Query Type Does Not Predict Engine Agreement
Whether a prompt is transactional, informational, or navigational, engines agree at roughly the same rate. Appearance rates range only from 36.9 to 38.6 percent.
38.3 percent
The finding
One might expect engines to agree more on constrained questions (like navigational queries asking about a specific company) than on open-ended questions (like transactional queries asking who to hire). The data does not support this.
| Prompt type | Appearances | Appearance rate |
|---|---|---|
| Transactional | 47628 | 38.3 percent |
| Commercial investigation | 8457 | 38.6 percent |
| Informational | 4469 | 38.7 percent |
| Navigational | 4687 | 36.9 percent |
The range is less than two percentage points. Navigational queries, which seem most likely to have "right answers," actually show the lowest appearance rate.
Why navigational queries do not drive higher agreement
Navigational queries in this context still involve company recommendations. A question like "How do I find a contractor in Austin?" is navigational in intent but still requires the engine to decide which contractors to name. The query type affects the framing, not the fundamental task of selecting companies to recommend.
If the sample included queries like "What is IBM's phone number?", where there is exactly one right answer, agreement would presumably be higher. But those queries do not appear in a dataset designed to study company recommendations.
What this suggests about engine disagreement
The consistency across prompt types suggests that disagreement is driven by factors that apply regardless of query framing:
- Training data differences persist across query types. An engine that lacks information about a company lacks it whether the query is transactional or informational.
- Retrieval system behavior is query-type-agnostic. The sources an engine consults may not change based on whether the user is investigating or ready to buy.
- The recommendation task is fundamentally similar. All these queries ask "who should I consider?", just with different framings.
Implications for optimization
This finding argues against query-type-specific optimization strategies. A company hoping to appear in transactional queries does not face fundamentally different engine dynamics than one hoping to appear in informational queries.
The 88.4 percent sole-mention rate applies broadly, not just to certain query categories.
Study context
Appearance rate is calculated as the number of appearances (company mentioned for a question on an engine) divided by the number of possible slots (questions times companies that could have been mentioned). The near-identical rates across prompt types suggest structural similarity in how engines approach recommendation across query types.
Other findings
How often does only one AI engine recommend a company when other engines do not?
The Sole-Mention Rate: Why Most AI Recommendations Are Unique to One Engine
When an AI engine recommends a company for a buyer question, that recommendation is unique to that engine 88.4 percent of the time. The other two engines simply do not mention the same company.
How often do all three AI engines agree on recommending the same company?
Only 3.2 Percent of Recommendations Are Unanimous
Unanimous agreement across ChatGPT, Claude, and Gemini is rare. When all three engines were asked the same buyer question, they all named the same company just 3.2 percent of the time.
Which AI engine recommends the most companies?
ChatGPT Recommends More Companies Than Claude or Gemini
ChatGPT mentioned 42.7 percent of the companies studied, compared to 39.9 percent for Claude and 33.6 percent for Gemini. The engines have different thresholds for recommendation.
Want to know how AI answers describe you?
We run the same measurement on your category. Fifteen minutes with founder Omar Jenblat, your own numbers, no deck.