Methodology

Engines and exact versions

A study that does not name the model version it ran against is not reproducible, because the answer changes when the model does.

Engine	Model version
Claude	rankxa-warehouse
Gemini	rankxa-warehouse
ChatGPT	rankxa-warehouse

Run window, UTC: 2026-07-01T23:15:00Z to 2026-07-31T17:15:00Z.

Method

The design goal was to reproduce, as closely as an automated study can, what an ordinary buyer sees: a plain question, put to an assistant, with no special prompting to surface a particular vendor. The study ran 22,875 distinct prompts through 3 engines, each represented by one production model configuration, 1 version tracked for ChatGPT, 1 for Claude, 1 for Gemini, over a run window from 2026-07-01 to 2026-07-31 (UTC timestamps). That produced 1,548,342 logged rows, of which 1,548,342 returned a usable answer, a completion rate the study treats as the denominator for every rate reported here.

Prompts were drawn across four intent types, informational (someone learning about a category), navigational (someone looking for a specific known company), commercial-investigation (someone comparing options before a decision), and transactional (someone ready to act). Sorting by intent matters because a company's absence from an informational answer means something different than its absence from a transactional one: the first is a content gap, the second is closer to a lost sale.

Every usable answer was parsed for company mentions and matched against a cohort of 153,251 companies drawn from 153,251 candidate records. A mention was logged with its position in the answer, first, second, and so on, because position is doing real work: a company named sixth in a list of eight is technically visible and commercially almost invisible.

Three corrections, referred to here as mitigations, were applied before publication (3 in total) to remove known sources of noise, such as an engine echoing the prompt's own wording back as a company name. The study does not claim these mitigations catch everything, and a reader auditing the dataset should check the mitigation log against their own category before trusting a null result for any single company.

Limitations we volunteer

Written by us, before anyone else found them.

Terms used in this study

Appearance rate

The share of answers, out of all usable answers for a given slice such as an engine or a prompt type, in which at least one company from the cohort was named.

Cohort

The fixed set of companies, 153,251 of them in this study, that every engine answer is checked against to determine whether a mention occurred.

Mention position

The rank at which a company appears within a single generated answer, first, second, third, and so on; a lower number means the company appeared earlier in the list.

Pair agreement

A measure of how many of the three engines named the same company for the same prompt, calculated only over company-prompt pairs where at least one engine made a mention.

Prompt type

A classification of the buyer intent behind a prompt: informational (learning about a category), navigational (looking for a known company), commercial-investigation (comparing options), or transactional (ready to act).

Run window

The start and end timestamps, in UTC, over which this month's prompts were issued and answers collected; here, 2026-07-01T23:15:00Z to 2026-07-31T17:15:00Z.

Status: ok

A row-level flag meaning the engine returned a usable, parseable answer for that prompt; rows not marked ok are excluded from every rate reported in this study.

Visible somewhere, not everywhere

The condition of being named by at least one of the three engines during the window, without necessarily being named by all three or named consistently.

Zero visibility

The condition of never being named by any engine, on any prompt, during the entire run window; 0 of 153,251 companies were in this state this month.

References

Sources this study reads against. Every link was fetched and confirmed reachable at publication.

1. GEO: Generative Engine Optimization arXiv (Princeton, Georgia Tech; KDD 2024), 2024 Foundational paper formalizing GEO and introducing GEO-bench, showing content-level interventions can shift visibility in generative engine responses.
2. E-GEO: A Testbed for Generative Engine Optimization in E-Commerce arXiv, 2025 Proposes a domain-specific benchmark for studying GEO effects in e-commerce, relevant context for our multi-category commercial cohort.
3. Deep-Research Agents Can Be Poisoned via User-Generated Content arXiv, 2026 Literature review section summarizes GEO findings that authoritative language, citations, and statistics affect source selection, and that engines show distinct citation preferences from traditional search.
4. Generative engine optimization Wikipedia, 2026 General reference for GEO terminology and definition used to orient readers unfamiliar with the field.
5. GEO: Generative Engine Optimization ACM Digital Library, Proceedings of KDD 2024, 2024 Conference of record for the GEO paper; cited for the peer-reviewed venue and abstract.
6. ChatGPT Search OpenAI Help Center, 2026 Vendor documentation describing how ChatGPT presents inline citations, used to explain mechanism behind engine-reported sources.
7. The Most-Cited Domains in AI: A 3-Month Study Semrush, 2026 Independent large-sample study of citation share across LLMs over thirteen weeks, used as a comparison point for cross-engine citation behavior.
8. AI Platform Citation Patterns: How ChatGPT, Google AI Overviews, and Perplexity Source Information Profound (tryprofound.com), 2026 Cross-engine analysis of citation patterns by top-level domain, relevant comparison for engine-level divergence.
9. How ChatGPT sources the web Profound (tryprofound.com), 2026 Large-sample (~730,000 conversation) study of ChatGPT citation behavior, used to compare scale and method against our own answer set.
10. How B2B Buyers Use AI to Choose Vendors Grey Matter, 2026 Summarizes Forrester's 2026 Buyer Insights survey finding generative AI is now buyers' most-cited research source, used to motivate why engine visibility matters commercially.
11. Web search tool Claude Platform Docs, Anthropic, 2026 Vendor documentation of Claude's web search and citation mechanism, cited to explain how Claude selects and attributes sources.
12. Anthropic Introduces Web Search Functionality for Claude Models InfoQ, 2025 Independent technical summary of the Claude web search API launch, corroborating the citation-generation mechanism.
13. Half of B2B Software Buyers Now Start Their Research with AI Chatbots: G2 Demand Gen Report, 2026 Reports G2's 'Answer Economy' finding on ChatGPT's dominant share of B2B software research starts, used as buyer-behavior context.
14. 100 Most Cited Domains in ChatGPT Ahrefs, 2026 Monthly-updated domain-level citation tracker for ChatGPT, cited for its finding on the concentration of citations among a handful of large domains.
15. Grounding overview Google Cloud Documentation, Gemini Enterprise Agent Platform, 2026 Defines grounding as tethering model output to verifiable sources, used to explain the general concept applied across all three engines.
16. 72% of B2B software buyers now use ChatGPT to evaluate vendors, and most brands aren't showing up MarketScale, 2026 Cites Forrester/Crackle PR Q2 2026 AI Citation Benchmark on buyer use of ChatGPT for vendor evaluation and the share of brands absent from citations, used to frame the stakes of non-visibility.
17. Grounding with Google Search Google AI for Developers, 2026 Vendor documentation of Gemini's grounding mechanism, cited to explain how Gemini attaches citations to generated answers.

Data

The complete row-level dataset is published open and ungated under CC BY 4.0. Every number in this study can be recomputed from it.