> ## Content Index > Fetch the complete content index at: https://intelligence.artificialrealitymedia.com/llms.txt > Use this file to discover other available public pages before exploring further. # AI is bad at media lists, and the reason is not data - URL: https://intelligence.artificialrealitymedia.com/ai-media-list-building-pr-gap/ - Published: 2026-09-02T12:02:00.000Z - Updated: 2026-09-02T12:02:00.000Z - Description: Media list building resists AI because the task is predicting an unwritten story from private signals. Sam Polstein's freshness threshold: three weeks. - Author: Galina Fendikevich - Tags: Public relations, #For AI Companies, #What needs to be built, #Import 2026-09-03 13:12 Media list building is the PR workflow AI has not taken, and the reason is not missing data. Sam Polstein, who leads communications and content for deep tech startups at HAUS, still has his team build lists by hand: "For whatever reason, I find AI to be pretty bad at this." His threshold for stale information is three weeks — beyond that, he calls the output basically useless. This is the failure he expected to be solved first. > "To be honest, I'm disappointed. Maybe I could put this in the 'let's somebody please fix this and get some engineers working on a better tool for this.'" ## Why does the media list decide the campaign? Because targeting dominates craft. Polstein puts list building above the pitch itself in importance. > "Building a media list in comms is incredibly important. Who you're reaching out to is — the same way context engineering is crucial to the output of using an LLM — building a media list is the hard work that makes or breaks a PR campaign. Because if you're reaching the wrong people, it doesn't matter how good your pitch is." The failure mode is specific. > "The amount of times over my career that I've gotten responses from journalists that are like: okay, fine, but I don't cover that. Or: cool, but my editor's not gonna let me talk about this." ## Why is this not a retrieval problem? Because the target is an article that does not exist yet. Retrieval finds what a journalist has already written, which Polstein says is often a reason to skip them rather than pitch them. > "Building a media list has a lot of tangential thinking involved. It's a little bit more of an art than a science. It's figuring out not what a journalist has covered before — because often if they've already covered it, they'll be like: I already wrote that story, sorry man, on to the next one." > "You're trying to base \[it on\] the vibe of their beat and what their editors are telling them to write, and then helping them with their next story that they've never written about, usually." An editor's direction to a reporter is not published anywhere. Neither is a beat a journalist is quietly trying to move into. The determining inputs are private, which is why a task that looks like search behaves like judgment. Polstein's workaround is lateral: look at competitors, and better still at "peer companies that are somewhat interesting in a similar vein but aren't directly competitive." ## How fresh does the data have to be? Three weeks is the line. Recency is not a quality bonus in this workflow; it is the entire signal. > "I don't care what a reporter's written about six months ago, for the most part. It's: what are the last few stories? Because I'm trying to guess what they're writing next. So what's top of mind for them, and what have they covered recently? If they've covered it really recently they might not want to cover it again, even if it's very relevant for them." > "If there's a three week long lag in the information that's feeding into [Claude](https://claude.ai/?ref=intelligence.artificialrealitymedia.com), well, that's basically useless." That is a sharper standard than most products in this category are built against. Not stale in months. Useless at three weeks. ## Do [Muck Rack](https://muckrack.com/?ref=intelligence.artificialrealitymedia.com) and [Cision](https://www.cision.com/?ref=intelligence.artificialrealitymedia.com) already solve this? Not in Polstein's experience, though they are closer than a general-purpose model. > "There are tools and platforms that we use like Muck Rack and Cision that offer these sorts of things. They are CRMs for exactly what we're describing, and I've used all of them throughout my career. I think though that that sort of information, it's still not that easy to come by, still somewhat out of date — although less so than if I was just prompting Claude or [ChatGPT](https://chatgpt.com/?ref=intelligence.artificialrealitymedia.com)." So the operator's actual choice is between a purpose-built database that is somewhat out of date and a general model that is more out of date. Neither does the reasoning step. ## What is the product he is describing? Not a ranked list. An argument. Polstein describes the real workflow as a back-and-forth on Slack with colleagues — someone proposes a name, someone else notes what that reporter just wrote and that they have moved outlets and started a new beat. Agreement is useless in that exchange, and agreeableness is what he gets. > "It can be a little too effusive as well. And I want somebody who's gonna push back and tell me when I'm wrong." > "I would love a Claude that was as good, or close to as good, as a PR person who has a few years of experience. And it's just not quite there in that task." ## What this means if you are building it Three requirements fall out of what Polstein describes, and they are unusually well specified for a product request: - **A freshness guarantee, stated in days.** Under three weeks, visibly, per source. Nobody in this market is selling against that number. - **Prediction, not history.** The output is a claim about the next story, with the evidence attached — recent coverage, beat drift, outlet moves — not a relevance score against past clips. - **Disagreement as a feature.** The system that says "not her, she filed that same angle nine days ago" is worth more than the one that returns two hundred ranked names. The lesson generalises past PR. Any task whose real inputs are relational and unpublished will look tractable and resist automation. If you are building for one of those tasks, the first question is not how to rank better. It is where you get the information nobody publishes. ## Where this comes from S1E4: He runs PR for deep tech startups — the full interview with Sam Polstein. Listen or watch: [YouTube](https://www.youtube.com/watch?v=Oo5rT9aS2RQ&ref=intelligence.artificialrealitymedia.com), [Spotify](https://open.spotify.com/episode/3QYcgc5ezkSMvH36WJd7qv?ref=intelligence.artificialrealitymedia.com) or [Apple Podcasts](https://podcasts.apple.com/us/podcast/he-runs-pr-for-deep-tech-startups-if-your-agency-isnt/id6780699025?i=1000773200320&ref=intelligence.artificialrealitymedia.com).