An AI PR agency earns coverage for companies building or applying artificial intelligence. It is the most competitive beat in technology right now, and the difficulty is not reporter interest — interest is enormous — but differentiation. Every company in every category now describes itself as AI-native, so the phrase carries no information. What earns coverage is a specific, verifiable claim about what the system does that something else cannot.
Why AI is the hardest category to stand out in
The claim is universal. When every pitch says AI-powered, the words stop functioning as a signal. Reporters filter on what is left after you remove them.
Reporters have become technically literate. The AI beat is now staffed by people who understand the difference between a model, a wrapper and a prompt. Vague architecture claims get caught.
Benchmarks are contested. Claiming state-of-the-art results invites scrutiny of your evaluation method. Publish the method or expect the number to be discounted.
The story moved from capability to consequence. Coverage has shifted toward employment, cost, energy, safety and procurement. A pure capability announcement competes with frontier lab releases and loses.
What reply rate should an AI PR agency get?
Responses to our outreach between December 2025 and August 2026, by company type:
| Company shape | Contacted | Open rate | Response rate |
|---|---|---|---|
| Established name, category past its news cycle | 269 | 54.5% | 11.4% |
| Pre-revenue, launching into a category the press doesn't cover yet | 351 | 54.3% | 10.6% |
| Late stage, recognised founder, crowded category | 700 | 48.3% | 8.5% |
| Market leader in a vertical with specialist trade press | 494 | 47.9% | 8.2% |
Every response, auto-replies included, across the four programmes we could classify this way.
AI companies sit squarely in the third row at 8.5% — late stage, recognisable founder, crowded category. This is the counterintuitive finding worth sitting with: the hottest category in technology produces a below-average reply rate, because attention and competition rise together. Being in a hot category does not make you easier to cover. It makes you one of forty pitches that reporter received that morning.
"We ran the same playbook for a pre-revenue startup nobody had heard of and a company approaching $100M ARR. Reply rates came out three points apart. Size doesn't buy you the reply and neither does a famous founder. That's not what agencies tell you in the pitch meeting."
Justin Mauldin, Founder, Salient PR
How far the funnel actually gets
Of every 100 pitches delivered, 51 register as opened. That understates it — a tracking pixel only fires if the client loads remote images, and privacy-first clients block or proxy them by default. So we work on roughly 7 in 10 rather than 51%. That is our estimate and not a measurement — every blocked pixel is a read nobody counted, which makes the recorded rate a floor.
"Our recorded open rate is 51%. The real number is closer to 7 in 10, because a big share of reporters read mail in clients that block tracking pixels. We know it happens because we get replies from people our system says never opened the email. Treat any open rate you're quoted as a floor, including ours."
Justin Mauldin, Founder, Salient PR
Follow-ups do less than you think
| Sequence step | Delivered | Open rate | Reply rate | Replies |
|---|---|---|---|---|
| First email | 2,063 | 47.6% | 6.0% | 124 |
| Follow-up 1 | 1,576 | 37.4% | 3.7% | 59 |
| Follow-up 2 | 69 | 40.6% | 4.3% | 3 |
Two-thirds of every reply came from the first email.
Using AI to do PR, which is a different question
Every agency now claims to use AI. Most of what that describes is drafting, which is the part that matters least.
"AI writes a fine first draft and a terrible pitch. The part that matters is knowing that this specific reporter wrote about this specific thing six weeks ago. No model knows that about your beat unless someone fed it."
Justin Mauldin, Founder, Salient PR
The useful applications are unglamorous: keeping a media database current, checking whether a reporter still covers a beat, classifying replies so the programme learns. Drafting the email is the easy part and doing it badly at scale is how you get blacklisted.
"Volume doesn't fix a bad list. Sending 200 pitches to people who don't cover you gets you the same result as sending 20, except now you're on more block lists."
Kayla Haines, Media Relations, Salient PR
What actually gets covered in AI
Deployment numbers from real customers. Not a demo. What happened when a named organisation ran it in production, including what broke.
Cost and energy. What inference actually costs you, what it costs to serve, how that has changed. This is under-covered relative to demand and you have the data.
Evaluation methodology. How you measure whether it works, published in enough detail to be criticised. This earns respect from technical reporters in a way benchmark claims do not.
A specific limitation. The single fastest way to be taken seriously in this category is to be precise about what your system fails at.
Watch for an AI premium
Be sceptical of a premium charged for "AI expertise" that turns out to be the same programme with different vocabulary. Ask what they will do differently for you than for a database company.
How Salient engages
We ask for a six-month commitment to start, because the first two months are relationship building and judging the work before then tells you nothing. After that there is a 30-day exit, no questions asked, either side. If it is not working we would both rather know.
How to judge an AI PR agency
Ask which AI reporters they placed in the last six months, by name. The beat is well-defined and checkable.
Ask them to explain your product back to you. In this category an agency that cannot articulate the technical difference will write pitches that get filtered.
Ask what they would refuse to claim on your behalf.
Ask for reply rates in AI specifically. Given the 8.5% figure above, a firm quoting you 20% in this category is not measuring the same thing.
Does AI search change AI PR?
Recursively, yes — and this category is where it matters most, because your buyers are the most likely of anyone to ask an AI which vendors to consider.
A peer-reviewed study of generative-engine optimisation (KDD 2024) found that adding direct quotations, statistics and cited sources measurably improved how often a page was surfaced in generative answers — by up to 40% on the authors' benchmark, with the size of the effect varying by subject area.
The practical version: if an AI system cannot find substantive third-party writing about what your model actually does, it will describe your category without you in it. Earned coverage is now training data, and in AI specifically your buyers are querying that training data before they ever reach your website.
The short version
- Everyone claims AI. The claim carries no information. Specificity does.
- AI companies see 8.5% reply rates — mid-pack, not top, because attention and competition rise together.
- Roughly 7 in 10 actually read it once you adjust for pixel blocking.
- Two-thirds of replies come from the first email.
- Deployment numbers, real costs, published evaluation methods and named limitations get covered. Capability claims do not.
- An agency using AI to draft pitches is automating the least important step.
Salient PR runs AI and technology PR programmes. All outreach figures come from our own database covering December 2025 to August 2026 and are updated quarterly.
Sources
- Salient PR outreach database, Dec 2025 – Aug 2026
- Aggarwal et al., GEO: Generative Engine Optimization, KDD
If you want this run by people who do it every day — Salient PR is a technology PR agency working with AI and machine-learning companies. See how we work or start a conversation.
Related reading
About Salient PR
Salient is a B2B tech PR agency. We help startups become the story — landing tier-one coverage, building executive visibility, and turning attention into pipeline.
Senior team. No hand-offs. Radical transparency. If you’re ready to talk, reach out.
How AI PR works when every company claims to be AI-native, what reporters actually cover, and the reply-rate data behind it.
