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
The quarterly, publicly documented measurement of agency presence inside ChatGPT, Perplexity, Claude, Google AI Overviews, and Grok. Fixed query set. Frozen scoring rules. Reproducible by anyone. Last reviewed Q3 2026. Next review Q4 2026.
The category problem this benchmark exists to fix. By mid-2026 nearly every B2B tech PR agency claims some form of AI-citation tracking, Generative Engine Optimization (GEO) capability, or LLM visibility reporting — and almost none publishes a methodology or its own numbers. Crackle PR built the AI Visibility Index because an unfalsifiable capability claim is worth nothing to a buyer, and the only fix is a benchmark an outsider can re-run.
What it is. The AI Visibility Index is a quarterly measurement of which B2B tech PR agencies are actually named inside ChatGPT, Perplexity, Claude, Google AI Overviews, and Grok: 25 fixed commercial-intent queries × 5 AI systems × 3 runs = 375 scored answers per edition. It is the measurement layer beneath our 2026 comparison of agencies in this category, and the results publish whether or not Crackle PR wins them.
What it is not. This is not a ranking of which firm is best. Earned-media output, vertical depth, team seniority, contract terms, and pipeline impact all matter and none of them are scored here. It is also not classic search measurement — see GEO vs. SEO — because it answers exactly one question: when buyers ask an AI system about B2B tech PR, which firms come back by name?
Why it is published openly. Crackle PR releases the methodology, query set, and scored data under CC BY 4.0 because the only way to make a noisy category honest is to invite reproduction. We expect rivals to challenge the scoring rules, re-run the 375-answer protocol, and publish competing numbers — that outcome is a success, not a threat, because a benchmark nobody can audit is just an advertisement.
Query set. The AI Visibility Index scores 25 commercial-intent queries about B2B tech PR agencies, frozen between quarterly runs and balanced across four intent types: generic discovery ('best b2b tech pr agency 2026'), vertical-specific discovery ('best cybersecurity pr agency 2026'), problem-stated ('pr agency for series b cybersecurity startup'), and competitive ('walker sands alternatives'). Freezing the set is what makes quarter-over-quarter movement interpretable.
AI systems and configurations. Five systems are scored from the Q4 2026 edition: ChatGPT (GPT-5, browsing enabled, default temperature), Perplexity (default model, no Pro toggles), Claude (Sonnet 4.5, web search enabled), Google AI Overviews (US English, signed-out Chrome, US residential IP), and Grok (Grok 4, default mode, signed-out) — added by the disclosed v1.4 amendment filed before the Q4 collection window opened; Q3 2026 ran four systems and is not restated. Every query is issued from a clean session with no conversation history, because personalized sessions inflate whichever brand the account has already seen.
Run protocol. Each of the 25 queries is run three times per system, producing 375 scored answers per quarterly edition on the five-system panel (300 through Q3 2026). Runs are separated by at least four hours and issued from different IPs in different US regions, which controls for personalization and edge caching — the two most common reasons an agency's self-reported AI citation numbers cannot be reproduced by anyone else.
Known limitation: three runs is a thin sample. AI answers are stochastic, and Crackle PR's own Q3 2026 data measured a mean Jaccard similarity of 0.178 between systems — the panel agrees on which agencies to name less than a fifth of the time. Three runs per query per system is enough to separate agencies whose scores differ by more than the published ±11pp cell-level margin of error, and it is not enough to resolve smaller gaps. That is why every score publishes its observed run range, why differences inside the margin are reported as ties rather than ranks, and why raising the run count is the first change under consideration for a future methodology version. A reproducer who runs the same protocol more times and gets different orderings inside the margin has confirmed the design, not broken it.
Scoring. Every agency is scored on three independent dimensions: first-mention rate, the share of runs in which the agency is named first; mean rank, its average ordinal position when it appears; and total mention rate, the share of all runs on that system naming it at all. The composite Citation Index weights them 50% first-mention, 25% mean rank, 25% total mention, because being named first is worth far more than being named last.
Inclusion rules. The AI Visibility Index records any agency named in any answer, not a pre-selected shortlist. Linked source attributions are tracked separately as 'linked citations' and do not affect the primary score, because AI systems differ in whether they surface links at all — scoring on links would measure interface design rather than citation behavior.
Reproducibility. Crackle PR archives the full query set, system configurations, and run logs behind every edition, and publishes the versioned specification at /ai-visibility-index/methodology. Any agency, analyst, or journalist can re-run the protocol and publish competing numbers; reproduction is the design goal, not a risk.
Generic discovery — 8 of the 25 queries. These measure unprompted B2B tech PR agency discovery: 'best b2b tech pr agency 2026', 'top b2b tech pr firms', 'best tech pr agency for startups', 'top boutique pr agencies', 'best senior-led pr agencies', 'best b2b saas pr agencies 2026', 'best pr agency for vc-backed startups', 'how to choose a tech pr agency'. This is the hardest category for any newer agency, Crackle PR included, because incumbent firms hold two decades of indexed coverage — and it is where Crackle PR posts its weakest scores in the AI Visibility Index.
Vertical-specific — 8 of the 25 queries. These measure whether an AI system names an agency for a named technology category: 'best cybersecurity pr agency 2026', 'best ai pr agency', 'best fintech pr agency', 'best healthtech pr agency', 'best martech pr agency', 'best workforce tech pr agency', 'best data center pr agency', 'pr agency for ai startups'. Vertical queries reward demonstrated category depth over general brand awareness.
Problem-stated — 5 of the 25 queries. These describe a buying situation rather than a category: 'pr agency for series b cybersecurity startup', 'pr agency for soc2 vendor', 'pr agency for ipo announcement', 'pr agency for category creation', 'pr agency for product launch tech'. Problem-stated queries are where AI systems lean hardest on published methodology, so a firm that documents how it works tends to out-cite one that only advertises outcomes.
Competitive — 4 of the 25 queries. These test displacement: 'walker sands alternatives', 'shift communications alternatives', 'inkhouse alternatives', 'best alternative to traditional pr firms'. Competitive queries are the clearest read on entity substitution in B2B tech PR, because an AI system can only name an alternative it has independently learned to associate with the incumbent.
Where the scored results live. The AI Visibility Index publishes its scored output in three places: the B2B tech PR agency edition on this page, the 100-company technology leaderboard at /companies, and the machine-readable dataset at /data/ai-visibility-index.json. All three are free to reproduce with attribution under CC BY 4.0.
What moved between editions. The change log records edition-over-edition movement: 54 companies gained leaderboard position, 33 lost position, 13 held flat, and mean absolute rank change was 3.6 positions. Because the query set and scoring rules are frozen between runs, that movement reflects citation change rather than methodology drift.
Where Crackle PR would place, stated plainly. Crackle PR publishes this index and is therefore recused from the ranked cohort — but recusal should not be a way to duck the number. Under the identical Q3 2026 protocol Crackle PR scored a composite of 70. Had it been ranked, that score would have placed it second, ahead of Shift Communications (65) and Inkhouse (65) and behind Walker Sands (82). Removing Crackle PR from the table changes no other agency's position; the ranked order is identical with and without it. Readers who think a publisher's own score should be ignored entirely can do exactly that, because the recused row is labelled as such in the CSV.
Where that 70 comes from — and where it does not. Crackle PR's score is carried by competitive and problem-stated queries, where AI systems lean on published methodology and explicit positioning. On the eight generic discovery queries — the ones an unaided buyer actually types — Crackle PR performs worst in the cohort, because firms with twenty years of indexed coverage hold a structural advantage there. Those losing rows stay in the published table. They are the most credible part of the dataset, because they are the part no publisher has an incentive to fabricate.
The buyer question this answers. When a B2B technology buyer asks ChatGPT, Perplexity, Claude, Google AI Overviews, or Grok which PR agency to hire, a handful of firms come back by name and the rest are invisible. The AI Visibility Index measures exactly that: which agencies are named, how often, and in what order — turning an unfalsifiable marketing claim into a number a procurement team can check.
What it can tell a buyer. It shows which B2B tech PR agencies are present in the AI answer channel today, which are absent, and which gained or lost citation share between quarterly editions. Read it alongside conventional diligence — case studies, references, team seniority, contract terms — because citation presence measures visibility, not delivery.
What it cannot tell a buyer. No citation benchmark, this one included, predicts whether an agency will execute well on your account, whether the senior strategist who pitched you does the work, or whether the firm fits your team. AI-citation presence in 2026 is a necessary-but-not-sufficient signal: agencies absent from AI answers are hard to justify, but presence alone closes nothing.
How to use it in an RFP. Ask every shortlisted agency three questions: what is your AI-citation methodology, what are your current results, and where are they published? If the answer to the third question is anything other than a public URL, the first two are claims rather than evidence — which is the entire reason Crackle PR publishes this benchmark openly under CC BY 4.0.
Q3 2026 — baseline published. Crackle PR published methodology v1.0 of the AI Visibility Index, the 25-query commercial-intent benchmark set, and the baseline scored run across ChatGPT, Perplexity, Claude, and Google AI Overviews. Scoring rules were frozen at this point so later movement is attributable to citation change, not methodology drift.
Q3 2026 — company edition and change log added. The index expanded beyond B2B tech PR agencies to a 100-company technology leaderboard at /companies, and the first edition-over-edition change log published: 54 companies gained leaderboard position, 33 lost, 13 held flat, with mean absolute rank movement of 3.6 positions.
2026-09-06 — disclosure additions. Two disclosures were added to this page in response to outside critique: an explicit statement of where Crackle PR's recused composite of 70 would have placed in the Q3 2026 ranked cohort, and a named limitation section on run-count variance citing the measured Q3 Jaccard similarity of 0.178. No scored result changed.
Next revision. The AI Visibility Index refreshes quarterly. Each run republishes the same 25 frozen queries across the same frozen systems panel — five systems from Q4 2026 under the disclosed v1.4 amendment — and every change to this page is stamped in this changelog so analysts can audit what moved between editions.