The Catalyst AI Visibility Index

What: Multiplier applied to Google search volume to estimate total AI query volume for a category.

Why: Adoption only. Google discloses roughly 14B searches a day. OpenAI discloses roughly 2.5B ChatGPT prompts a day, about half of them information-seeking, per the OpenAI/NBER usage study. That puts search-like AI volume near 9% of Google's economy-wide. We hold 0.5 for B2B software research, where buyers reach for an assistant more often than the general population does. G2 buyer research finds most B2B buyers now use AI somewhere in a purchase evaluation, and Gartner projects a 25% decline in traditional search volume driven by assistants. Going from the 9% economy-wide figure to 0.5 for B2B is a judgment call on our part and we would rather label it than bury it. We previously published 2.0, built by multiplying adoption against session consolidation and a long-tail correction. Both are gone. Consolidation says each ask carries more buying intent, so multiplying the ask count by it and then applying an unchanged per-ask citation rate counts the same intent twice, and the long-tail factor had no source. Read every dollar figure against the 9% economy-wide floor if you want the harder bound.

What: Estimated percentage of AI responses that result in a click-through to a vendor website.

Why: AI models frequently answer questions without linking out. When they do cite a vendor, a fraction of readers click through. We lowered this from 2.5% to 1.0% in August 2026. At 2.5% the modeled pipeline came out high enough that the return multiples read as implausible to buyers, and a number nobody believes is worth less than a smaller one they will argue with. 1.0% is a deliberately low floor. It will rise as models adopt more link-forward formats.

What: Estimated percentage of citation visits that become a qualified pipeline opportunity.

Why: Calibrated to B2B SaaS inbound conversion benchmarks. A buyer arriving via an AI citation is already in active research mode, higher intent than average organic traffic, but not every visit converts to a qualified opportunity.

What: Annual contract value used to translate opportunity volume into dollar pipeline.

Why: Derived in priority order: (1) pricing page, if the company publishes per-seat or per-agent prices, we multiply by a typical enterprise seat count; (2) Apollo enrichment, employee count and funding stage provide a reliable proxy when pricing is hidden; (3) category benchmarks, industry median ACV for the classified product category as a last resort.

Context-free measurement

Voice share is measured via standardized, context-free API calls, no conversation history, no account state, fresh sessions per query. This is the most reproducible and comparable baseline, but it differs from what individual users see. A buyer who has been researching your category for 20 minutes will get different results than a cold query. Our score measures baseline citation behavior, not personalized results.

Non-determinism

LLM outputs are stochastic. The same query can return different results across runs. We run each prompt twice per AI platform and take the union of both runs, so a company mentioned in either run receives credit for that question-and-platform pair. Each run uses a different buyer persona (decision-maker vs. practitioner). The union reduces variance and does not eliminate it. We measured this on repeat scans of an identical 10-prompt set and observed voice share moving by roughly 15 points between runs, so read the score as a reading with a range around it, and read a small gap between two companies as a tie.

Query coverage

We ask 10 buyer questions per scan, written for your company's category and buyer. Each one asks AI to recommend, compare or choose, the way a buyer asks before shortlisting, and each is tested once before it is kept. Real buyers ask thousands of variations, so the set represents the highest-intent questions rather than all of them, and 10 questions carry roughly ±15 points of spread between runs. A company keeps its questions and its company profile on every rescan, so two scans compare like with like.

Google sample

The Google part checks up to 20 searches, desktop results, in one market, the country your domain points to or the US for a .com. It reads the top 10 organic results only, so a page at position 11 counts as not ranked, and it does not see personalised or logged-in results. Paid results, AI Overviews and map packs are not counted here. AI Overviews is already one of the five AI answers in the AI part.

Coefficient estimates

The LLM shift coefficient (0.5), citation CTR (1.0%), and opportunity rate (5%) are calibrated estimates rather than audited figures. B2B AI citation behavior is an emerging measurement area with few published benchmarks, and no study we can find measures AI query volume against Google volume for B2B categories specifically. Anyone citing one is usually quoting a growth rate as though it were a ratio. The shift coefficient carries the widest uncertainty, so we publish the arithmetic behind it and a harder floor of 9% of Google volume, the economy-wide adoption figure. Voice share is measured directly from live AI answers. Only the volume and the dollars are modeled. We update these coefficients as more data becomes available and will publish a changelog at this URL.

Point-in-time snapshot

LLM training data, retrieval augmentation, and citation behavior all change over time. A score today reflects today's AI landscape. Companies that improve their entity authority, publish more original research, or earn more citations will see scores improve on future scans. Scores are not permanent rankings.

Pipeline figures are estimates

The monthly pipeline figures are directional estimates intended to size the opportunity, not audited revenue projections. They represent the estimated value of AI-driven buyer attention in your category, not a guarantee of recoverable pipeline. Treat them as a signal for prioritization, not a forecast.