PR Skills AI Cannot Replace: Six Moments From the Real Work

The PR skills AI cannot replace are the ones that depend on context nobody wrote down, consequences somebody has to own, and history between two people. In practice that means six things: news judgment, timing, reporter trust, saying no to a founder, staying calm in a crisis, and translating a hard idea for a specific audience. Writing isn't on the list. I use AI for drafts every week, and none of these six has moved an inch.

Lists like this usually stay abstract. "Judgment" and "relationships" sound nice and mean nothing until you see the moment they decide something. So for each skill I'll describe the kind of moment where it shows up in my work, what a model gives you instead, and how a person actually gets better at it.

Which PR tasks can AI actually do?

Before the six skills, it helps to be honest about the other column. AI is already good at a long list of PR tasks, and pretending otherwise makes the rest of the argument look defensive.

• First drafts of releases, Q&A documents and briefing notes
• Summaries of a reporter's recent articles before a pitch
• Turning one founder interview into several posts
• Cleaning up a media list's formatting and spotting duplicates
• Producing ten subject-line variations so you can reject nine

Every item on that list is production. It happens after someone has decided what the story is, who it's for and when it goes out. That deciding is where the six skills live. If you want the full task-by-task split, I've mapped it in how to use AI in a PR workflow.

The replaceability test: three questions for any PR task

I use a quick test when someone asks whether a piece of comms work can be handed to a tool. It's three questions, and a single yes keeps the task with a person.

1. Does it need context nobody wrote down? Market mood this week, what a reporter said off the record last month, which investor is sensitive about which topic.
2. Does it carry consequences someone has to own? If the output goes wrong, who answers for it to the founder, the community or the press?
3. Does it depend on history between two people? A reporter taking your call, a founder trusting your "no".

A press release draft fails all three questions, which is why AI handles it fine. Deciding whether to send that release this week passes the first two. The six skills below are the ones that pass at least one question every single time.

The six PR skills AI cannot replace

The scenes are typical moments, not one specific client story.

Skill 1: News judgment, deciding what's actually a story

The scene: a founder sends me a launch brief on a Monday. It's a feature update, the team is proud of it, and they want CoinDesk and The Block by Thursday. My job in that first hour is to decide whether there's a story in it at all, and if not, whether there's a story next to it.

every other launch is important to the company & promotional to the media of it doesn't have a high impact value associated w it! Shilika Jain on X, 11 Oct 2024

A model will happily write a pitch for anything you hand it. It has no reason to tell you the announcement is promotional, because it can't see the forty other pitches in a reporter's inbox that morning. News judgment is knowing what an editor will consider new, and that comes from reading coverage daily and watching which pitches land.

What good looks like: when we did launch PR for Rari Chain's mainnet, the whole push hung on one angle, creator royalties protected at chain level, and that single angle carried it into CoinDesk, The Block and Cointelegraph in the same week. Choosing that angle over the ten other true things about the launch was the work. I broke the sequence down in how Rari Chain landed its tier-1 placements.

Skill 2: Timing, reading the week before you announce

The scene: everything is ready. Embargo set, quotes approved. Then the market drops hard overnight, or a competitor gets exploited, or a big regulatory headline eats the news cycle. Do you hold, shift by a week, or reframe?

AI can tell you what happened in the news. It can't weigh how your announcement will read next to it, how much attention your target reporters have left, or whether moving the date costs you an exclusive you promised.

Timing cuts the other way too. For Fluence, a global Microsoft outage turned into an opening: a piece in e27 tied the outage to the case for decentralized compute, and that newsjack gave an infrastructure story a reason to run outside crypto media. Spotting that kind of window while it is still open, and having the founder's view ready, is a timing skill. A model would have needed someone to ask it the question first.

How you build it: keep a running calendar of what your target outlets covered each week, and note what you'd have done differently. After a few months you start to feel the rhythm.

Skill 3: Reporter trust built over years

The scene: a reporter you've worked with for years messages you directly. They're writing a piece on your sector and want a founder who'll say something real, on a tight deadline. They came to you because last time you sent someone useful, and the time before that you told them honestly that your client had nothing worth their time.

lot of PR folks are only focused on driving coverage, only a handful focus on building genuine media relationships Shilika Jain on X, 25 Jun 2026

Nothing about that message can be generated. AI can write a polite email to a reporter. It can't be the person the reporter already trusts, and it can't decide which of your clients deserves that slot. I go deeper on how this relationship works in journalists are not on your payroll.

How you build it: send fewer pitches, make each one specific to the reporter's beat, answer their questions fast even when there's nothing in it for you, and never oversell a client to them twice.

Skill 4: Saying no to the founder

The scene: a founder wants to post a sharp reply to a critic at 11pm, or announce a partnership that isn't signed, or put a big number in a release that legal hasn't checked. The most useful sentence I'll say that week is "we're not putting that out."

your values & ethics take you a long way Shilika Jain on X, 29 Apr 2023

A drafting tool does what it's asked. It will polish the angry reply and make it land harder. Pushing back takes standing with the founder, a clear reason, and an alternative they can live with. I usually bring all three: why it's risky, what it could cost, and what we can say instead that gets them most of what they wanted.

The cost is real. Saying no sometimes ends an engagement. I'd still rather lose a client than let them put out something I know will hurt them.

Skill 5: Crisis calm while facts are still moving

The scene: something has gone wrong. Facts are incomplete, the community is loud, reporters are asking for comment, and every sentence will be read by lawyers, investors and users at once.

AI helps here with monitoring and with drafting variants fast. It doesn't decide what you can say before you know it's true, who hears first, or when silence is the better answer. I cover that split in detail in whether AI can handle crisis communications, and the full response sequence lives in the crypto crisis communications playbook. The skill itself is mostly temperament: slowing a panicking room down without slowing the response.

Skill 6: Translating a hard idea for one specific reader

The scene: a founder explains their product in protocol terms. A business reporter needs it in two sentences a general reader can follow. A crypto-native reporter needs the technical detail that shows it's real. Same product, two different explanations, and both have to be accurate.

AI can produce a "simple" version. It often gets the simplification wrong in ways only someone who understands both the product and the reader will catch. Translation is judgment about what you can safely drop.

How do the six skills compare?

When the skills argument doesn't hold

I don't want to overstate this. If your comms work is mostly volume, like pushing the same release to a long list or producing daily social copy from fixed templates, a lot of it can be automated, and the people doing only that work should be learning new skills now. Very early teams with no news and no reporter relationships may also get more from a founder writing well with AI help than from hiring anyone. These six skills earn their cost once there's real news, real risk or a real audience to protect.

FAQ

Which of these skills is hardest to learn?

Reporter trust, because it only compounds with time and consistent behaviour. You can study news judgment and timing, and you can practise translation. Trust needs years of being useful and honest to the same people.

Can AI help me practise these skills?

Partly. You can use it to role-play a skeptical reporter's questions or stress-test a message for weak spots. It's a sparring partner for preparation, not a substitute for the real conversation.

Do founders need these skills themselves?

Some of them. Founders should develop translation and a basic sense of news judgment, since they're the ones speaking to reporters. Timing, reporter relationships and the discipline to say no are usually easier to get from a senior comms person who isn't emotionally attached to the announcement.

Should I hire for these skills or for writing speed?

Hire for the skills. Writing speed is now cheap because every candidate has the same tools. Ask candidates about a time they talked a founder out of something, or picked one angle over others, and listen for the reasoning.

If you want a senior operator making these calls on your next announcement, see how I run AI startup PR or book a 30-minute teardown.