managed AI environments — Microsoft 365 Copilot, government-approved systems, or enterprise deployments where tool access is limited. You are not forgotten. Where a story or tool has direct relevance to closed-system users, I flag it. I’ll do the same where something simply isn’t accessible inside a managed environment. Sign up here.
One of the more challenging aspects of writing this brief is deciding what is relevant to communicators and what isn’t. Frankly, everything kind of relates to communicators. The new AI models, the governance issues, the inevitable screw-ups, the myriad ways agentic tasks can revolutionize everything we do.
Yet here is what I’m seeing — in our workshops, in our monthly sessions, in my day-to-day with other communicators.
This is becoming a game of model-chasing, instead of asking ourselves the real question: do I know how to use the tools I already have?
Exhibit A: you don’t need Mythos to write a set of media lines.
It’s like hiring a construction crew to put up an IKEA shelf. Sure, they’ll do it for you. But you’re paying top dollar for 50 people to stand around while your daughter just sighs and puts it up (unless things are different at your house than mine).
The point? I’m narrowing the kind of stories that show up here.
Five stories. Five things you can do this week.
Drawn, quartered, delivered fresh.
The journalists have it right: set task-level AI rules instead of a general policy
Nothing gets a good gripe going in our workshops quite like participants describing their company’s “AI Policy”. In short — most are useless. This shift to task-specific rules points in another direction.
The Associated Press has just updated its newsroom standards to name the specific uses it permits: early-stage research, document summarization, transcription, translation, headline suggestions, story summaries, shot lists, and help with grammar, spelling and search optimization.
Journalists must review and edit anything AI touches, and generative AI remains barred from creating or altering news photography. Material AI involvement must be disclosed where it is relevant to understanding how the content was produced. There is even a section on AI coding assistants — the part most comms policies never think to write.
Source: Associated Press — updated newsroom standards for artificial intelligence, July 2026. Coverage free.
We’ve talked about this a lot in our workshops. Bring AI into the open. Discuss how we use it and how we don’t — not as a catch-all, but task by task. For a speech: condensing absurdly long policy reports, checking claims back against the source document, checking grammar. For the first draft? For the final sign-off? For speaking in our principal’s voice? Never.
Directly relevant. Microsoft 365, Copilot and bespoke enterprise systems still need written rules defining permitted tasks, mandatory review, disclosure thresholds and prohibited content types. The platform being locked down is not a policy.
Audit your public-interest content for AI disclosure before 2 August
Nothing says “read this” like “audit”, right? But bear with me — this stuff matters to communicators.
On 20 July the European Commission adopted its final guidelines — 51 pages, non-binding — on Article 50 of the EU AI Act. Providers must tell users when they are interacting with AI and apply machine-readable markings to AI-generated or manipulated content. Deployers must disclose deepfakes and certain AI-generated public-interest text where it has not passed through human review or editorial control. The obligations apply from 2 August 2026.
Here is the detail for us. For text on matters of public interest, the clock runs on the date of publication, not the date of generation. Something drafted in July and published in August still needs labelling, unless it qualifies for the editorial-control exception. So the piece sitting in your queue right now counts.
One caveat. The Digital Omnibus on AI, signed 8 July and awaiting publication in the Official Journal, would push the date to 2 December for systems already on the market. Until it lands, 2 August is the date you plan against.
Source: European Commission, 20 July 2026 — Guidelines on transparency obligations for providers and deployers of AI systems. Primary source, free.
Relevant. The disclosure test attaches to the published content and its review process, not to the system that produced it. An internal Microsoft, government or bespoke environment does not exempt you.
Test whether your AI summaries keep attribution and meaning intact
Delivering the Andrew Olle Media Lecture at ABC Studios in Sydney on 21 July, Reuters Editor-in-Chief Alessandra Galloni said the quiet part loudly: “Our journalism should be licensed, not taken.” Her argument is that news organizations should sell verified reporting to AI companies, provided it is properly represented, attributed and compensated.
The backdrop: the Reuters Institute Digital News Report 2026 puts weekly use of AI chatbots for news at 10% globally, up from 7%, and at 16% among under-35s.
Source: Alessandra Galloni, 2026 Andrew Olle Media Lecture, 21 July 2026, Sydney. Coverage free. Audience figures originate in the Reuters Institute Digital News Report 2026.
Directly relevant, and closer to home than it looks. Reuters licenses its text archive to Meta and Microsoft Copilot — so the licensed news your internal assistant is summarizing may well be hers. Run a check on whether those summaries retain source attribution, qualifications and intended meaning.
Prepare your lines now on which AI tools touch personnel decisions
Twenty-six Meta employees filed suit in federal court in Oakland on 13 July, alleging that internal AI systems — keystroke and activity monitoring, AI token-usage dashboards, algorithmically assisted performance rankings — were used to select people for a layoff round covering roughly 10% of the workforce.
The complaint argues those metrics cannot, by design, be accumulated by someone on protected medical or family leave, or by someone whose output is reduced by a disability, and that no leave-neutral individual review was run before the scores were acted on. The plaintiffs, who filed anonymously, were notified in May that their jobs would end on 22 July, and asked the court to block the terminations while they pursue their claims in arbitration. Meta says the claims lack merit and that workforce decisions were made by people, not AI.
Source: Reuters, 14 July 2026 — Meta used AI to target workers with medical conditions for layoffs, lawsuit claims. Free. Fuller detail at CBS News.
These are allegations. Untested in court. At the same time, questions that look HR-only today land on comms desks tomorrow. Get this one on your radar now, while there is nothing to answer for.
Closed environments are more exposed here, not less. Managed platforms generate exactly the telemetry at issue — usage dashboards, activity logs, seat-level adoption reports — and those numbers are usually commissioned by IT or HR without anyone asking whether they will one day need a public explanation.
Strip “it’s not X, it’s Y” out of every executive draft this week
Will Oremus reports in The Atlantic that negative parallelism — the “it’s not X; it’s Y” construction — has become the most recognizable tell of machine-written prose, and that models from every major lab converge on it without anyone being able to fully explain why.
Barron’s counted the construction in Fortune 500 filings, climbing from 50 in 2023 to more than 200 in 2025. In one Washington Post dataset, variations of “not just X, but Y” turned up in roughly 6% of all messages in a single month. Elyas Masrour, a founding engineer at the detection firm Pangram, says every major chatbot leans on it to some degree, and estimates it appears about three times as often in AI writing as in human writing (it’s not just a hypothesis, it’s a pattern! — sorry).
Worse: German researchers now find the same tics turning up in spontaneous human conversation. We are catching it from them.
There is a deeper version of this, and it’s the one worth carrying into a room. Sandra Matz and colleagues at Columbia Business School tested what happens when people hand identity-defining choices to LLM agents, using 110,000 real choices from 1,000 US users. Both generic and personalized agents pulled choices toward more popular options, compressing how distinctive any one person’s preferences were. The counterintuitive finding: the personalized agent — the one built to preserve your distinctiveness — compressed the range more sharply than the generic one, by optimizing for the most likely option inside that person’s own typical set.
Which is to say: the tic is the visible symptom of a much larger flattening. Watch for the construction, because you can see it. Worry about the flattening, because you can’t.
Sources: Will Oremus, The Atlantic, July 2026 — It’s Not Just Annoying; It’s Inescapable (paywalled). Supporting research: Matz, Horton & Goethals, The Basic B*** Effect, arXiv 2509.02910, September 2025. For the writing-specific version of the finding, see Doshi & Hauser, Generative AI enhances individual creativity but reduces the collective diversity of novel content, Science Advances, 2024 — peer-reviewed.
Directly relevant, and the one item closed-system readers can act on today without any external tool. Copilot drafts inside Microsoft 365 carry the same tic, and a bespoke model fine-tuned on your executive’s back catalogue is precisely the personalized-agent case Matz found compresses range hardest.
An argument I’ve been making since AI began is that far from erasing speechwriters from the comms landscape, it’s going to raise the value of great speechwriters even further. In the land of AI-flattened speeches, the writer who can hold originality and voice is king. Story 5 underlines it.
I made that argument in a workshop this week and got an excellent response: do speakers want their voices to stand out, or are they fine with AI speech slop?
As with most things, it depends. A speechwriter got in touch with us a few weeks ago to say their principal wondered aloud why he couldn’t simply replace the writers with a chatbot. Other, more clear-eyed leaders value the role of a — you know — human with real thoughts, strategic insight and a distinct voice.
A hell of an answer, I know. Some yes, some no. That’s why I’m paid the millions.
What I do know is this. The chatbots will last about as long as the first screw-up.
Readers of this fine weekly screed will recall last week’s news about Ford “unretiring” a feral pack of humans (long live the greybeards!) because the AI wasn’t up to snuff. Or the Canadian province of Quebec, which has abandoned automation and artificial intelligence initiatives across the public sector after encountering “significant challenges.” These are a few from pages of examples.
As I wrote in response to a LinkedIn post last week:
Keep the faith, everyone! Firms are still hiring communicators, will continue to hire communicators, and communicators will continue to thrive — regardless of tech or challenge. Non ducor, duco!
One core message, three angles.
This is the audience stress-test from my free 45-minute workshop earlier this month. If you missed it, the recording is yours.
Get the workshop recording →Here’s the setup. A technology-sector leader is speaking on a climate panel at a conference hosted by a small island developing state. The core message on the page reads:
“Data centres powering AI are beneficial for people and planet — the quickest way to leapfrog to a sustainable future. It’s time to stop complaining about the energy these centres use, and start celebrating the progress they make possible.”
Confident. Clear. And it will land very differently depending on who is sitting in row three.
Paste this into whichever model you use:
Role: You are a senior communications strategist helping test a message against audience resistance. Core message: "Data centres powering AI are beneficial for people and planet — the quickest way to leapfrog to a sustainable future. It's time to stop complaining about the energy these centres use, and start celebrating the progress they make possible." Context: A technology-sector leader is delivering this on a climate panel at a conference hosted by a small island developing state. Task: Read this message as three specific members of that audience: (1) a small-island climate delegate who is sceptical; (2) a clean-energy investor who is broadly sympathetic; (3) a local resident whose community the message overlooks. For each: 1. What is their likely objection? 2. What are they listening for that the message doesn't address? 3. What specific phrase most likely triggers resistance? 4. How should it be reframed for this audience without changing the core truth? Reframe, don't spin — same truth, different doorway. Name the blind spot even when it's uncomfortable. Be specific. Use realistic objections based on how these stakeholders actually behave, not hypothetical concerns. Do not hallucinate. Do not infer.
The instruction that does the heavy lifting is reframe, don’t spin. Without it, the model will quietly soften the message until it means nothing. With it, the truth stays put and only the doorway moves.
Run it on your own core message this week. Then read what comes back as the person you’d least like to have in the audience.
Works anywhere. No external tool, no web access, no upload. Copilot will run this happily inside Microsoft 365.
We’re here for you.
Have a question or concern? Interested in a workshop? Need a speech written? We have what ails you.
I’ve built a custom assistant that helps me pull stories daily. I then read the list, hack 90% of the useless and boring B.S. out, add some comments (snide and otherwise), and verify whatever makes the final copy with my human brain. If you’re not seeing the “bigger” stories, that’s because I’m more interested in narrowing down what matters to communicators.
I also use AI to design this newsletter and this portion of my website, because you definitely don’t want me designing anything. Initials carved into tree trunks look better than my design skills.
And don’t forget to ask others exactly how they use AI in the products they send to you.
AI for Leadership Communications · Geneva
Further details and all weekly newsletters available at brentkerrigan.com/what-it-means-for-us.