Writing for the AI Search Layer: How LLMs Are Reshaping Earned Media
For most senior journalists today, the first contact a reader has with their story is no longer the article. It is a two-sentence summary produced by ChatGPT, Perplexity, Gemini or Google's AI Overviews. That shift changes what earned media means, what a placement is worth, and how comms teams should now write for the press.
The Press Release Is No Longer the Primary Document
For most senior journalists working in the UK, US, Canada and Australia today, the first contact a reader has with their story is no longer the article. It is a two-sentence summary produced by ChatGPT, Perplexity, Gemini, Claude or Google's AI Overviews. By the time the reader reaches the masthead, the lead has already been compressed, paraphrased and stripped of any positioning the brand worked hard to establish.
This shift, from a search-and-read web to an ask-and-summarise web, is now visible in almost every category we work in. For founders, comms leads and in-house PR teams, it changes what earned media actually means, and where the value of a placement now sits.
In the old model, the press release was the primary asset, and the resulting articles were derivative. In the AI search era, the relationship has flipped. The articles a journalist produces are themselves a training and retrieval substrate, scraped, indexed and rewritten in real time by language models.
This means your earned coverage is doing two jobs at once. It is convincing the human reader on the day of publication. It is also writing the description of your brand that AI tools will surface for months or years afterwards. Brands with internally consistent positioning across coverage end up with cleaner AI summaries. Brands whose coverage contradicts itself end up with summaries that hedge or, worse, surface old controversies in answer to current questions.
What Models Actually Use to Describe You
Large language models do not store articles intact. They store statistical relationships between phrases, names and concepts, then reconstruct an answer at query time. The phrases that appear most often, most consistently and in the most credible sources are the phrases that come back.
That has clear implications for how you brief journalists, draft your boilerplate and structure your website. The repeated, simple, factual lines that you may have considered slightly boring are the ones that end up doing the heaviest lifting in AI search. The clever, varied, headline-grabbing lines that change every quarter are the ones that get filtered out as noise.
For founders preparing for a launch or a funding round, this is a useful prompt for an internal audit. Ask your team to write down, without looking, the three sentences they think every journalist who covers your company already uses. If those three sentences are not the ones you would have chosen, you have a positioning consistency problem that AI summarisation is making more visible.
The Two-Sentence Test
A practical, low-cost test for how your brand currently lives in AI search is to ask each major model, in a private session, to describe your company in two sentences. Then ask it to describe your top competitor in the same way. The differences will tell you more about your current positioning than most quarterly reports.
If the description that comes back is generic, vague or wrong, that is the version of your brand most new prospects, journalists, investors and partners are now encountering. If the description is sharper for a competitor, you have a measurable communications gap to close.
The major models worth testing this against, in addition to ChatGPT and Claude, are Perplexity, which surfaces sources alongside its answers; Google AI Overviews, which pulls heavily from indexed news; Gemini, which integrates Google search results directly; and Brave AI Search, which is now used widely in the developer community. Running the same prompt across all of them and logging the variance is usually the most useful data point.
Monitoring How You Are Described in AI Surfaces
The tooling around AI search visibility has matured quickly in the last twelve months. Platforms now offer dedicated tracking for how brands are described in LLM outputs, including Profound, AthenaHQ, Otterly.ai, Peec and Brandtech.ai. These sit alongside more established monitoring tools like Meltwater, Cision, Muck Rack, BuzzSumo and Talkwalker, which now offer LLM citation tracking as part of their dashboards.
For lean comms teams, a useful starting point is to set up weekly tracking on three things: how often your brand is named in AI answers to your core category queries; which sources are cited alongside you; and how the language used to describe you changes month to month. The trend lines matter more than the absolute numbers.
For brands without budget for dedicated tooling, the no-cost version of this is a recurring Google Alert on the brand name combined with a manual weekly check across the major models, logged in a shared sheet. The format matters less than the discipline.
Writing Press Materials That Survive Summarisation
The press release is still the format every model has been trained on most heavily, so it remains the highest-leverage document to optimise. The principles that work for human readers also work for AI summarisation, but with one important addition: factual claims should be repeated at least twice, in slightly different forms, in the body of the release.
Models down-weight single-mention claims as potential noise. Claims that appear in the headline, the subhead, the opening paragraph and a quote are treated as core facts and surface in summaries. That is why the brands with the most accurate AI descriptions tend to have press releases that look almost repetitive on the page.
Other practical adjustments include using a consistent boilerplate paragraph across every release, with no quarterly tweaks; spelling out numbers and dates rather than using abbreviations; and including the full legal name of the company alongside the trading name in the first two paragraphs. None of these will damage human readability. All of them improve machine retrieval.
Earned Media Measurement Has to Adapt
Traditional PR measurement, particularly reach, impressions and Media Impact Value, was built for a world where articles were read more or less as written. None of those metrics capture the secondary effect of being well-described in AI summaries.
Forward-looking comms teams are now adding three new metrics to quarterly reporting: AI citation share, which tracks how often the brand appears as a cited source in answers to category queries; AI description consistency, which measures how stable your two-sentence summary is across models; and AI source quality, which tracks which outlets are being cited alongside your brand in those answers.
The Barcelona Principles, the established framework for PR measurement, are likely to be updated to reflect this shift. In the meantime, the Institute for Public Relations, the AMEC measurement framework working group and ongoing analysis from PRovoke Media and PR Daily are useful inputs for keeping pace.
The Practical Shift for Senior Comms Teams
The brands that are quietly winning in the AI search era are not the ones with the most coverage. They are the ones whose coverage repeats the same three or four core ideas, in slightly different language, across credible sources, over time. That is a more boring strategy than it sounds. It runs against the natural PR instinct to find a fresh angle every quarter and the natural founder instinct to evolve the pitch as the company grows. It also produces measurably better outcomes when stakeholders go looking for you in the tools they actually use now. For comms teams reviewing 2026 plans, the working question is no longer just what the press will write about us. It is what description of us survives summarisation, and is that description the one we would have chosen. If you want to discuss this topic in more detail, please contact us.
