Earned Media: The Coverage ai Engines Actually Read

Earned media is coverage about your brand that you neither paid for nor published: press articles, analyst notes, reviews, community threads, podcasts, videos, and third-party listings created by people under no obligation to be kind. It has become one of the main inputs to what AI engines say about your category, because a generated answer can draw on independent sources that corroborate or contradict what you publish yourself.

Ahrefs tested how closely a mention on heavily linked pages tracks with appearing in AI answers. Across the fifty most-mentioned websites in each system, it found a rank correlation of 0.70 in Google's AI Overviews, 0.40 in Perplexity, and 0.12 in ChatGPT. Ahrefs published the study in July 2025, drawing on roughly 76.7 million AI Overviews and about a million prompts in each of the other two systems. Similarweb read US ChatGPT prompts and put the share of answers carrying any citation at about 1.6% in June 2025, rising to roughly 6.8% by May 2026. Professional services still sits under 4% (Similarweb, July 2026). Most of the time, any influence your coverage has will arrive with no visible link attached to it.

Key Takeaways

• Earned media is the part of your story you cannot edit: someone else wrote it, someone else hosts it, and that lack of control is exactly what gives it weight beyond your own website.
• Paid, owned, and earned describe who paid for or controls the coverage. AI adds another question: can the system reach and read it?
• Chasing citations is the wrong target. Under 7% of US ChatGPT responses carried a citation in May 2026, so citations are too sparse to measure the full influence of earned coverage. A source can help shape the sentence without appearing in the footnote.
• Coverage in one major publication will not carry equally across every engine. The same accurate claim repeated across independent sources is less dependent on any one engine’s retrieval preferences.
• A publisher’s robots.txt affects your distribution. OpenAI says an opted-out page will not appear in ChatGPT search answers, while Anthropic says blocking its search crawler may reduce visibility. The publisher controls that access, not you.

What Earned Media Is

Earned media is any coverage of your brand published by someone under no obligation to publish it.

Three conditions define it.

1. You did not pay for the placement.
2. You do not host the page.
3. You cannot edit the wording.

PR is the work of earning attention. Earned media is the coverage that work produces. Confusing the two is why a PR team and an SEO team can report on the same asset for a year and never agree on its worth.

What makes the coverage earned is who says it. Google's Search Quality Rater Guidelines, dated 11 September 2025, tell raters to research reputation from independent reviews, expert references, and news articles. The instruction is blunt. Where a site's account of itself disagrees with reputable independent sources, trust the independent sources.

That document describes human raters, not a ranking system. Its own opening states that no single rating can directly change how a page appears in Google Search. It does not prove how an AI engine selects its sources. It does make the hierarchy plain: independent reputation evidence carries weight that self-description cannot.

Earned, Owned and Paid: The Model Rebuilt for AI Answers

Marketers have sorted channels into earned, owned, and paid media for a decade. The split answers a budgeting question: who pays, and for what. AMEC, the industry body for communication measurement, already treats three buckets as too few. Its Barcelona Principles 4.0 (AMEC, June 2025) tells practitioners to define the paid, earned, shared, and owned channels that influence each audience. Its channel list now names "search data or AI search engine outputs" outright. A measurement standards body is treating the generated answer as a channel you report on.

For AI visibility, two more questions matter: did an independent source say it, and can the system retrieve it? Your own website is usually easy to fetch, but it is not independent. Earned coverage is independent, but it may sit behind a paywall or block the crawlers trying to reach it.

The final column explains why PR belongs beside the blog. Independent sources can confirm or challenge what you publish. Everything on your own site is still your version of the story.

What Counts as Earned Media Now

Earned media now appears in more formats. What matters is not the format but whether the person publishing it was free to say no. The earned media examples below run from the most formally accountable author to the least.

The order is not an impact ranking. Our separate breakdown of the off-site signals that move an answer already compares their contribution, so repeating that ranking here would add nothing.

Four Kinds of Earned Coverage

• Editorial coverage. A journalist, analyst, or trade editor writing under a masthead's standards. The most accountable author in the set, and usually the hardest coverage to secure.
• Practitioner accounts. Someone who used the product and wrote about it in a forum, a newsletter, or a video, with no relationship to you. Low formality, high specificity, and the phrasing matches how buyers ask.
• Structured third-party records. Review profiles, category directories, and encyclopedic entries, where the platform fixes the format and a customer supplies the content.
• Aggregated verdicts. Comparison articles and shortlists where inclusion was somebody else's judgment call. A shortlist page already resembles the answer a shortlist query calls for.

An AI answer is a downstream result, not a fifth kind of earned media. Another system may encounter a published or indexed answer later, but the answer itself does not meet the definition above.

Two things often get mistaken for earned media. A sponsored post may be easy for an AI engine to retrieve, but the placement is still paid rather than independent. A customer quote you drafted and a customer approved is still your wording on someone else's letterhead. Neither is disqualified from working. Both belong in a different column.

Why Corroboration Across Sources Beats Any Single Placement

Coverage in one major publication can feel like enough. The data says otherwise. The relationship between prominent coverage and AI visibility looks different in every engine tested. Ahrefs took the fifty most-mentioned websites in each system and measured how closely being named on heavily linked pages tracks with appearing in that system's answers. The result runs from strong to almost nothing.

Source: Ahrefs, July 2025, June 2025 data. These are Spearman rank correlations across the top fifty domains in each system, not causes, and the study's author says so directly.

Read the ChatGPT row twice. Optimizing only for prominent placements follows the strongest pattern in Google AI Overviews but a much weaker one in Perplexity and ChatGPT. Coverage in one major publication will not carry the same weight with every AI engine. That is the practical case for corroboration: the same accurate claim repeated across independent, retrievable sources is less exposed to any one engine’s source preferences.

Brand authority here is not a score anyone assigns you. It grows when independent sources agree about what you do. It compounds slowly, and no single purchase buys it. The related guide to why on-site work hits a ceiling explains why making your own site readable is only the starting point.

How to Value Earned Media When the Reader Is a Machine

PR teams used to value earned coverage by calculating what the same space would have cost as an advertisement. This is called the advertising value equivalent, or AVE. AMEC has rejected it since 2010, and the fifth principle of the current edition is blunter than the versions before it: "Invalid measures such as advertising value equivalents (AVEs) should not be used. Instead measure and evaluate the contribution of communication by its outcome and impact." The same section extends the ban to any multiplier applied in the calculation.

A machine reader makes the AVE worse, not better. An advertisement runs once and stops. An earned page can be retrieved and reused across related questions long after publication. Comparing its value with a one-time advertising rate measures the wrong thing.

I would rather report five simple numbers than one combined score, because a combined score hides what actually changed.

What to Record Instead

• Independent sources naming you, counted. The denominator for everything else, and the number that moves brand authority.
• Sources naming a competitor but not you. Count the gap in pages. This is the line that gets a budget approved.
• Domains appearing in the answers to your buying questions. Run the questions, record the domains, and mark which ones mention you.
• How each source describes you. Read the page, not only the AI answer. If the source gets your brand wrong, the answer may repeat it.
• Referral sessions from AI assistants. Configure your analytics to identify them before you need to report on them.

A useful measure of earned media value is how many accessible, independent sources name your brand. That is a count another team can check against the same prompt set, and it does not pretend to be revenue. The metrics worth tracking covers how these counts turn into a dashboard, and setting a baseline worth reporting covers taking it to a board that still thinks in impressions.

Getting Earned Coverage in Front of the Crawlers

One part of content distribution is easy to overlook. The page carrying your best coverage is governed by somebody else's robots.txt. You cannot edit that file. You will not see it unless you go and look. OpenAI and Anthropic both document what happens when it says no.

Sources: OpenAI’s published crawler documentation and Anthropic’s crawler and opt-out page, both read in August 2026. OpenAI notes the switches are independent, so a publisher can allow search and refuse training, and that a robots.txt update takes about 24 hours to reach its search systems. One exception is documented and it matters. OpenAI says ChatGPT-User fetches are started by a person, so "robots.txt rules may not apply" to them. A block is a strong signal rather than a sealed door.

Publishers were already using these switches at scale by the end of 2023. When the Reuters Institute last ran a cross-country count, 48% of the most widely used news websites across ten countries were blocking OpenAI's crawlers by the end of 2023, rising to 79% in the US. Not one site in the study reversed the decision that year.

Read the scope before you use the number. That study counted blocks affecting GPTBot, ChatGPT-User, or both. OAI-SearchBot did not exist yet. The count is from 2023, so do not present it as the current blocking rate. What it establishes is that crawler access became a publishing decision years ago. Check the host before you value the coverage, favor venues that stay readable, and never assume a prestigious byline makes the full page retrievable.

Building an Earned Media Program

An earned media strategy that survives a full quarter has three moving parts and one sequencing rule. Start from the source list, not the target list. Run the twenty questions closest to a buying decision across the assistants your buyers use. Record every domain the answers name, and mark which of them mention you. That list is the plan. The patterns in this article are a starting point; your own source data decides what happens next.

Then separate direct requests from longer-term relationship building. Review profiles, directory entries, and inclusion in comparison articles are requests you can take to a specific person. Editorial coverage and trust within a community take time to build. Earned visibility is harder to measure in professional services. Its buyers ask peers first, and it has the lowest citation rate among the categories Similarweb tracked. That makes visible citations a particularly weak proxy for whether earned coverage is influencing the answer.

The Three Parts, and Who Owns Each

1. Coverage. Earning coverage worth citing through pitches, contributor work, and review requests. Campaign mechanics belong in a dedicated digital PR discussion rather than this one.
2. Access. Checking whether an AI engine can retrieve the coverage and choosing venues with that in mind. This is where content distribution extends beyond your own channels.
3. Measurement. Recording the five counts above on a fixed schedule and comparing them with the previous quarter.

Two failure modes account for most stalled programs. The first is having an accurate source list but nobody available to pursue the coverage. That is where monitoring stops short. The second is treating community work as a campaign with an end date. A thread becomes citable through community trust that can take months to build. A 90-day operating plan keeps monitoring and execution on the same schedule.

How Klarivo Earns Coverage That Engines Cite

Klarivo does not stop at tracking third-party coverage. It does the work needed to earn more of it. Klarivo Monitor runs targeted buyer queries across ChatGPT, Claude, Perplexity, Gemini, and Grok, updated on a schedule you control. It reports the domains the engines drew on rather than a single score. That domain list is the source list this article keeps pointing at, so next quarter's priorities come out of last quarter's evidence. Klarivo’s published FAQ makes the distinction clear: SEO improves the website you control, while Klarivo builds visibility across the third-party sources you do not.

Most teams can identify the sources they need to appear in. Far fewer have the people to earn that coverage. That is why a source list can sit untouched for two quarters. Klarivo’s global contributor network works across dozens of subreddits, with human contributors taking action every week and following each community’s norms. The work is not automated. Klarivo says clients can begin seeing commercial movement within one to two months. Our published FAQ puts corrective content and review support in the same service, because a wrong description on a source an engine already reads costs more than an absence.

A team already earning consistent coverage may only need tracking. If it has the capacity to act on what the data shows, it can keep the work in-house. Klarivo may also be a poor fit when nobody discusses the problem publicly. The first week should establish whether those conversations exist.

Want to see which domains your category's answers are built from? Book a 15-minute Klarivo discovery call. Choose a time that works for you and receive instant confirmation. Tracking runs in Klarivo Monitor, while the Reddit program covers community and review work. See how the full program works.

Frequently Asked Questions About Earned Media

Is earned media the same thing as PR?

No, and the difference matters when you are deciding who owns a budget line. PR is broader than earned media. It also covers crisis response, spokesperson management, and internal communication. Earned media is one result of that work, and PR teams are not the only teams that can produce it. A customer who leaves an independent review is generating earned media. So is a practitioner who discusses the product without being paid, or a partner who chooses to list it on their site. Not all of it has to sit under PR.

What does earned media cost if nobody pays for the placement?

The placement is free and the program is not. You are paying for the people, the time, and the assets that make coverage worth publishing, and the market prices that as a retainer rather than a media buy. Ask any vendor which of the three parts above it actually handles. An agency that only pursues coverage may still leave access checks and measurement to your team. Category rates vary too widely to quote a useful figure, and anyone quoting one without seeing your category is guessing.

Does a paywalled or gated article still count?

Partly. It may reach subscribers while remaining inaccessible to the systems you want to influence. Neither OpenAI nor Anthropic documents how its agents handle paywalled text, so the safe working assumption is that what sits behind the wall is not retrievable. Treat a paywalled placement as a reputation asset rather than a retrieval one. Then ask the publisher whether an excerpt, a summary page, or a syndicated version sits outside the wall. This varies by outlet, so you have to check each one.

How long does earned coverage take to show up in an AI answer?

Nobody has published a reliable figure. The closest published benchmark is OpenAI’s statement that a robots.txt change takes about 24 hours to reach its search systems. Everything after that depends on crawl frequency for that host, whether the engine is retrieving live or answering from training, and how contested the query is. Measure it on your own prompt set rather than accepting a vendor's number.

What happens when a publisher deletes or rewrites the page?

You lose the asset, and you find out late. That loss of control is the price of earning independent coverage. It is also why a program built on five placements is fragile and one built on fifty is far less so. Keep a dated record of every placement and re-check the high-value ones quarterly. Treat durability as a selection criterion when you choose where to pitch.