Guide
How to Measure AI Search Without Pretending It Has Fixed Rankings
Traditional rank tracking can assign a position to a query and compare it over time. AI answers are more variable. The same commercial question can produce different brands, ordering and sources depending on the platform, mode, wording and collection time.
How this guide is meant to be used
- Method
Baseline: Measure what is true today before changing anything.
A more honest measurement system tracks observations: Was the brand mentioned? Was it recommended? Was it cited? Which source supported the answer? How often does that happen across a repeatable prompt set?
1. Build the Commercial Prompt Set First
Start with questions real customers might ask: recommendations, comparisons, service-specific needs, product research and market questions. Do not change the prompt until the brand finally appears.
2. Separate Mention, Recommendation and Citation
A brand mention is not automatically a recommendation. A recommendation is not automatically a citation. A citation to the company website can occur even when the company is not recommended. Record those states separately.
3. Record the Context
Platform
The answer system being tested.
Mode or model
Search/browsing mode or model version where visible.
Prompt
The exact question, not a paraphrase.
Date
The collection date because outputs change.
Brand state
Absent, mentioned, compared or recommended.
Citation
The exact source where available.
Source ownership
Owned site, directory, review platform, publisher, association or another third party.
4. Repeat Important Prompts
Repeated observations help distinguish a consistent pattern from a one-time answer. The objective is recommendation and citation frequency, not an invented average position.
5. Build a Source-Gap Map
When competitors appear and the client does not, inspect the sources supporting the answer. The missing evidence may be first-party content, reviews, a relevant directory, an association, a publisher, comparison content or stronger third-party corroboration.
6. Report the Denominator
If a brand appeared in 18 of 60 tracked prompts, say 18 of 60. That is more useful than calling the company highly visible without showing the sample.
What Not to Do
• Do not present one favorable screenshot as broad AI visibility.
• Do not change prompts mid-test and compare the outputs as though the test stayed the same.
• Do not call a mention a citation when no source was cited.
• Do not guarantee an AI recommendation.
• Do not attribute revenue to an AI answer without supporting attribution evidence.
Can AI search be tracked like Google rankings?
Not reliably as one fixed position. Recommendation frequency, brand presence and citation/source tracking are more defensible measures.
How many prompts should a business track?
Enough to represent the commercial decisions customers actually make. A narrow local service business may need dozens; a broad ecommerce catalog may require substantially more.
Build an AI Search Baseline
Explore AI Search Optimization or schedule a consultation to define the commercial prompt set before measuring improvement.
Start with clarity
Schedule a Consultation
A conversation with a strategist — not a sales script. We review your position and tell you the honest next move.