The gap between what your engineers say in a briefing and what a journalist writes about you the next day is where the AI PR story lives. Most AI founders, especially the technical ones, think their job is to close that gap by being more precise. The reverse is usually true. The job is to make the gap useful, not to remove it.
I have spent the last decade working on the line between technical companies and the people who need to write about them. The pattern is consistent enough that I am going to lay it out plainly, with worked examples for foundation models, machine learning infrastructure, and autonomous systems.
A generalist tech reporter at the FT or TechCrunch is not the only person reading your coverage. There are three people in the audience and they want different things from your story.
The engineer wants the description to be accurate enough that her own team would not roll their eyes at it. She does not need the reporter to use the right technical term for every concept. She does need the reporter to not say something obviously wrong.
The journalist wants a story the desk will commission. That story usually has the shape of a structural claim ("AI is shifting where compute matters") supported by a concrete proof point (your company doing the thing the structural claim predicts). Without the structural claim, the story is a product release. Without the proof point, the story is an op-ed.
The buyer is reading because the journalist's framing tells them whether to take the meeting. They want to know whether the company is credible, whether the technology is differentiated, and whether other companies they respect are working with you. They will skim the article and pull two or three impressions. Those impressions decide whether you get a call back.
The translation work is mostly about making each of those three readers walk away with what they need from the same set of words.
The reflex when briefing a generalist reporter is to simplify. "Think of it like a brain." "It is like a really smart auto-complete." These analogies are everywhere and they almost always make the coverage worse.
The problem is not that they are inaccurate. The problem is that they hand the reporter a frame that flattens what is interesting about the company. If a foundation-model company describes itself as "a really smart auto-complete", the reporter has no reason to write about it differently from any other foundation-model company.
Translation, by contrast, finds the structural property of the technology that the journalist can hold onto without losing what makes it specific.
The single most useful sentence in an AI briefing is the one that tells the reporter what the technology cannot do. It earns trust faster than any claim about what it can.
The bad framing: "Our model is more accurate than [competitor's] on benchmarks." This frame fails because the reporter cannot evaluate benchmark claims and the buyer does not care about benchmarks they have not heard of.
The better framing: "We train on data the others cannot access. Our model is consequently the only one that gets [specific high-value use case] right." Two sentences. The first is a structural fact. The second is the consequence the buyer cares about. The journalist now has a story shape ("the data access becomes the moat") and the buyer now has a reason to call.
The bad framing: "We are faster and cheaper than [hyperscaler]." This is true for many ML infra companies and is also approximately useless. The reporter has heard the claim eighty times.
The better framing: "The hyperscalers are optimised for general-purpose workloads. We are optimised for one specific workload pattern that is becoming common because of where AI is going next. For that pattern we are six times faster, but for other workloads we are slower than the hyperscalers." The second framing turns a generic competitive claim into a structural observation about the market. It also says what the company is bad at, which is almost always the sentence that earns trust.
The bad framing: "We are building self-driving cars." Setting aside the timeline credibility issue, this is a frame the reporter has heard from twenty companies that are now gone.
The better framing: "We focus on a narrower category than full autonomy: warehouse logistics with structured environments. In structured environments the engineering problem changes in [specific way], which is why we are deployed in production at three customers while the broader autonomy market is still in research phase." Now the reporter has a way to write about you that is not "another Waymo" or "another Wayve". MIT Technology Review's coverage of structured-environment autonomy shows what good treatment of this framing looks like.
Generalist tech reporters cover dozens of AI companies a year. The companies they go back to repeatedly are the ones whose senior people are useful as experts on topics broader than their own product. The CTO who can explain why the latest open-weights release matters for the whole category, not just for their company, gets called back. The founder who only talks about their own roadmap does not.
This is a strategy, not a personality trait. It can be planned. The senior team chooses two or three topics on which they are genuinely contributing thinking (not just product positioning) and commits to commenting on those topics whenever they come up in the news cycle. Within twelve months, the senior team is one of the reporter's first five calls on those topics. Within twenty-four months, the company is the implied subject of half the reporter's articles in the area.
A short list of patterns that almost always backfire with non-technical media:
The foundation-model companies that have landed strongest in tier-one press over the last year share three patterns:
First, they led with a structural claim about where the field was going (data access becoming the moat) rather than a benchmark number. The reporter could write a category story, not a product story.
Second, they had the CTO available for follow-up questions for the full week after launch, not just on launch day. Six follow-up articles got written that would not have existed if the CTO had gone back into research mode the day after.
Third, they were honest about what their model did badly. The reporter quoted that directly, which gave the rest of the coverage a credibility floor that the competitors did not have.
Talking to non-technical media as an AI company is not a translation problem. It is a framing problem. The work is to find the structural claim that the engineers, the journalist, and the buyer can all walk away with, and to be the company that earned the right to make it. That is the discipline. Everything else is craft on top of it.
Should AI founders avoid making any technical claims to generalist media?
No. Avoid claims that depend on the journalist understanding the underlying technique. Make claims about consequences, market structure, and specific use cases. Those translate cleanly.
How important is benchmark performance in AI PR?
Less than founders expect. Benchmarks matter to analysts and to technical buyers; for press coverage they are usually noise. The structural claim about the company's position matters far more.
Should CEOs of deep-tech companies be media-trained before any briefing?
Yes, even informally. A two-hour session covering the three or four common framing pitfalls (over-claiming, comparing to the wrong reference companies, refusing to name limitations) prevents most of the coverage problems we see. More on Beachhut media training →
When is the right moment for a deep-tech company to start a press strategy?
Six to nine months before the round you want covered, or twelve months before the product launch you want to land. The relationships and the framing both take time, and there is no compressing them. See the Beachhut deep tech offer →
Keep Reading
Five questions to ask in a pitch meeting and the red flags that look like green ones.
15 April 2026
What the senior team actually does in the sixty minutes after a story breaks.
28 April 2026
Most founders call us 48 hours before close. Here's what six weeks actually buys you.
8 March 2026Work With Us
If you are planning a launch, a fundraise, or a category-defining campaign, the press strategy and the product strategy belong on the same page from the start. Talk to our deep tech team.
Let's chat