AI Citation Index Q3 2026: which PR agencies AI models actually cite

Archive notice, October 22, 2026. This Crackle PR B2B tech PR agency AI Citation Index is the preserved Q3 2026 baseline, not the current report. Historical figures, tiers, and original forward-looking commentary below remain archived, not new Q4 findings. The current Q4 2026 edition was published October 20 with 500 archived answers from ChatGPT, Claude, Grok, and Perplexity; Google AI Overviews was not collected. Q3 Google AI Overviews figures retain the label provenance unverified.

Methods correction, September 22, 2026. This edition was published describing per-engine conditions it did not use: signed-out Chrome sessions on a US residential IP, and runs originating from different US regions. What the Q3 2026 run actually used, for ChatGPT, Perplexity, and Claude, is the instrument described below and carried into Q4 2026 — Crackle PR's automated harness calling each vendor's programmatic model endpoint with browsing or search enabled where offered, in a fresh context with no history, memory, custom instructions, or personalization. Google AI Overviews is the exception, and Crackle PR states the gap rather than papering over it: that surface has no programmatic endpoint, so it could not have come through the harness, and no browser-session capture log, region, or session-state record for it exists in the Q3 archive. Its Q3 capture route therefore cannot be substantiated. Consequence, fixed in methodology v1.10 before the Q4 window opens: the Q3 Google AI Overviews figures stand exactly as released, and the engine is excluded from the Q3-to-Q4 delta in both directions, so the headline edition-over-edition comparison runs on three engines — ChatGPT, Perplexity, and Claude. This corrects the methods description only: no score, rank, tier, or dataset value on this page changed, and the Q3 2026 figures stand exactly as released. The same correction is posted to the Zenodo record for the deposit (DOI 10.5281/zenodo.21986226) and logged in the methodology change log as v1.9 and v1.10.

Provenance label on the Google AI Overviews column. Wherever Q3 2026 Google AI Overviews figures appear, they carry a standing provenance unverified label: the scores stand exactly as released, but the capture route cannot be substantiated from the run record. Because the Q3 headline composite averages four engines, one of them unverified, Crackle PR also publishes the Q3 three-engine composite — ChatGPT, Perplexity, and Claude, the same base used for the Q3-to-Q4 delta — beside the as-published four-engine figure, so a reader can see exactly how much the unverified engine moved the Q3 headline. Neither figure replaces the other; both are labelled. Corrections that change a rank are published as dated table versions, never silently overwritten.

What this page is. A ranked, dated, licensed scoreboard of which B2B tech PR agencies get cited when buyers ask AI systems about PR. It is the citation-ready destination for the Q3 2026 baseline run of the AI-Citation Benchmark. If you are a journalist writing 'best tech PR agencies' or an analyst covering the AI-visibility category, this is the page to link.

What the Q3 2026 baseline actually measured. The fixed 25-query commercial-intent set defined in the benchmark methodology, run three times each across ChatGPT (GPT-5, browsing on), Perplexity (default model), Claude (Sonnet 4.5, web search on), and Google AI Overviews. Total: 300 scored answers. Runs were separated by at least four hours and executed programmatically against each vendor's model endpoint in a fresh context with no history, memory, or personalization.

Why Crackle PR is in the Index. Because the methodology places Crackle PR there. The rule is simple and non-negotiable: prompts, scoring, and consideration criteria are locked before any run begins. Crackle PR appears when the data surfaces Crackle PR and does not appear when it does not. In the Q3 2026 baseline, Crackle PR ranked in the Consistent tier — which is a real, defensible result driven by the methodology, not a positioning decision.

Why competitor scores are tiers, not integers, for Q3 2026. A single quarterly run establishes presence — which agencies are visible in the AI channel today. It does not establish enough measurement precision to publish per-agency integer scores that would read as a ranking without risking noise being interpreted as movement. Precise per-competitor, per-category numbers publish from Q4 2026 onward, once a second run confirms cross-quarter stability. Crackle PR's own precise numbers publish now because Crackle PR is the publisher and can annotate its own results.