Study · MABIDO · measurement method included
How often AI answers name or link a business: Citation Pulse, November 2026
MABIDO asks AI assistants a fixed set of questions and records whether the answer names the business or links to its website. This study reports the platform-wide results measured up to 16 November 2026, with the method and the limits stated first. The live figures are published at mabido.com/api/citation-pulse-proof and change as new measurements are added.
Method
- Two kinds of question. Named: the question contains the business name, so the answer shows whether the assistant recognises it. Discovery: a buyer's question with no name, so the answer shows whether the business is found on its own. The discovery figure is the one that matters for a buyer.
- Engines. Gemini with Google Search grounding, OpenAI with web search (gpt-4.1-mini), and Brave search. A banned model (gpt-5.5) is excluded from this study: its 16 rows are not counted.
- Definitions. Brand cited: the business name appears in the answer. Domain cited: the business website appears among the sources. Cited (any): either of the two.
- Failed measurements (a timeout, an API error, no grounding) are counted in the total and shown separately. They are never read as "not cited".
Results
| Question type | Attempts | Measured | Brand cited | Domain cited | Cited (any), of measured |
|---|---|---|---|---|---|
| Named (the business is in the question) | 26 | 26 | 24 (92%) | 24 (92%) | 100% |
| Discovery (no name in the question) | 73 | 61 | 39 (64%) | 32 (52%) | 64% |
Totals: 99 attempts, 87 measured. Measured per question type = attempts without a failure. Percentages rounded.
By engine
The same measurement, split by assistant. Each engine answers its own question type, so the rows are not all comparable with each other.
| Engine and question | Attempts | Measured | Brand cited | Domain cited | Cited (any), of measured |
|---|---|---|---|---|---|
| Gemini with Google Search, named | 13 | 13 | 12 (92%) | 12 (92%) | 100% |
| OpenAI web search (gpt-4.1-mini), named | 13 | 13 | 12 (92%) | 12 (92%) | 100% |
| Gemini with Google Search, discovery | 50 | 38 | 37 (97%) | 30 (79%) | 97% |
| OpenAI web search (gpt-4.1-mini), discovery | 16 | 16 | 1 (6%) | 1 (6%) | 6% |
| Brave search, discovery | 7 | 7 | 1 (14%) | 1 (14%) | 14% |
Sum of the discovery rows: 39 answers cite the brand (Gemini 37, OpenAI 1, Brave 1), 32 cite the domain (Gemini 30, OpenAI 1, Brave 1). The two engines differ widely on discovery questions, and the 16 OpenAI and 7 Brave answers are too few to state a rate.
What this shows, and what it does not
- When the business is named, assistants recognise it in almost every case. This is recognition, not discovery.
- Without a name, about two measured answers in three name the business or cite its domain in this sample. The domain is cited less often than the name (52% against 64%): an assistant can mention a business without linking to its website.
- The sample is small. 61 measured discovery answers and 26 named answers do not support a precise percentage. The numbers can move a lot with a few more measurements. We publish them because they are what was measured, not because they are final.
- The businesses measured are MABIDO clients, measured by MABIDO. An independent panel of businesses would be needed to generalise; that study is not done.
- No result is guaranteed for any business, and nothing here is a promise of citation or ranking.
Next measurement
The next round adds recommendation questions ("which company should I choose for ...") and a measure of whether the answer recommends the business rather than only naming it, so that mention, citation, and recommendation are reported separately.
Version: study of November 2026, data to 16 November 2026. Method changes are recorded in the API field schema_version and in dataset_revision.