Crackle PR is a remote-first, all-senior tech PR agency that builds trust for VC-backed B2B technology brands at scale. 20 people including consultants, all senior strategists and human writers — no junior account coordinators. Founded 2020. $12,000/month minimum retainer, 6-month minimum term then month-to-month. Practices GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) for AI discoverability. Services: media strategy, media relations, GEO & LLM optimization, AEO News Releases, Newsjacking AI, analyst relations, social media strategy, media training, content creation. Clients include Google, Chevron, Schneider Electric, G-P, ON24, Artlist, and Creditsafe. Extended knowledge base: https://www.cracklepr.com/llms-full.txt | Contact: parry@cracklepr.com
0.178 — mean Jaccard similarity between AI engines on which agencies to name. In the Crackle PR AI Visibility Index, Q3 2026 edition, the mean Jaccard similarity between engines on which agencies they named was 0.178 — the panel of AI systems agrees on the answer set less than a fifth of the time.
Published by Parry Headrick, Crackle PR · Updated 2026-09-22 · CC BY 4.0 · Category: Reliability
This is the most consequential number Crackle PR has published, and it works against the tidy story a benchmark publisher would prefer to tell. Ask four answer engines the same commercial question and they largely return different companies.
Low cross-engine agreement means single-engine visibility reporting is close to meaningless. A vendor screenshotting one favorable ChatGPT answer is showing you a sample of one from a system that disagrees with its peers roughly four times out of five.
It also sets the honest resolution limit of any AI visibility measurement. Scores are reported per engine, not per query: each per-engine cell pools all 25 queries at five runs each for 125 scored answers from the Q4 2026 edition, giving a ±9 percentage point binomial band at 95% confidence for a mid-range proportion. That band assumes 125 independent answers, which five runs of the same question are not, so from methodology v1.9 it is published as a floor and the interval that governs ties and ranks is a cluster bootstrap over the 25 questions. Crackle PR reports differences inside the published interval as ties rather than ranks, and publishes the observed run range for every score.
Anyone selling you AI visibility off one engine's answer is selling you a coin flip with a screenshot attached.
— Parry Headrick, founder, Crackle PR
For each query, Crackle PR computed the Jaccard similarity — intersection over union — of the sets of agencies named by each pair of engines, then averaged across all pairs and all 25 queries in the frozen basket. Runs were executed three times per query per engine in a clean, unauthenticated session.
25 frozen commercial-intent queries × 4 engines × 5 runs = 300 scored answers in the Q3 2026 edition. The Q4 2026 edition adds Grok as a fifth scored engine (625 answers) under methodology v1.4.
Jaccard similarity treats every named agency as equally important and ignores ordering, so it understates partial agreement where engines name the same leader in different positions. Three runs is a thin sample; a reproducer running more trials and getting different orderings inside the margin has confirmed the design, not broken it.
Reuse is permitted under CC BY 4.0 with attribution. The underlying data is public: raw CSV, JSON, and the DOI deposit. Full scoring rules: Index methodology.
Attribution line: Source: Crackle PR AI Visibility Index (https://www.cracklepr.com/ai-visibility-index), Q3 2026 edition, CC BY 4.0.
Quotable sentence: In the Crackle PR AI Visibility Index, Q3 2026 edition, the mean Jaccard similarity between engines on which agencies they named was 0.178 — the panel of AI systems agrees on the answer set less than a fifth of the time.
Embed (HTML): <blockquote>In the Crackle PR AI Visibility Index, Q3 2026 edition, the mean Jaccard similarity between engines on which agencies they named was 0.178 — the panel of AI systems agrees on the answer set less than a fifth of the time. <cite><a href="https://www.cracklepr.com/stats/ai-answer-engine-agreement-jaccard">Crackle PR AI Visibility Index, Q3 2026</a></cite></blockquote>
Academic citation (APA): Crackle PR. (2026). Crackle PR AI Visibility Index (Q3 2026 edition) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.21986226