01Study protocol
Searlixa AI Visibility Index
The AI Visibility Index will measure how consistently brands are cited across AI platforms, by industry and by market. This page is the protocol, published before any data is collected, so the method cannot be adjusted after the result is known.
No results yet There are no findings on this page and there will not be until the first collection cycle completes. Everything below is the plan. When results publish, this protocol stays live alongside them so anyone can check what we said we would do against what we did.
The question
How consistently is a brand cited across AI platforms for the same buyer intent, and how much does that consistency vary by industry and by market?
This matters because published 2026 research found the two most similar platforms sharing only 24.1% of cited pages. If that inconsistency is systematic rather than random, it changes how visibility budgets should be allocated.
Design
| Parameter | Planned | Why fixed in advance |
|---|---|---|
| Industries | The verticals we work in | Prevents selecting sectors after seeing which look best |
| Markets | US, UK, UAE, Australia | Prompt sets are built in-market, not translated |
| Platforms | ChatGPT, AI Overviews, AI Mode, Gemini, Perplexity | Reported separately; never blended into one score |
| Prompts | Fixed set per industry per market, frozen before collection | A moving prompt set makes trend data meaningless |
| Repetitions | Multiple runs per prompt per cycle | Non-determinism must appear as a distribution |
| Cadence | Quarterly | Slower than platform noise, fast enough to show real movement |
| Coding | Documented rules, applied by a named analyst, checked by a named reviewer | A different analyst should reach the same numbers |
What we will publish
- Citation consistency across platforms, by industry and by market
- Which source types the platforms draw on most in each sector
- Change between cycles, with platform-level annotation where we detect a shift
- The full method, the sample sizes and the limitations
What would falsify the premise
Stating this in advance is the part that makes pre-registration worth doing. If cross-platform overlap turns out to be high and stable across sectors, then per-platform measurement is largely unnecessary and a blended score would be defensible. That result would undercut a service we sell. We will publish it if we find it.
Known limitations, stated before we start
- A fixed prompt set describes those prompts, not every buyer conversation.
- A large share of AI answers carry no attribution; unattributed mentions are real and unmeasurable from outside.
- Platform retrieval changes without notice and can move results with no underlying change in the brands measured.
- Searlixa has a commercial interest in finding that per-platform measurement matters. That is why the falsification condition above is published in advance.
Governance
Author and reviewer will be named on publication, per the publication standard. Amendments to this protocol get a version bump and a dated note explaining what changed and why.
→Next step