"Cited by AI" and "cited in the answer your buyer sees" are two different claims

The claim, and the disclosures that belong next to it

The primary source is the 22 July 2026 announcement and the AI Citations Report it draws on (sample: 8,000 GlobeNewswire press releases, engines: ChatGPT and Claude). The framework is SOAR — Structure, Originality, Authority, Recency. Three disclosures belong in the same breath as the 99.3%, and none of them make it false: GlobeNewswire is Notified's own distribution wire, so this is a vendor measuring the product it sells; the full report is gated, so the citation-detection method cannot be independently checked; and "cited" is not defined in the public materials — a citation surfaced in any query context counts the same as one surfaced in a buying question. We are an AI-visibility vendor publishing a competing number, which is the fourth disclosure, and it is ours.

What we counted, and how

Our corpus is 173 completed audits — 1,024 answers — covering 131 distinct businesses (some re-audited over the period) across med spas, personal-injury law firms, dentists and home-services businesses in US metros, 2026-06-12 to 2026-07-11, with synthetic QA rows removed by exact-name filter. The citation-bearing half is the 511 ChatGPT web-search answers; we extract every linked domain and count unique domain-per-answer citations — 2,815 of them across 1,107 distinct domains, averaging 5.5 sources per answer. Against a wire list of 21 press-release distribution domains (GlobeNewswire, Business Wire, PR Newswire, PRWeb, EIN Presswire, ACCESSWIRE, and the smaller networks), the hit count is 6: two PR Newswire, two PRWeb, one PRLog, one IssueWire. That is 0.21% of citations and 1.2% of answers. GlobeNewswire and Business Wire: zero.

Both numbers are probably right

This is not a debunk, and the two measurements are not in contradiction — they answer different questions on different query classes. Notified's corpus is corporate news: earnings, product launches, executive moves, the kind of release an S&P 500 comms team distributes, evaluated against whether an engine ever cites it. Ours is local consumer buying intent: "best med spa in Dallas," "personal injury lawyer near me" — the questions where a business either gets recommended or does not. A release can be ingested, indexed and cited within eight hours in a corporate-research query and be absent from every answer your customers actually generate. Our result says nothing about B2B, finance or brand-research queries; we did not test them.

What to do with this if you are buying or selling AI visibility

1. 1Make anyone quoting a citation rate name the query set. "99.3% cited" is a statement about supply — the release got picked up. The number that decides revenue is demand-side: of the questions your buyers ask, in how many does your name come back? Those two can differ by two orders of magnitude, as they do here.
2. 2Wire distribution is an authority-and-recency play, not a local-recommendation play. SOAR's Originality and Recency signals match what we see — third-party, freshly-written coverage is what the engines read. But for a local business, the documents doing that job in our corpus are review aggregators and vertical directories, not the wires. Budget accordingly, per vertical, per query class.
3. 3Measure before and after, on your own questions. A release costs money and a citation report is not a measurement of your visibility. Record which businesses the engines name for your buyers' questions in the week before distribution and the week after; if nothing moves, the channel is not wrong, it is just not the one your buyers touch.

The demand-side version of this test is cheap to run for one business — put the buyer's question to several engines and record which names and which sources come back. That is what our free AI visibility check does, and the full source map for the corpus above is published at where AI looks before it recommends a local business.