Software companies increasingly want the models to know them: to describe the product accurately, place it in the right category, and recommend it when a buyer asks. The instinct to reach for PR is correct, but the phrase train the model needs unpacking, because getting the mechanism right is what makes the tactic work. You cannot open ChatGPT and teach it facts about your software directly. What you can do is shape the enormous body of web content that models learn from when they are trained and retrieve from when they answer, and digital PR is one of the best levers for that. Done well, it results in models that genuinely know your software from credible, independent sources. Here is how to do it deliberately.
The short answer
You cannot directly train a third-party model, but you can shape what it knows by influencing the two inputs that matter: the web corpus it learns from during training, and the sources it retrieves at query time. Digital PR shapes both by producing genuine, authoritative, independent coverage, exactly what models absorb and trust, and Ahrefs found that web mentions correlate with AI visibility far more strongly than backlinks (Ahrefs). So leveraging digital PR means earning real coverage that teaches the corpus, not uploading facts to a model. Influence the sources, and you influence what the model says.
The two channels PR feeds
To use PR deliberately, understand the two ways your coverage reaches a model. First, training: models learn from a large snapshot of the web, and OpenAI documents a dedicated crawler, GPTBot, that gathers content for that purpose (OpenAI), so coverage that exists when a model is trained can become part of its baseline knowledge. Second, retrieval: when a model searches live, it pulls current sources through crawlers like OAI-SearchBot, so recent coverage can be surfaced at answer time. Digital PR feeds both channels, which is why it is uniquely powerful: good coverage helps you now via retrieval and later via training.
Why digital PR, not link-building
The instinct of many SEOs is to treat this as link acquisition, but that is the wrong tool. What models weight is mentions and credible coverage, not the link graph, and Ahrefs found link metrics correlate very weakly with AI visibility while mentions correlate strongly (Ahrefs), a distinction explored in do ChatGPT links count as backlinks for SEO. Digital PR is about earning genuine coverage and mentions that describe your software, which is precisely the currency models trust, whereas chasing links for their own sake optimises a signal the models largely ignore. Aim for coverage, not link count.
Tactic one: earn authoritative coverage
The core move is to earn real coverage in outlets the models trust. Ahrefs’ analysis of the most-cited domains in ChatGPT shows models lean heavily on authoritative sources (Ahrefs), so a genuine article about your software in a respected publication is worth far more than a mention on a low-value site. Pitch real stories, product news with substance, milestones, genuinely useful angles, to journalists and outlets your buyers and the models both read. One authoritative feature that describes your software accurately can shape the model’s understanding more than a hundred thin mentions.
A key distinction here separates digital PR from the old press-release habit. Blasting a release across wire services produces many identical, low-value copies that models discount, whereas earning a journalist to write an original story produces a unique, credible mention that actually teaches the corpus something. The difference is pickup, not distribution, exactly the point made in do press releases still boost brand presence inside AI Overviews. So aim your PR effort at earning genuine, original coverage, and treat any wire distribution as a means to that end rather than the goal, because syndicated duplicates teach the model nothing new about your software.
Tactic two: be the quoted expert
Getting your people quoted as experts is a high-leverage PR tactic for AI, because expert commentary ties your brand to a topic in a credible, third-party voice. When your founder or specialists are quoted in articles about your category, the model absorbs your brand as an authority in that space, reinforcing the topical association that drives recommendation. Offer genuine expertise to journalists and commentary platforms, and each quote becomes a small, credible lesson to the corpus about who you are and what you know, which compounds into the authority that answer engines reward, as covered in how do answer engines evaluate authority compared to page rank.
Tactic three: land in comparisons and roundups
Models building recommendations lean on comparison and roundup content, so getting your software genuinely included in credible comparisons directly shapes whether it appears in the model’s shortlist. Pursue legitimate inclusion in category roundups, best-of lists, and comparison articles on trusted sites, by being genuinely worth including. This teaches the model that your software belongs in the consideration set for its category, which is exactly the association you want it to learn. It is earned placement through merit, not paid or fake inclusion, which models and readers see through.
Tactic four: publish original data worth citing
One of the most durable PR assets for AI is original data. When you publish genuine research, surveys, or proprietary statistics, others cite it, and cited data spreads your brand across many authoritative sources as the origin of a fact. Profound’s analysis shows models favour authoritative, well-referenced sources (Profound), and being the original source of a widely-cited data point makes you exactly that. Original data is PR that keeps working, because each new citation of your research is another credible mention teaching the model who produced it.
Tactic five: keep entity and naming consistent
None of this lands cleanly if the model cannot consolidate your coverage into one entity. Use one exact, consistent name for your software across all coverage and your own pages, so the mentions reinforce a single, clear picture rather than fragmenting. Inconsistent naming splits your hard-won coverage and can cause misattribution. Consistency turns scattered mentions into a coherent, cumulative signal, which is the difference between coverage that teaches the model clearly and coverage that confuses it. Include your exact product name, and ideally your category, in the pitch materials you give journalists, so the coverage that results is already using the language you want the model to learn.
PR tactic and what it teaches the model
This table maps each tactic to the lesson it imparts.
| Digital PR tactic | What it teaches the model |
|---|---|
| Authoritative coverage | Your software exists and is credible |
| Expert quotes | You are an authority in your category |
| Comparisons and roundups | You belong in the consideration set |
| Original cited data | You are a trusted source of facts |
| Consistent naming | All of the above is one coherent entity |
Genuine coverage versus manufactured
A warning that determines whether this works: it must be genuine. Fabricated coverage, syndicated boilerplate, fake reviews, and astroturfed mentions are exactly what models and platforms are built to discount, so manufacturing volume does nothing for the corpus and risks backfiring. Real earned media is trusted precisely because it is independent and credible; manufactured mentions are the opposite of the corroboration models look for. Invest in being genuinely worth covering, and let the coverage be real, because only real coverage teaches the model anything durable.
You influence, you do not control
Set expectations honestly. Digital PR influences what the model learns and retrieves, but it does not give you control over its exact output, because the model synthesizes from many sources and its own reasoning. You are shifting the weight of evidence in your favour, not dictating a script. This is the same influence-not-control reality behind all AI visibility work, and it means the goal is to make the accurate, favorable, credible picture of your software the dominant one across the corpus, so the model’s synthesis reflects it, rather than expecting a guaranteed line.
How to measure it
Because PR feeds a slow-moving corpus, measure it patiently and on the right signals. Track your coverage and mentions growing across authoritative sources, watch your presence and how you are described in AI answers over time, and connect improvements to specific PR wins where you can. Live-retrieval effects can show up relatively soon after coverage is crawled; training effects appear only as models are retrained on newer web data, so read this as a compounding trend, not an instant result, using the presence-tracking mindset from how to get cited in ChatGPT.
A worked example
Say your B2B software is barely known to the models. You run a genuine digital-PR program: you publish an original industry survey that several trade outlets cover, get your founder quoted as an expert in category articles, and earn inclusion in two respected comparison roundups, all under one consistent product name. Over the following months, live AI answers begin describing your software accurately and including it in relevant recommendations, drawing on that coverage, and as models retrain, the baseline knowledge improves too. You never trained a model directly; you fed the corpus genuine, authoritative, consistent coverage, and the models learned from it, which is exactly how this works.
Common mistakes
The biggest mistake is thinking you can upload facts to train a model directly; you shape the corpus it learns and retrieves from instead. The second is treating this as link-building when mentions and coverage are the currency. The third is manufacturing coverage, which models discount and which can backfire. The fourth is inconsistent naming that fragments your coverage. The fifth is expecting instant results from a process that compounds as the corpus updates. Earn genuine authoritative coverage consistently, and the models learn your software over time.
The bottom line
How do you leverage digital PR to train language models on your software? Reframe train as shape the corpus, because you influence what models learn during training and retrieve at query time rather than programming them directly. Earn genuine authoritative coverage, get quoted as an expert, land in credible comparisons, publish original cited data, and keep your naming consistent, and you feed both the training and retrieval channels with exactly the trusted, corroborated coverage models absorb and reward. Keep it real, accept that you influence rather than control, measure the trend patiently, and over time the models come to know your software accurately and favorably, from sources they trust rather than from claims you made about yourself.
Frequently asked questions
Can digital PR actually train a language model on my software?
Not directly, there is no way to upload facts into a third-party model. But digital PR shapes the two things that decide what the model knows: the web corpus it learns from during training, and the sources it retrieves at query time. Earning genuine authoritative coverage is how you influence both, so the model ends up knowing your software accurately.
How does digital PR influence what AI says about my software?
By creating independent, authoritative mentions across the web. Models absorb that coverage into training and retrieve it live, and they weight credible, corroborated sources heavily. So real coverage that describes your software accurately and favorably shapes how the model represents and recommends you, far more than links or self-published claims do.
What kind of PR works best for AI visibility?
Genuine earned coverage in authoritative outlets, expert quotes, inclusion in credible comparisons and roundups, and original data others cite, all with consistent naming. It is the opposite of syndicated boilerplate or fake reviews, which models discount. Real, corroborated, third-party coverage is what teaches the model.
How long does PR take to change what a model knows?
It is gradual. Live retrieval can reflect new coverage relatively quickly once crawled, but influence on trained knowledge updates only as models are retrained on newer web data. So treat digital PR as a compounding investment whose effect builds over time, not an instant switch.
