The ibs Fulcro AI Search Visibility Methodology: A Canonical Reference

The 6 AI Engines We Test

Every ibs Fulcro AI Search Visibility engagement runs real buyer prompts against 6 live, named AI engines: ChatGPT, Perplexity, Google AI Overview, Google AI Mode, Google Gemini, and Microsoft Copilot. We don't test a single engine and extrapolate; different engines cite different sources, weight different signals, and produce genuinely different visibility outcomes for the same brand, which is itself real, useful diagnostic information.

The prompt set itself is built through a real, calibrated process: real keyword data from SEMrush is expanded across different buyer-intent stages inside Velivo Radar's Playground (our preferred AEO/GEO tracking, analytics, and recommendation platform), weighted toward the intent stages most likely to actually convert, not just the highest-volume search terms.

The Single Most Important Decision: Brand-Anchored vs. Unprompted Prompts

A prompt that already names a brand (“JK Tyre vs Apollo for Toyota Taisor”) will mention that brand in the response most of the time, almost by construction, that's not a real measure of AI visibility, it's closer to a check that the AI engine can read. A genuinely generic prompt with no brand name at all (“best budget tyre for Skoda Kylaq under 5000”) is the honest test: the brand has to earn its way into the answer with no help from the prompt itself.

Blending these two prompt types into one number is a real, common mistake, and one we've made and corrected in our own published work. An early version of the JK Tyre audit reported a single blended mention rate of 40.2% across all scans. Splitting the two categories apart revealed the real, honest, unprompted mention rate, materially lower, and the number we now report as the primary, honest competitive metric in every engagement.

How We Define Coverage vs. Owned-Citation Share

Two genuinely different metrics answer two different questions, and conflating them hides real problems. Coverage is the share of scanned prompts in which a brand appears in the AI engine's answer at all, whether cited from the brand's own domain, a retailer, a review site, or a competitor's comparison page. Owned-citation share is the share of all citations behind those answers that came specifically from the brand's own domain.

A brand can have real, strong coverage while having almost no owned-citation share, meaning it's being talked about, but almost never from a source it actually controls. Across five real, independent AI visibility scans we've published, brands with under 1% owned-domain citation share averaged single-digit AI recommendation coverage, while brands above 10% owned-citation share averaged 30%+, a real, measured relationship between the two, not an assumption.

How We Score Position: Rank-Decay, Not Average Position When Mentioned

Most AI visibility reports, including some of our own earlier published work, lead with “average position when mentioned,” a number calculated only across the prompts where a brand appeared, silently dropping every prompt where the brand was absent from the calculation entirely. That produces a real, specific distortion: a brand mentioned twice, both times at rank 1, reports a “perfect” average position of 1.0, indistinguishable from a brand that dominates every single prompt in the batch.

We now score every prompt in the batch, absences included, using a real, specific rank-decay method.

Real, Honest Limitations of This Methodology

This approach has genuine, specific limitations worth stating directly, not glossing over. AI engine outputs are not static: the same prompt run on different days, or even at different times of day, can return different citations and different rankings, since these models and their retrieval systems are continuously updated. A single scan is a real, honest snapshot, not a permanent fact.

Prompt sampling itself involves real judgment calls. The specific mix of buyer-intent stages, and the weighting toward the ones most likely to convert, is a deliberate choice that shapes which real buyer questions get asked, and a different, equally reasonable sampling strategy could produce somewhat different numbers. We calibrate this against real SEMrush keyword data specifically to reduce that risk, not eliminate it.

Sentiment scoring, when included, is inherently more subjective than presence or position, since it requires judging whether an AI-generated answer's tone toward a brand is genuinely positive, neutral, or negative, a real, harder call than counting a citation.

How Often We Recommend Re-Measuring

Given that AI engine outputs genuinely shift over time, a single scan should be treated as a real, dated baseline, not an evergreen fact. For an active AEO/GEO engagement, we recommend re-scanning on a real, regular cadence, typically quarterly, closely enough to catch genuine movement, not so frequently that normal day-to-day variance gets mistaken for a real trend.

Where This Methodology Is Already Applied

This isn't a theoretical framework. It's the exact method behind the specific, published numbers across our real, named engagements and research: the JK Tyre AEO/GEO audit's honest, corrected 17.8% unprompted mention rate, the Owned-Citation Gap study's five real brand scans, and the rank-decay scoring used throughout our AEO/GEO practice.