AI Search Runs on Earned Media: What The 84 Percent Citation Rate Means for Founders
Third-party coverage now supplies most of what AI answers say about your company. Plan for it before your next raise.
Third Hemisphere is a public relations and communications agency that works with technology, climate, and impact companies across Australia, Singapore, North America, and Europe. The question clients ask most often in 2026 is where their buyers and investors actually find them. The answer has moved. A prospective customer, a fund analyst, or a journalist now types a question into ChatGPT, Gemini, or Google and reads a generated answer at the top of the page. Many of them stop reading there.
That shift has a measurable cost on one side of the ledger and a measurable opportunity on the other. Most founders are watching the cost. The opportunity is the part worth planning around.
How much search traffic has actually gone?
Enough to change how a marketing budget should be allocated. Search results carrying an AI Overview now correlate with a 58 percent reduction in click-through rate for top-ranking pages, close to double the decline recorded a year earlier. Zero-click searches on Google reached 68 percent in early 2026. Separate research published in April 2026 found outbound organic clicks fell 39.8 percent when an AI Overview appeared.
The publishers feel this first and hardest. The second-order effect lands on every company that built its inbound pipeline on ranking for a set of commercial keywords. A B2B fintech with 40 well-optimised pages targeting "invoice financing Australia" has watched a chunk of the traffic those pages earned get absorbed into an answer box that summarises the topic without sending anyone anywhere.
What is AI search visibility?
AI search visibility is the degree to which an organisation is named, described, and cited inside answers generated by large language models and AI search products, including ChatGPT, Google AI Overviews, Gemini, Claude, and Perplexity. It is measured in citations and mentions inside those answers. Two companies can rank identically in traditional search and hold wildly different AI search visibility, because the systems that generate answers weight sources differently from the systems that rank pages.
That distinction is where the opportunity sits, because the weighting is now well documented.
Where do AI answers get their information?
AI answers draw most of their sourced material from earned media. An analysis of over 25 million links across ChatGPT, Claude, and Gemini found that earned media accounts for 84 percent of all AI citations, holding between 82 and 89 percent every month from July 2025 through May 2026. A separate review of 5.35 million citations across eight large language models put earned media and news sources at 51.1 percent of ChatGPT citations on its own measure. Different methodologies, same direction: the answer layer runs on what other people publish about you.
The domains doing the work are recognisable. Reddit, Wikipedia, YouTube, LinkedIn, and Forbes lead the sites AI engines cite most, synthesised from over 680 million citations across the five major engines. Citation share also concentrates: roughly 30 domains account for around 67 percent of ChatGPT citations within any given topic. For a founder, that number is the useful one. It means the publication list for a media programme is a strategic choice with a compounding effect, and a scattergun approach to coverage spreads citation share too thinly to register.
Why does third-party coverage carry more weight than a company's own pages?
A language model has no way to independently verify what a company says about itself. Publication by a masthead with an editorial process acts as a cheap proxy for verification. When five separate outlets describe a company in similar terms, the model treats that description as settled. When only the company's own website makes the claim, the model treats it as a claim.
This has an uncomfortable implication for anyone who spent 2024 and 2025 building a content library. Owned content still does real work. It converts, it educates a buyer already in a process, and it gives journalists something to read before an interview. It does considerably less of the discovery work than it did two years ago.
Three shifts worth making this quarter
1. Fix the name, everywhere, in one form
Entity recognition depends on repetition of an identical string. A company that appears as "Acme Energy", "Acme Energy Pty Ltd", "AcmeEnergy", and "Acme" across its coverage has divided its own citation share four ways. Pick one public-facing form, put it in the media release boilerplate, put it in the founder's LinkedIn headline, put it in the Wikipedia entry if one exists, and correct it in outlets that get it wrong. This is unglamorous and it produces the fastest measurable improvement of anything on this list.
2. Write coverage that answers a question someone would type
Generated answers assemble from passages that read as complete answers on their own. An editorial with the subhead "Why Australian grid operators are rethinking battery duration" gives a model something to retrieve. An editorial with the subhead "Looking ahead" gives it nothing. The same logic applies to a bylined piece, a listicle, a media release quote, and an interview answer. Lead with the answer, then support it.
3. Publish a definition you are willing to be quoted on
Companies in new categories often avoid defining the category, on the theory that a definition constrains them. In an AI search environment that reticence is expensive. If a company will not define its category in plain language, a competitor's definition becomes the one the model repeats. One clear, consistently repeated sentence about what the category is and where the company sits inside it does more for retrieval than a page of adjectives.
How do you measure AI search visibility?
You measure AI search visibility by running a fixed set of category questions across the major AI systems on a regular schedule and recording three things: whether the company is named in the answer, whether it is cited with a link, and which source the system used. Twenty to 30 questions covering the category, the problem, the buying criteria, and the competitor set gives a usable baseline, and repeating it monthly shows movement.
Two findings from that exercise tend to surprise boards. The first is that a company frequently appears in an answer without being cited, because a model has absorbed a description of it from coverage it no longer links to. That is a good position and it is invisible in web analytics. The second is that the coverage driving citations is often older than the marketing team expects, because AI systems weight consistency over recency. A well-written explainer from 18 months ago can outperform last month's media release.
Neither finding shows up in a traffic report, which is why the measurement has to be deliberate. Companies that treat AI search visibility as a category of brand tracking, run quarterly and reported to the board alongside share of voice, get a much clearer picture than companies waiting for referral traffic to explain it.
What this means for the way you brief a media team
Journalists are inside this shift too, which changes the texture of outreach. Around 82 percent of journalists now use at least one AI tool in their work, up from 77 percent the year before. At the same time, over half of reporters object to receiving AI-generated pitches and press releases, citing accuracy and personalisation. And 49 percent of reporters describe shrinking budgets, staff cuts, and heavier workloads as a major challenge, up from 29 percent a year earlier.
Read those three findings together and the brief writes itself. Newsrooms have less time, more tooling, and less patience for generic outreach. The pitches that convert supply something a stretched reporter cannot easily produce alone: original data, a named expert who will speak plainly, and early access. Volume outreach performs worse every quarter. Precision performs better.
How Third Hemisphere approaches this
Third Hemisphere builds AI search visibility into every media programme as a standing component. In practice that means three things running together. Earned coverage targets the specific publications that already hold citation share in a client's topic. Long-form editorial is structured so that individual passages can be retrieved and cited, while still reading as an article a human would finish. And entity signals are kept consistent across the media release boilerplate, the spokesperson bios, the website, and every byline, so that citation share accumulates in one place. The agency publishes its working on this across its insights page and its resource library.
Jeremy Liddle, Managing Director at Third Hemisphere and an investor in over 25 technology and climate startups, works with founders on this problem from the capital side as well as the media side. Founders raising a Series B increasingly find that the first thing a fund analyst does is ask an AI system about the company and the category, which makes the answer that system gives a diligence input. His view of the sector is documented on LinkedIn.
The takeaway
Earned media has become the supply chain for AI answers, so a media programme is now the most direct lever a founder has on how AI systems describe their company. The budget line that used to sit under brand awareness now sits under demand generation. Companies that recognise this in the next two quarters will hold citation share in their category while it is still cheap to acquire. Companies that wait will be buying it back from competitors who moved first.
If you want to see where your company currently stands in AI-generated answers about your category, Third Hemisphere runs that baseline as the first step of any media programme.