--- title: "How I Would Buy Performance PR Now" description: "I would tie a PR fee to citations now that earned media can be counted. Muck Rack put earned media at 84 percent of AI citations and paid at 0.3 percent." canonical: "https://jaxonparrott.com/blog/performance-based-pr-ai-citations-measurable-2026" date: "2026-08-07" last-updated: "2026-11-04" --- # How I Would Buy Performance PR Now I would buy PR against a citation rate I can recount, and I would keep a retainer only for work whose outcome cannot be counted yet. Performance-based PR was a slogan while nobody could show what a placement did after it went live. Citation data gives the fee a unit: how often ChatGPT, Claude, Perplexity, and Gemini use the placement when a buyer asks a question in the category. The account of the model I would send a finance lead is [Performance PR in 2026: How It Works and What It Covers](https://authoritytech.io/blog/performance-pr-future-media-relations-2026). That page is the deeper method. This essay is how I would use it as a founder: which number has to move, which risk sits with the firm, and what would send me back to a retainer. ## What the retainer was selling [About 85 percent of PR agencies charge monthly retainers](https://www.jenniferbett.com/the-spin/how-much-does-pr-cost-for-startups-in-2026), typically $5,000 to $25,000 a month for startups, with four- to six-month minimums. The fee is due whether or not a piece runs. Pitching, lists, and reporting calls are real labor. They are not the outcome I need, which is simpler: when a prospect asks an assistant who leads the category, the company is in the answer, attached to a claim. In my judgment, earned media in a publication a buyer already trusts is the strongest trust signal in this market. The broken piece was the commercial wrapper. Time was for sale because results were not countable. They are countable now, imperfectly, and imperfectly is enough to move the fee. ## The counts I would put in the contract [Muck Rack's Generative Pulse report](https://aimagazine.com/globenewswire/3290268), May 2026, analyzed more than 25 million links cited by ChatGPT, Claude, and Gemini across 17 industries. Earned media drove 84 percent of the citations. Paid and advertorial content accounted for 0.3 percent. Journalism alone was 27 percent of cited sources. Earned media stayed between 82 and 89 percent across the three editions since July 2025. The October 2026 edition kept earned and third-party sources above 84 percent and paid and advertorial at 0.3 percent, with journalism down to 22 percent from 27 ([Muck Rack, October 2026](https://muckrack.com/blog/what-is-ai-reading-october-2026)). [Meltwater's GenAI Lens](https://www.meltwater.com/en/blog/ai-search-visibility-april-may-2026), on more than 8 million citations across eight major LLMs, found earned and news media at 37.6 percent of AI citations in May 2026, and press releases at 0.2 percent. Muck Rack and Meltwater define "earned" differently and measure different citation sets, so I would not average 84 and 37.6. I would read each against its own definition. Meltwater puts the press release near zero, and Muck Rack puts paid and advertorial content near zero. The distribution wire is not the asset. The earned placement is the asset. When someone asks an assistant for the best vendor in a category, the answer follows whatever credible publication wrote a claim the engine was willing to lift. That mechanism is old. The new fact is that the lift is countable. [LeadCoverage's quarter-long experiment](https://www.globenewswire.com/news-release/2026/07/27/3333617/0/en/Data-Led-PR-Programs-Earn-3-5x-More-Artificial-Intelligence-Citations.html), a self-reported program from a logistics-sector agency, published July 2026, shipped one release a week and tracked citations. In its own release, LeadCoverage said releases leading with a specific economic datapoint earned 3.5 times the AI citations of releases that did not; announcement, personnel, and award releases earned almost none. The tools cited the company's data 1,058 times in the quarter. Search impressions rose 83 percent. Kara Brown, the CEO, told GlobeNewswire: AI cannot invent a number, so it cites whoever published one. I would put a specific figure, not a range and not the phrase "significant growth," in anything I expected to be cited. The unit I would write into a performance fee is citation count, citation rate, and share of citation inside the category, tied to named placements. The terms for those units are in the [glossary](/glossary). The loop from placement to retrieval to a rate I can re-read is the work on [Machine Relations](/machine-relations). [A guest opinion piece by PenVine CEO Jennifer Schenberg in Marketing Dive](https://www.marketingdive.com/news/a-cmos-guide-to-machine-relations/826421/), August 2026, uses the term, which is one sign the name has left my own pages. The fee structure does not depend on the name. It depends on the count. ## How I would score a firm | What I am comparing | Retainer | Pay per placement | Pay tied to citations | |---|---|---|---| | Price | Fixed monthly fee | Fee when a piece runs | Fee when the citation outcome lands | | Measured object | Impressions, ad-value equivalents, clips | Placement count and outlet tier | Citation rate, confidence, share of citation | | Success | Activity completed | Coverage secured | The brand is cited when prospects ask category questions | | Proof | A report of links | Confirmation the URL is live | Rates across ChatGPT, Claude, Gemini, Perplexity | | Who holds the risk | The founder | Split | The firm, on the portion tied to the rate | Five questions, in the first meeting: 1. Do you measure citation rates for the placements you secure? A no is a 2019 product with a 2026 adjective. 2. Which publications do engines cite in this category? A Forbes piece the engines never retrieve is worth less, for this fee, than a trade publication ChatGPT uses on the category question. 3. Can you show citation data from a current engagement? A firm that measures this can open the file. A firm that changes the subject is answering the question. 4. What is your placement rate? [Propel, a PR software vendor, found journalists responded to 3.15 percent](https://www.bizcommunity.com/article/report-journalists-respond-to-315-of-pitches-podcast-pitch-volume-surges-by-22-789482a) of the 425,000-plus pitches sent through its platform in Q4 2023. A firm that cannot state its own rate is not tracking the top of its funnel, which makes me doubt the bottom. 5. Do you price on outcomes or on time? The answer tells me who carries the miss. ## How I would stop a clip from hiding inside the fee The two big counts do not match, and I would write both into the contract rather than pick the friendlier one. Muck Rack's Generative Pulse, May 2026, read more than 25 million links cited by ChatGPT, Claude, and Gemini across 17 industries. Earned media was 84 percent. Paid and advertorial was 0.3 percent. Journalism was 27 percent. Earned media stayed between 82 and 89 percent across three editions since July 2025. Meltwater's GenAI Lens read more than 8 million citations across eight models and, in May 2026, put earned and news media at 37.6 percent and press releases at 0.2 percent. Muck Rack and Meltwater define "earned" differently and measure different citation sets, so I would read each against its own definition. Both put a self-published or paid category near zero, though the categories differ: zero-point-three for paid and advertorial, zero-point-two for press releases. The thing that was paid for, or the release left to travel alone, is barely in the answer. LeadCoverage's quarter makes the warning operational. One release a week. By LeadCoverage's own count, releases leading with a specific economic datapoint earned 3.5 times the citations of releases that did not. The tools cited the data 1,058 times. Search impressions rose 83 percent. Kara Brown's line, that an engine cannot invent a number so it cites whoever published one, is the editing rule I would put above the fee schedule. A placement with no checkable figure can still be famous. It is a weak unit to pay an outcome fee on. The retainer this fee is replacing, or sharing the contract with, has a published shape. About 85 percent of agencies charge monthly, typically $5,000 to $25,000 for startups, four to six month minimums. Propel, a PR software vendor, found journalists responded to 3.15 percent of the 425,000-plus pitches sent through its platform in Q4 2023. That is one vendor's platform average, not a current industry rate. I would ask the firm for its own rate before I treated 3.15 as theirs. A firm that does not know its pitch conversion is unlikely to know its citation conversion, and the second number is the one the outcome fee depends on. A Marketing Dive guest column from August 2026 is evidence the name of the discipline has traveled. It is not evidence that a particular firm can produce the 1,058-style count. I would ask to see a current engagement's queries, engines, URLs, and rates. A subject change is an answer. I would split the invoice so a clip cannot settle the outcome portion. Retained money can pay for the reporting rhythm. The outcome portion pays when a named URL moves citation rate on the signed query list, on ChatGPT, Claude, Gemini, or Perplexity. A Forbes URL that never appears in an answer and a trade URL that appears on the category question are not the same deliverable, which is why "which publications do engines cite here" is question two and not a courtesy. Question four, the placement rate, keeps the firm from describing effort as a result. Question five, time versus outcomes, tells me who holds the miss when the rate is flat. Three editions of the 82-to-89 percent range are why I would sign as if the citation layer stays. Each new edition is why the contract should have a reread at renewal; the October 2026 edition kept paid and advertorial at 0.3 percent. If paid and advertorial ever take a share that looks like earned media, on a denominator as plain as 25 million links, I would price paid distribution as a citation channel and rewrite this essay's fee. Until an edition shows otherwise, 0.3 percent is the number I would not let a clip book talk over. ## The decision I would move the next PR contract off a pure retainer and onto a split: a smaller retained amount for reporting and maintenance, and a defined fee that pays when named placements move citation rate on a query list we both sign. I would not pay the outcome fee for a clip that never appears in an answer. Scope: ChatGPT, Claude, Gemini, and Perplexity, the category questions we list, and placements the firm can tie to a URL. Muck Rack's 25 million links and three editions since July 2025 are the reason I trust the direction. Meltwater's 0.2 percent for press releases is the reason a wire does not qualify as the outcome. LeadCoverage's self-reported 3.5 times, from one agency's own quarter, is the reason the piece has to carry a number. What would change my mind: a category where paid or advertorial sources, on a count with a denominator, took a share comparable to earned media, rather than Muck Rack's 0.3 percent. I would then price paid distribution as a citation channel and say so. A firm that shows me our URLs cited, with the query, the engine, and the date, earns the outcome fee even if I dislike the outlet. A firm that delivers famous outlets and a flat citation rate does not. If Propel's 3.15 percent pitch-response figure turned out to be the wrong baseline for a firm that only pitches outlets engines already cite, I would replace the baseline. I would not drop the requirement that they have one. Earned media has stayed between 82 and 89 percent across three editions. I would write contracts as if the citation layer stays, and I would reread the latest edition before I renewed. ## FAQ ### What do I mean by performance-based PR? A fee tied to a countable outcome. The older version paid per placement. The version I would buy in 2026 pays, at least in part, when those placements are cited by ChatGPT, Claude, Gemini, or Perplexity on the queries in the contract. ### Which study would I cite in the agreement? Muck Rack's May 2026 Generative Pulse report: more than 25 million links, ChatGPT, Claude, and Gemini, 17 industries, earned media at 84 percent (between 82 and 89 percent across three editions since July 2025), paid and advertorial at 0.3 percent, journalism at 27 percent (25 to 27 percent across those editions). I would cite Meltwater beside it, 8 million citations, earned and news at 37.6 percent, press releases at 0.2 percent, so nobody treats one study's definition as the only one. ### Is this cheaper than $5,000 to $25,000 a month? It is a different risk. The retainer is due either way, usually for four to six months. The outcome fee is due when the rate moves. Total spend can be higher in a year when the work hits, and it should be, because the alternative is paying the retainer for a stack of clips nobody can connect to an answer. ### Who coined the name I am using for the loop? I coined Machine Relations in 2024 for the discipline of making a brand legible and credible to AI-mediated discovery. A guest column in Marketing Dive in August 2026 is an outside use of the term. The contract can say "citation rate" and never use the name. I would still run the work in that order: a placement worth citing, a number inside it, a rate we both can recount. --- *Published August 7, 2026* By [Jaxon Parrott](https://jaxonparrott.com) — Founder & CEO, [AuthorityTech](https://authoritytech.io) Updated 2026-11-04 Part of [Machine Relations](https://jaxonparrott.com/machine-relations) · [Glossary](https://jaxonparrott.com/glossary) · [Founder judgment](https://jaxonparrott.com/topics/founder-judgment) Tags: Machine Relations, AI Search, PR ## Continue reading - [The Category Ownership Trap](https://jaxonparrott.com/blog/category-ownership-trap): Owning a category definition does not mean owning the buyer journey. How I connect Machine Relations to the questions people ask before they know its name. - [Whether I Would Spend to Raise Entity Mass](https://jaxonparrott.com/blog/entity-mass-why-ai-engines-recommend-brands-2026): I would buy cross-domain evidence before more pages, because brand mentions tracked AI visibility more closely than links did in Ahrefs' study. - [I Would Not Stop at Getting the Page Selected](https://jaxonparrott.com/blog/founder-ai-strategy-2026-citation-systems): I would treat a selected citation as unfinished until the answer reuses the page's evidence, and I would build the source layers that make reuse possible. --- *Machine-readable version of [https://jaxonparrott.com/blog/performance-based-pr-ai-citations-measurable-2026](https://jaxonparrott.com/blog/performance-based-pr-ai-citations-measurable-2026)*