Why Companies Are Increasing PR Budgets in 2026: The AI Citation Effect
92% of surveyed CEOs reported increasing PR investment because of AI search. Here is what is driving the budget shift, what citation studies measured, and how to evaluate results without treating source composition as an engine-selection rule.
Public relations budgets are rising in 2026 as executives respond to a change in how brands are researched and discovered. The budget evidence is clearest about executive behavior: companies are increasing investment because AI-mediated search now matters to their buyers. Citation studies add useful observations about the sources appearing in measured answer sets, but they do not disclose a universal rule for which brand, page, or publication an engine will select.
Delight Labs released the State of PR 2026 report in March 2026, a survey of CEOs and executives at growth-stage companies. 92% of respondents have increased their PR investment directly because of the rise of AI-powered search engines like ChatGPT, Gemini, and Google AI Overviews. Of those, 49% described the increase as significant or extensive. Earned media ranked as the number one tactic respondents believe drives AI citation, chosen by 29% of executives, ahead of technical SEO at 22%.
This is the AI citation effect: executives are treating third-party coverage as part of the evidence layer that AI systems may retrieve when buyers ask for vendor recommendations, category leaders, or product comparisons. PR still serves reputation and human audiences; it also creates independently published material whose presence, retrieval, citation, and relationship to recommendations can be measured. No cited source in this article establishes that a placement determines whether a brand appears in an answer.
Key Takeaways
- 92% of CEOs have increased PR investment specifically because of AI search, according to Delight Labs' State of PR 2026 report (March 2026)
- Muck Rack's December 2025 Generative Pulse analysis reported sources brands neither own nor pay for at 82% of cited links in its measured sample, and non-paid sources — a wider category that still includes brand-owned content — at about 94%; its May 2026 edition put the non-owned, non-paid share at 84%. Those are source-composition shares, not a promise that earned coverage causes citation
- 83% of B2B marketing decision-makers expect marketing investments to rise over the next 12 months, according to Forrester's 2026 CMO Budget Planning Guide
- The median revenue threshold at first PR investment has dropped sharply since 2023, signaling that earned media is no longer viewed as a late-stage luxury
- Consistent coverage creates a larger body of independently published material to test for retrieval, citation, brand mention, and recommendation across a dated query set
- This shift is a Machine Relations measurement problem: track placements, retrieval, citations, recommendations, referrals, and business outcomes as separate units rather than assuming one produces the next
The Data Behind the Budget Shift
The Delight Labs survey captures something that has been visible inside individual companies for 12 to 18 months but has not been named clearly at scale: executives are not increasing PR spend because the traditional arguments for brand awareness have gotten more compelling. They are increasing it because they have run the query themselves.
Open ChatGPT and ask who the top three players are in your category. Then ask Perplexity for a vendor comparison. Record which brands appear, which claims are attributed, and which exact URLs are cited. Comparing that answer set with each brand's owned and earned source footprint can reveal useful associations and gaps. It does not, by itself, identify the engine's selection mechanism or show that adding a placement will cause a recommendation.
The Muck Rack Generative Pulse analysis examined more than one million cited links in AI responses as of December 2025. Within that measured citation sample, 82% of links came from sources the brand neither owned nor paid for, and about 94% were non-paid — a wider category that still includes brand-owned content; more than half came from sources published in the preceding 12 months, and the highest observed citation rate was within seven days of publication. The measured unit is the composition and age of links already present in the sampled responses. It does not establish that earned coverage or recency caused selection, that a specific publication will be retrieved, or that a placement will produce a brand recommendation.
The Fullintel-UConn study, presented at the International Public Relations Research Conference in March 2026, measured source types across 400 prompts on a single AI platform, all on one health topic. Third-party news and informational sources represented 47% of cited links in that sample; corporate, university and health-network sites represented 48%. These are source-composition observations within the study's sampled responses. They do not establish an engine-selection mechanism, prove that journalism is the primary input, or show that coverage causes a brand citation, recommendation, or business result.
Why Earned Media Works Differently for Machines Than It Did for Humans
Earned media differs from owned content in a practical way: it is published by a third party and can supply independently attributable claims for both human readers and machine-mediated research. Whether an engine retrieves or cites a given placement depends on the engine, query, index, freshness, and source set. The studies cited here observe outputs; they do not reveal an identical human-and-machine trust mechanism.
AI systems can use training data, retrieval indexes, and other provider-specific components when generating answers. The Ahrefs analysis of ChatGPT's most cited pages reported that 65.3% of the already-cited pages in its bounded inventory came from domains with a domain rating of 80 or above. The measured unit is the domain-rating distribution of pages that had already been cited, not a domain-rating mechanism. The inventory does not establish that a high domain rating causes selection, that third-party validation is the primary signal, or that a DR80+ placement will be cited or recommended.
Gartner predicted in February 2024 that traditional search engine volume would decline 25% by 2026 as AI chatbots and virtual agents absorb research tasks. That forecast explains why teams are measuring AI-mediated discovery now; it does not specify how an engine weights third-party, owned, recent, or high-domain-rating sources. Those roles must be observed in a named engine and query set rather than inferred from a general trust analogy.
The practical implication is to maintain a durable, attributable record across owned and independent sources, then test how that record appears in current answer systems. Historical coverage may be present in training or retrieval corpora, but publishers and brands generally cannot observe its model weight. The defensible unit is the measured answer: brand mention, attributed claim, cited host, exact cited URL, and recommendation language at a recorded time.
How the Budget Math Has Changed
The shift in when companies are investing in PR is as significant as the shift in why. The Delight Labs State of PR 2026 report found that the median revenue threshold at first PR investment dropped sharply from 2023 to 2026. That survey documents an earlier budgeting decision among respondents. It does not establish that earlier PR spend causes AI visibility, pipeline, or revenue; those outcomes require separate measurement against a dated baseline.
The Cision Inside PR 2026 report, drawn from nearly 600 PR professionals across the U.S. and UK, shows the same pressure from the practitioner side. 91% of PR professionals reported using generative AI in their workflows, while 32% of senior executives named revenue and ROI as their top priority. That makes measurement more important, not causality automatic: placements, retrieval, citations, referrals, and commercial outcomes remain distinct stages.
Forrester's 2026 B2B marketing data adds the budget context. 83% of B2B marketing decision-makers expected marketing investments to rise over the next 12 months. Within that, Forrester's 2025 B2B Brand and Communications Survey found that the share planning to increase content and creative services investment dropped from 53% in 2024 to 44% in 2025. Together, those survey results show uneven budget intentions; they do not by themselves prove that the difference moved into PR or that any reallocation produced AI citations.
The Budget Reallocation Pattern
| Channel | 2023 Direction | 2026 Direction | Driver |
|---|---|---|---|
| Content and creative services | Increasing (53%) | Declining (44%) | AI tools replace volume production |
| Earned media / PR | Late-stage, optional | Earlier, required (92% increasing) | 92% of surveyed CEOs reported increasing PR investment because of AI search |
| Website and digital programs | Top priority (64%) | Declining (60%) | AI answers replace direct website visits |
| Paid advertising | Stable | Shifting to earlier funnel | AI discovery precedes intent signals |
| Third-party earned coverage | Nice to have | Infrastructure | Muck Rack observed sources brands neither own nor pay for at 82% of cited links in its measured sample, and 84% in its May 2026 edition |
Sources: Delight Labs State of PR 2026 (March 2026); Forrester B2B Brand and Communications Survey 2025; Muck Rack Generative Pulse December 2025.
The Citation Architecture Behind AI Recommendations
The GEO paper (Aggarwal et al., KDD 2024) published in SIGKDD 2024 (Aggarwal et al.) measured visibility changes after applying content features in a controlled benchmark, reporting gains for statistics and source citations under those experiment conditions. It supports testing specific, sourced, attributed claims. It does not establish a universal ranking rule, isolate publication authority as the cause, or predict the result of a PR placement in a live commercial query set.
Practically, teams can improve the evidence available for testing by making each placement specific, sourced, attributable, and current. Then they should measure whether the item is indexed, retrieved, cited at the host or exact-URL level, associated with the brand, and used in recommendation language. Neither the Ahrefs inventory nor the controlled GEO benchmark establishes that publication domain authority is the primary trust signal or that recency determines retrieval weight.
The Signal Genesys LLM citation study, which analyzed 179.5 million citation records across six LLM platforms, reported 88.4% domain citation coverage and different citation volumes by platform. Its cross-platform differences argue against one universal optimization strategy. Use a portfolio of relevant sources as the population to measure, not as a guarantee that sustained coverage will make a brand appear.
Stacker and Scrunch's earned-versus-owned pilot, reported on Machine Relations, observed a 4.25× citation-rate difference between earned and owned records across its compared topics. The measured unit is a bounded observed rate in that dataset. It is not evidence of causality or primacy, does not recommend one distribution strategy for every brand, and is not a forecast, guarantee, or business outcome. The useful bridge to Machine Relations is measurement: compare source role, placement, retrieval, citation, recommendation, referral, and commercial outcome separately.
The 30-to-1 Gap Between Recognition and Recommendation
A specific data point clarifies the urgency behind the budget shift. Research testing AI system behavior at the category level found that when AI systems were asked about a product by name, recognition rates reached 99.4%. When asked category-level discovery questions, such as "What are the best tools in this category this year?", the discovery rate on ChatGPT dropped to 3.32%. A 30-to-1 gap between being recognized when named and being recommended unprompted.
The study reported zero correlation between its GEO content optimization scores and discovery rates, while referring domains and third-party editorial presence were associated with visibility in its dataset. That is a cohort-level relationship, not proof that editorial coverage caused recommendation or that owned-page optimization has no effect. It gives teams variables to measure against their own query set.
This gap is where the budget increase becomes rational even to executives who have historically been skeptical of PR as a business investment. The operational question is whether the brand appears in important discovery queries and which sources support that appearance. Earned media, owned content, paid activity, entity data, and engine-specific retrieval can all be measured; the cited studies do not assign the result almost entirely to one channel.
AI Search Adoption Is Outpacing Budget Adjustments
The budget shift is not happening in isolation. The Bain 2025 AI search consumer study found that 80% of search users rely on AI summaries at least 40% of the time, and approximately 60% of searches end without the user progressing to a website. Forrester's State of Business Buying 2026 found that 30% of all buyers viewed generative AI tools as a meaningful interaction type during the final commit stage of their purchase, compared to just 17% in the previous year. The buyers are moving to AI search faster than most marketing stacks are adjusting.
When 60% of searches end with the AI summary, the brand cited in that summary wins the impression. No click required. No landing page required. No form fill required. The brand that is cited gets associated with the category in the buyer's mind. The brand that is not cited does not exist in that moment. Marketing budgets are chasing that reality: the budget shift toward PR is not optimism about earned media as a craft. It is a response to where brand discovery has moved.
What the Brandi AI Research Shows From Inside PR
In March 2026, Brandi AI published research on AI discovery restoring PR's strategic role. The core finding: "Public relations is the infrastructure of AI visibility. The outputs created by PR -- media coverage, expert commentary, and institutional validation -- are exactly the signals AI systems prioritize." The Brandi AI report also cited Gartner projections indicating PR spend could double by 2027. The researchers found this projection consistent with what practitioners inside PR firms are already observing in their own client portfolios.
This finding from inside the PR industry is notable because it documents how practitioners are framing their work. Muck Rack separately observed an earned-media-heavy source composition in its measured citation sample. One is an industry argument and the other is a citation inventory; neither demonstrates that PR is an engine-selection mechanism. Together they explain why teams are investing in measurement of whether independently published material is retrieved, cited, and connected to a brand.
Why Technical SEO Is Not the Answer
The budget data shows a specific pattern: investment is shifting toward earned media and away from technical content production, not away from marketing entirely. Understanding why requires understanding where technical SEO and GEO fall short for AI citation.
The Moz 2026 AI Mode analysis of 40,000 queries found that 88% of AI Mode citations were not in the organic top 10 in traditional search results. Only 12% of AI citations overlapped with traditional organic rankings. The optimization playbook for traditional search does not translate to AI citation. A brand can rank on page one of Google for its target keywords and still be absent from every AI answer in its category. The measured unit is source composition in the cited sample. Boundary to preserve: this does not establish a universal source-selection mechanism, citation causation, a guaranteed future citation, recommendation lift, pipeline, revenue, or any other business outcome.
Traditional search rankings and AI citations are different observable outputs. Technical SEO can affect discoverability, while independently published coverage can add attributable evidence about a brand. The current studies do not identify one primary factor in AI citation probability. A defensible program measures technical access, entity clarity, owned-source retrieval, third-party retrieval, host citation, exact-URL citation, and recommendation language instead of collapsing them into one trust score.
That is a PR problem, not an SEO problem. It requires earning placements in publications the AI engine already trusts, not optimizing pages the AI engine may or may not retrieve. Technical SEO and GEO are necessary layers. They are insufficient without the foundation.
The Compound Effect of Consistent Coverage
One of the less-discussed aspects of the budget shift is the portfolio dynamic. Each placement creates another independently published artifact that can be checked for indexing, retrieval, citation, and attribution over time. It may also enter future training or retrieval corpora, but that inclusion and its weight are generally not observable. Portfolio growth is therefore a measurement opportunity, not proof of compounding model authority.
University of Washington and Stanford research on training-data imprints in large language models, published in March 2025, examined memorization and recall patterns. Applying that research to a specific brand's media record remains an inference unless the relevant training inclusion, model behavior, and query outcomes are measured. Sustained coverage can build a durable public record, but this article does not claim an observable or irreversible model weight.
The Zhang et al. AI citation behavior study (arXiv, December 2025) found that 37% of AI-cited domains are absent from traditional search results entirely. The editorial layer and the SEO layer are pulling from different pools of sources. Companies that optimize only for traditional search are invisible in a significant portion of the AI citation ecosystem.
What the Budget Increase Actually Buys
The 92% of CEOs increasing PR spend are not all buying the same thing. Understanding what the investment actually produces at different levels helps clarify why the budgets are moving where they are.
At the baseline, a consistent earned media program creates independently published records that can be tested in category-relevant queries. This is the infrastructure layer: secure appropriate placements, confirm that the pages are accessible, and measure whether named engines retrieve, cite, or attribute them. A placement does not determine an unprompted recommendation.
At the next level, consistent coverage in relevant publications builds what earned authority means in the Machine Relations framework: a pattern of third-party evidence that can be evaluated across multiple engines and query types. The measured result is not the existence of coverage alone; it is whether that evidence is retrieved, cited accurately, and associated with the intended entity.
At the highest level, teams compare a publication portfolio with citation density and recommendation presence across a stable query panel. Repeated appearance can be a valuable observed outcome, but neither coverage volume nor publication authority guarantees that a brand becomes the default answer.
The PR Model That Matches the New Reality
The traditional PR retainer model charges whether or not a placement is secured. A company on a monthly retainer gets outreach activity, relationship management, and the probability of placements. The placement itself is not guaranteed. Neither is its quality or the authority of the publication it lands in.
The Delight Labs survey found average monthly PR retainer spend across its CEO respondents at approximately $18,000. That figure does not capture placement rate, placement quality, retrieval, citation, or commercial impact. Two companies at the same monthly spend can produce different media portfolios and different observed answer-set results, but the survey does not assign those differences to publication authority or consistency.
Performance-based PR, where payment occurs only when a placement is confirmed in a publication that meets the client's criteria, aligns payment with a verifiable media output. That output is a live placement, not an AI citation or business result. Retrieval, citation, recommendation, referral, and revenue should be reported as subsequent measured outcomes rather than promised consequences of the commercial model.
This is why Jaxon Parrott, who coined Machine Relations, built AuthorityTech around the outcome-based model: payment in escrow until the placement is live. Eight years of direct editorial relationships with 1,600+ publications. 10,000+ AI-cited articles secured for clients including 27 unicorn startups. The model works because the relationships are real and the placements are guaranteed, not promised.
Frequently Asked Questions
Why are companies increasing PR budgets specifically in 2026?
The clearest evidence is budget behavior: according to the Delight Labs State of PR 2026 survey, 92% of responding CEOs reported increasing PR investment because of AI-powered search. Muck Rack's Generative Pulse analysis separately found earned media at 82% of cited links in its measured response sample. That source-composition share does not show that earned coverage caused the budget change, causes citation, or makes PR sufficient for AI visibility.
How does earned media drive AI citations?
Earned media can add an independently published source about a brand, but the placement still has to be accessible, retrieved, and cited in the relevant answer set. Muck Rack measured more than one million cited links and reported an earned-media-heavy, mostly non-paid source composition; Fullintel/UConn measured a health-focused sample in which journalistic sources represented 47% of cited links. Those observations do not establish a universal selection mechanism, source primacy, or that a specific placement will cause a citation or recommendation.
What is Machine Relations and how does it connect to PR budget increases?
Machine Relations, coined by Jaxon Parrott in 2024, is the discipline of earning AI citations and recommendations for a brand by making that brand legible, retrievable, and credible inside AI-driven discovery, and measuring the resulting citations and recommendations. GEO and AEO are Layer 4 (distribution) of the five-layer Machine Relations Stack. The Stacker and Scrunch earned-versus-owned pilot, reported on Machine Relations, observed a bounded 4.25× citation-rate difference across its compared topics; it does not establish causality, primacy, a recommended budget allocation, a forecast, a guarantee, or a business outcome. The bridge from PR investment to Machine Relations is therefore a measurement chain: placement, retrieval, brand mention, host citation, exact-URL citation, recommendation language, referral, and commercial outcome.
Is PR spend replacing SEO and paid advertising?
No. The budget data shows changing investment intentions, not elimination of other channels or proof that all reduced content spend moved into PR. Technical access, owned information, paid distribution, entity data, and third-party coverage can each contribute different inputs. The Ahrefs inventory found that 65.3% of already-cited pages in its sample came from DR80+ domains, but that bounded correlation does not make domain rating a selection mechanism or prove that PR is the primary trust signal. The measured unit is the domain-rating distribution among pages already present in Ahrefs' citation inventory. Boundary to preserve: Ahrefs does not establish that Domain Rating causes citation, that DR80+ is required for citation, or that changing Domain Rating changes source selection.
How does earned media compound over time for AI visibility?
A growing publication record creates more independently published artifacts to test over time. Muck Rack observed that more than half of cited links in its December 2025 sample came from sources published in the preceding 12 months, with the highest observed citation rate within seven days. That recency distribution does not establish that publication timing caused retrieval or that each placement compounds model weight. Measure the same query panel over time and separate new placement, retrieval, citation, recommendation, referral, and business outcomes.
The Infrastructure Argument
The reason PR budgets are rising is the same reason any infrastructure budget rises when the infrastructure becomes load-bearing. For most of the past decade, earned media was valuable but optional. A brand could grow on SEO, paid acquisition, and content marketing without investing heavily in PR. The AI citation layer did not exist or was not yet material to business outcomes.
That has changed. When buyers use AI tools for vendor research, procurement teams use agents to shortlist suppliers, or a prospect asks ChatGPT about a category, companies need evidence beyond their own claims. Earned media can contribute that independent record, but it does not determine whether the brand appears. Teams still need to measure engine access, entity resolution, retrieval, citation, and recommendation across the queries that matter.
The Fullintel-UConn study observed third-party news and informational sources at 47% of cited links in its health-focused response sample, against 48% for corporate, university and health-network sites. The measured unit is source composition inside that sample. The result does not establish a fixed trust hierarchy, predict that the shares will persist, or show that editorial journalism causes a brand to be selected, cited, recommended, or purchased.
This is what Machine Relations contributes to the budget discussion: a measurement architecture connecting communications outputs to machine-observable results without assuming the mechanism. PR practitioners can report the placement; technical teams can test discovery and retrieval; analysts can record citation and recommendation language; commercial teams can measure referral and business outcomes. The chain is useful precisely because no source-composition study proves that one stage causes the next.
Next Step
For a brand that has not yet measured this infrastructure, the starting point is a dated baseline across the category queries that matter. Start your visibility audit to record current brand mentions, cited hosts, exact URLs, and recommendation language before changing the program.