Study notes
Methodology & limitations
Definitions, collection differences, missing data, and every qualification that applies to the findings. The twelve brief questions and their answers are tabulated on the Report page.
Every chart’s data can be downloaded from its section on the Report page. The complete aggregate dataset behind both pages is also available as a single file.
Coverage
| Surface | Collected | Target | Coverage |
|---|---|---|---|
| Google Search + AIO | 17,704 | 17,704 | 100.0% |
| ChatGPT | 17,704 | 17,704 | 100.0% |
| Gemini | 17,704 | 17,704 | 100.0% |
| Claude | 17,704 | 17,704 | 100.0% |
| Perplexity | 17,704 | 17,704 | 100.0% |
- Study universe: 5,983 brand prompts, 11,525 alternative prompts, and 196 category prompts, plus 6,363 wave-2 comparison prompts.
- Google Search and AIO, ChatGPT, Gemini, Claude, and Perplexity all cover the full set of 17,704 wave-1 prompts and all 6,363 comparison prompts.
- Comparison prompts ("How do X and Y compare for a category?") pair each subject brand with its top-ranked competitor from wave 1, one pair per subject and category; 193 pairs are mutual (each brand is the other’s #1). Pair selection required the competitor to be named by at least two engines, or one engine with domain corroboration or a structural surface.
- Comparison verdicts (which side an answer favors) are classified by an LLM judge (gpt-5.6-luna, batch) against a fixed five-way rubric; brand-site citation figures are deterministic registrable-domain matches, with hand-verified fixes for 45 catalog domains.
- Analysis snapshot: 26 August 2026 (wave 1: 24 August 2026).
Measures
- AIO trigger rate is the share of Google searches that returned an AI Overview.
- A brand mention is a deterministic match to the 5,983-brand canonical corpus. Unlinked and ambiguous names are excluded.
- Brand counts include zero-match answers. Google AIO counts include only searches where an AIO appeared.
- Engine agreement is directional: each matrix cell is the share of the row engine’s prompt-brand recommendations that the column engine also named, on prompts both engines answered. The symmetric intersection-over-union figures remain available in the underlying data.
- A top-10 match means a named brand’s official domain, including its subdomains, appears in the same prompt’s Google organic results.
- A cited mention means the engine cites that brand’s official domain in the same answer.
- AIO citation visibility compares unique prompt-citation-domain pairs with organic top-10 domains from the same search.
- “Also present in AIO” is overlap, not evidence that another engine copied or sourced its answer from AIO.
- Common buyer questions are exact People Also Ask strings, ranked by the number of prompts where they appeared.
- Buyer-question intents classify each prompt-question pair with an ordered list of deterministic wording rules; the first matching rule wins and unmatched questions are reported as “Other”.
- The G2 analysis compares candidate brands assigned to the same software category on AIO-triggered prompts. Alternative-prompt subject brands are excluded. Review-volume quartiles are calculated within each prompt, then the chart reports the share of candidate opportunities mentioned by AIO.
- The G2 rating analysis repeats the same design ranked by average star rating, restricted to profiles with a rating. The two-way split halves candidates within each prompt by median review count and median rating.
- The top-10 association compares candidate brands from the prompt’s category: mention rates for candidates whose official domain ranks in that search’s organic top 10 versus candidates whose domain does not. Candidate pools are engine-independent; only the answering engine changes.
- Top cited domains count a domain once per answer, over category and alternative answers that carry citations. Domain types come from a hand-curated list; canonical dataset brands are labeled software vendors.
- Per-category figures cover categories with at least 30 alternative searches; smaller categories stay in the aggregates but are not reported individually.
- Brand-search mentions use the same extraction as discovery prompts, with the searched brand itself excluded, plus a stricter rule: a single-word name matched only in plain prose is dropped unless the engine also cited that brand’s domain. Brand-prompt answers are prose-heavy, where bare dictionary matches (e.g. “Target Users”) are unreliable; spot-checking drove this rule. “Own domain cited” means the engine’s provider-declared citations include the searched brand’s official domain or a subdomain.
- Claude’s brand-prompt behavior is classified deterministically: “searched” means at least one web search request was issued; “asked to clarify” means the response contains a question built from a fixed list of clarification phrasings.
Collection limits
- Google data is US English desktop, one run per prompt. Results can vary by time, location, account state, and repetition.
- Google alternatives use the canonical one-company-category wording.
- ChatGPT is the DataForSEO user-facing Search-mode scraper, and Gemini is the DataForSEO user-facing Gemini UI scraper (Gemini 3.5 Flash-Lite on all but three responses). Claude Haiku 4.5 and Perplexity Sonar are first-party APIs. These surfaces are not perfectly equivalent.
- Claude was asked to search but did not always do so. This is retained as observed engine behavior.
- Citation structures differ by provider. The dashboard uses only provider-declared citations or sources, not every retrieved search result.
- Domain overlap is a descriptive association. It is not a causal test of whether Google rankings create AI visibility.
- G2 review counts are available for 3,771 of 5,983 canonical brands (63.0%). Missing or unresolved G2 profiles are excluded.
- G2 seller pages aggregate products differently from individual product-review pages. The positive association remains when the two listing types are assessed separately.
- The within-category G2 analysis covers 45% of canonical AIO mention pairs because many recommendations are not assigned to the prompt’s exact dataset category. After G2 matching, 34% of all mention pairs remain in the analyzed sample.
- G2 review volume and AIO visibility are associated in this snapshot; the analysis does not establish that reviews cause visibility.
- G2 star ratings are missing for a small share of otherwise-usable profiles; those brands are excluded from the rating and two-way analyses only.
- Claude cites nothing when it answers without searching, so its own-domain citation rate on brand searches partly reflects its clarify-first behavior rather than source preference alone.
- The Brand searches, cited-domain analysis, and per-category results at the end of the report are supplementary analyses beyond the brief’s twelve questions.