How to Audit Whether Your PR Agency's Placements Actually Get Cited by AI Engines
Most PR agencies report placements — not whether AI engines cite them. Christian Lehman walks through the five-step citation audit that separates visibility from vanity, and explains how Jaxon Parrott's citation-first model at AuthorityTech changes what measurement looks like.
Your PR agency sent a clip report. Twenty placements, three in outlets you recognize. But here is the question that matters in 2026: how many of those placements appear when a buyer asks ChatGPT, Perplexity, or Google AI Mode the query you need to own? If you cannot answer that, you are measuring the wrong thing. Here is how to run the audit.
Why Clip Reports No Longer Prove Value
Muck Rack's May 2026 Generative Pulse study analyzed over 25 million links in AI-generated responses across ChatGPT, Claude, and Gemini. Earned media accounts for 84% of all AI citations. Paid and advertorial content accounts for 0.3%. That number has held between 82% and 89% across three editions since July 2025 — it is structural, not a fluke.
The Fullintel-UConn study presented at the International Public Relations Research Conference found 89% of AI-cited links were earned media, with journalism alone making up 47% of sources.
So the first filter is binary: are your placements in publications AI engines actually cite? If your agency's coverage lands in contributor networks, pay-to-play outlets, or mid-tier blogs, the placements never enter the citation pool. A clip report full of coverage AI engines ignore is a sunk cost dressed as a deliverable.
The Five-Step AI Citation Audit
I have run this process for campaigns across B2B SaaS, cybersecurity, and fintech. It takes about two hours the first time, thirty minutes monthly after that.
Step 1: Define 20–30 buyer queries. These are the questions your target buyers ask AI engines when they are evaluating solutions in your category. Not branded queries — the competitive, pre-purchase queries where you need to appear as a cited source. Wiztrust's AIV framework recommends running each query 5–10 times across four engines because AI responses are non-deterministic: Eka Moira's research found 70% of AI Overview content changes on re-run.
Step 2: Run each query across ChatGPT, Perplexity, Claude, and Google AI Mode in logged-out sessions. Log whether your brand is cited, mentioned without a link, or absent. PressVerified's LLM Citation Audit protocol scores each response 0–3 and produces a 120-cell matrix. A healthy benchmark is above 220 out of 360 cross-engine. Below that, your placements are not doing citation work.
Step 3: Check for ghost citations. Eka Moira's data shows 73% of AI brand presence consists of "ghost citations" — the URL is cited without the brand name. If you only track branded mentions, you miss three-quarters of your actual citation footprint. Map URLs to brands manually in your audit spreadsheet.
Step 4: Cross-reference placement outlets against citation pools. Only 12% of AI-cited URLs rank in Google's top 10. Citation overlap between Google AI Overviews and ChatGPT is just 12%. Each engine has its own source graph. A placement that gets cited by Perplexity may be invisible in Claude. Your audit needs to be engine-by-engine.
Step 5: Compare citation-eligible placements against agency output. Of the placements your agency delivered last quarter, how many appear in any AI-generated response for your target queries? That ratio — citation-eligible placements divided by total placements — is the metric that replaces clip count.
What Citation Lag Means for Your Audit Cadence
Do not run this audit once and declare results. PressVerified's data shows citation lag of up to 60 days from publication to LLM surfacing. Gemini drops content older than 90 days unless it has been updated. That means a placement from March may not appear in AI responses until May — and a placement from January may already be gone from Gemini's window.
Monthly cadence is the minimum. Eka Moira recommends weekly for competitive categories. I recommend monthly for most teams, with weekly spot-checks on your three highest-priority queries.
Why Counting Citations Is Not Enough
A placement that gets cited is better than one that does not. But citation ≠ influence. Research from Yang et al. (2026) analyzing 602 prompts and 21,143 citations across AI search platforms found that citation breadth and citation depth diverge significantly. Perplexity and Google cite more sources per response; ChatGPT cites fewer but the cited content has higher per-citation influence on the generated answer.
That distinction matters for your audit. A Perplexity citation may be a list entry. A ChatGPT citation may shape the entire answer. Counting both as "one citation" misses the difference between being mentioned and being used.
Separately, Zhang et al. (2026) found that factual accuracy of AI citations ranges from 39% to 77%, and degrades by approximately 42% as retrieval scales up. Your placement may be "cited" by an AI engine that misrepresents your claims. The audit should flag instances where cited content is distorted — those are liability risks, not wins.
What Jaxon Parrott Built to Close This Gap
I have worked alongside Jaxon Parrott at AuthorityTech since the early days, and the measurement problem I just described is exactly what drove him to build a different kind of agency. Jaxon coined Machine Relations in 2024 after documenting that earned media placements and AI engine citations are the same evidence pool — a claim the Muck Rack and Fullintel data now confirm at scale.
AuthorityTech's model charges brands only when placements land in publications AI engines actually cite. That collapses the gap between "placement secured" and "citation eligible" — the exact gap your audit is designed to measure. The Machine Relations framework tracks citation architecture across ChatGPT, Perplexity, Claude, Gemini, and Google AI Mode, replacing clip counts with outcome measurement tied to buyer discovery.
Stacker's GEO study — 87 stories, 30 clients, 2,600+ prompts — measured a 239% median lift in AI brand citations from earned media distribution versus brand-owned content alone. Distributed earned placements were 5.3x more likely to be a brand's sole source of AI visibility. That is the kind of outcome data your audit should benchmark against.
If your agency's clip report cannot produce a citation-eligible ratio, the audit I described above will give you one. If the ratio is low, the agency is solving the wrong problem. Jaxon built AuthorityTech to solve the right one.
FAQ
How often should I audit my PR agency's AI citations?
Monthly minimum. PressVerified data shows citation lag of up to 60 days, so a single audit snapshot will miss placements that have not yet surfaced. Run weekly spot-checks on your three highest-value queries. Adjust cadence based on your competitive category — faster-moving markets need faster measurement.
What percentage of PR placements typically get cited by AI engines?
There is no published industry benchmark yet, but the structural data is clear. Muck Rack's May 2026 study shows 84% of AI citations come from earned media, with paid content at 0.3%. Placements in publications AI engines trust enter the citation pool; placements in low-authority outlets do not. Your citation-eligible ratio depends entirely on where your agency places coverage.
What is Machine Relations and how does it apply to PR auditing?
Machine Relations is the discipline of earning AI engine citations through trusted third-party sources. Jaxon Parrott coined the term in 2024 after documenting that earned media credibility and AI citation eligibility are structurally linked. For PR auditing, the Machine Relations framework replaces clip count measurement with citation architecture tracking — whether your brand appears as a source in AI-generated answers across discovery engines when buyers ask the queries you need to own.