Related: AI Communications & GEO pillar · Crisis PR pillar · Citation Share Is the New KPI · Deepfake Defense · The Crisis Communications AI Citation Share Study
An AI-native communications team isn't five specialists surrounded by colleagues who opted out. It's a whole team operating from a raised baseline. The roles built into the AI-native org chart — Chief AI Officer, Head of GEO, AI Visibility Lead, AI Visibility Analyst — are for a few people. The skills below are for everyone.
Four baseline skills now define the job. None of them is technical. All four are table stakes for any communications professional working a brand brief in 2026.
Using an AI tool is typing a request and taking what comes back. Directing it is supplying the context, the constraints, the prior approved examples, the voice samples, and the boundary conditions it needs to produce something usable. The first produces generic prose. The second produces work the client signs off on.
The gap is measurable. Across 5W internal training cohorts in Q1 2026, the same brief produced materially higher first-draft quality when the prompt included structured context — brand voice samples, prior approved deliverables, an explicit do-not-mention list, an audience persona — than when the prompt was a bare "write me a press release about X." The tool capability is identical in both cases. The operator gap is the variable.
Directing is a learnable craft. Anchor the prompt with the brand's actual prior work. State the audience explicitly. Name the format. Give the tool a kill list of phrases the brand never uses. Iterate at least three rounds before accepting output. This is project management adapted to a new collaborator — not engineering.
2. Verifying AI output
Every communications professional now needs reflexive skepticism toward AI-generated facts, quotes, statistics, and named sources. This is not cynicism about the tools. It is basic craft — the same instinct a good professional already applies to any unconfirmed source. The difference is volume: an AI tool can produce thirty plausible-but-wrong facts in ninety seconds.
Two cases set the precedent. The 2023 Mata v. Avianca matter in the Southern District of New York saw a lawyer sanctioned for filing a federal brief that cited six fabricated cases generated by ChatGPT — complete with invented case names, fake docket numbers, and hallucinated judicial reasoning. The 2024 New York Times v. OpenAI complaint surfaced examples of fabricated quotes attributed to real Times reporters. Both cases are now business-school assignable. Neither outcome had to happen. Verification at the source would have caught every fabrication.
The discipline is procedural. Any fact, quote, statistic, or named source that originates from an AI tool must be confirmed in a primary source before it ships. The verification step is not optional. It is the work.
3. Understanding AI visibility
Not every person on a communications team owns the AI visibility program. Every person should understand it. Buyers now research inside ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews. Brands surface in those answers or they don't. The content the team produces either compounds into the citation graph or it doesn't.
The shift is not directional. By late 2025, roughly a third of US consumers began product research inside an AI tool rather than Google. That share is climbing each quarter. A brand that scores below median Citation Share in its category is leaving discovery on the table — and most CMOs don't know the score because the score is not on any dashboard they have ever looked at.
The measurement framework is in Citation Share Is the New KPI. The category benchmarks are in The Crisis Communications AI Citation Share Study and The Beauty Citation Share Index 2026. The team-level implication is simpler than either: every byline, every press release, every social post, every brand page either feeds AI citation share or doesn't. The work is to make the feed deliberate.
4. Knowing the confidentiality line
Everyone who touches client work needs to know, without asking, what can and cannot go into an AI tool. The line is simple — if it isn't public, it doesn't go in unprotected — but it has to be reflexive.
The 2023 Samsung incident set the template. Engineers at Samsung Semiconductor were caught pasting proprietary source code, internal meeting transcripts, and chip-design notes into ChatGPT to debug and summarize work. Samsung banned generative AI tools across the company within weeks. Walmart, Verizon, JPMorgan, Goldman Sachs, Citigroup, Bank of America, and Apple followed with their own restrictions in the months after. The breach surface was not technical. It was a baseline-skills failure on what counts as confidential information.
Defensible practice: never paste client-confidential material into a consumer AI tool. Use only enterprise instances with no-training clauses written into the contract. Treat any prompt as if it could be subpoenaed. The communications professional who treats the AI tool the way a junior associate treats a deposition transcript is the one who keeps the client.
None of this is technical
The four baseline skills above are judgment, verification, awareness, and discretion. There is no coding. There is no engineering. There is no tool-building. They are communications skills applied to a new instrument — the same craft, raised to the level the work now requires.
That is precisely why they are the baseline, not a specialism. A team that treats AI fluency as the job of "the GEO person" or "the AI lead" is leaving ninety-five percent of the work unimproved. The lift compounds only when the floor moves.
How the baseline gets built
Teams don't reach this baseline by sending a memo. They reach it through deliberate, scheduled training on actual client work. Two-hour weekly skills sessions for eight weeks, anchored to real briefs the team is already running, produce measurable change. Memo-based "AI policy" rollouts produce essentially none.
The leadership job is to make the training non-optional and to measure it. Pre/post writing samples. Prompt audits on live deliverables. Verification spot-checks. Confidentiality tabletop exercises. The training is the work, not extra to it.
Continue Reading
