Answer Engine Optimisation: The Math Behind AI Visibility
Three formulas (Citation Rate, Share of Voice, and Citation Share) are all Indian brands need to measure AI visibility rigorously. Most programmes track only one.
Most AEO programmes are under-querying, not under-optimising. Expanding a query set changes the numbers without any content intervention.
A fintech founder in Bengaluru is looking at a monthly AEO report. The citation rate has moved from 18% to 22% across a four-week period. The question that follows is exactly the one Magnent hears most often when Indian brands start tracking their AEO math AI visibility: is that a real improvement, or is it just month-to-month noise? The answer is not instinct. It is arithmetic.
AEO math AI visibility reduces to three formulas: Citation Rate, Share of Voice, and Citation Share. Each measures a different dimension of how AI engines engage with a brand. A citation rate moving from 18% to 22% across a 20-query set is within the noise range; the same move across a 50-query set, sustained over two consecutive four-week windows, is a genuine signal. Magnent applies all three formulas across client citation tracking programmes to separate random fluctuation from structural change.
What Are the Three Formulas Behind AEO Math and AI Visibility?
Three formulas cover most of what matters in AEO measurement. Most Indian brands track only one and draw conclusions that do not hold up over time.
Citation Rate is the foundational formula:
Citation Rate = (number of queries where the brand was cited / total queries tested) x 100
A brand cited in 14 out of 21 tested queries has a Citation Rate of 66.7%. This metric answers the question: across the queries the brand cares about, how often does AI actually surface it as a source?
Share of Voice shifts the frame from absolute presence to relative presence:
Share of Voice = (answers mentioning your brand / total answers in the query set) x 100
A query set of 20 questions returns 20 AI responses. If a brand appears in the text of 8 of those responses (whether cited as a source or just named), its Share of Voice is 40%. A competitor appearing in 12 of the same responses holds 60%. Share of Voice counts mentions: the brand's name appearing anywhere in the response, even without a direct source citation.
Citation Share is the most precise of the three and the most commonly omitted:
Citation Share = (citations of your domain / all citations across the query set) x 100
This captures the proportion of the total citation pool belonging to the brand. A brand appearing in 5 out of 20 responses but earning only 3 citations out of 40 total citations across those responses has a Citation Share of 7.5%. The distinction matters: a brand can hold a high Share of Voice while holding a low Citation Share. The gap between those two numbers is often where the content trust deficit sits.
Is Share of Voice Actually Different From Citation Rate?
These two metrics are frequently conflated, which produces misleading data. The table below separates all three by what they measure, when to use each, and what a high score actually signals for an Indian brand running an AEO programme.
| Metric | What It Measures | Best Used For | What a High Score Signals |
|---|---|---|---|
| Citation Rate | % of queries where the brand was cited as a source | Tracking actual sourcing by AI engines | AI engines trust the brand's content |
| Share of Voice | % of AI responses mentioning the brand | Competitive positioning and brand awareness in AI answers | AI engines know the brand exists in this category |
| Citation Share | % of all citations in the query set belonging to the brand | Understanding source authority relative to competitors | AI engines prefer the brand's content over alternatives |
A brand with high Share of Voice but low Citation Share is being named but not trusted as a primary source. Optimising for this gap (moving from mention to citation) is the most direct intervention target in AEO programmes.
What Does a Realistic AI Visibility Score Look Like for Indian Brands?
There is no industry-standard benchmark as of mid-2026. Any tool that publishes a single "AI visibility score" is building on methodology the industry has not formally agreed on. Directional benchmarks from aggregate AEO data provide a working reference:
- B2B SaaS brands in competitive categories: 10-15% citation rate as a baseline
- Category market leaders: 30%+ citation rate sustained across multiple measurement windows
- Fintech brands facing stricter AI source scrutiny: meaningful performance typically begins at 8-12%, given that regulated categories face higher sourcing standards from AI engines
For Indian brands specifically, benchmarks shift depending on how well-represented the category is in English-language AI training data. A category with limited third-party coverage will show lower absolute citation rates across all players. Reaching 15% in a thin-coverage category may represent stronger relative performance than 25% in a saturated one. The benchmark is always category-relative, not universal.
How Many Queries Do You Actually Need Before the Numbers Mean Anything?
Citation data varies 40-60% month-to-month for the same query on the same brand, without any brand-side change (Discovered Labs, 2026). The variation reflects how AI engines weight sources differently across model updates and retrieval cycles. The practical implication is that sample size and measurement window matter more than the absolute percentages in any single reporting cycle.
Minimum working thresholds for reliable AEO measurement:
- Query set size: At least 20 queries per model tested; 50 or more for B2B categories with competitive depth
- Measurement window: Rolling 4-week cycles, with at least two consecutive windows before drawing directional conclusions
- Model scope: Measure ChatGPT, Gemini, and Perplexity separately. The three models behave differently enough that aggregating their data into a single number misrepresents performance on each
The non-obvious finding here: most AEO programmes are under-querying, not under-optimising. A 21-query test set cannot reliably distinguish a genuine 4-percentage-point improvement from random variation at the model level. Expanding the query set is a measurement improvement, not a strategy change. It will change the numbers without any content intervention. Any brand seeing unexplained swings in its monthly AEO data should check the query set size before revising the content plan.
The AI visibility audit Magnent runs across Indian brand categories treats query set construction as a distinct first step because the measurement architecture determines what the data is actually capable of showing.
Can You Tell Whether a Content Change Caused the Numbers to Move?
When a brand implements a schema update, publishes a structured FAQ, or secures a new third-party mention, the question is whether the intervention drove the subsequent number change or whether the metric would have moved regardless. A before/after snapshot comparison rarely answers this cleanly given baseline volatility.
A more defensible approach is a controlled comparison: a 21-day observation window post-intervention, compared against the same window in the prior cycle, across a stable and consistent query set. The intervention is the variable; everything else is held constant.
Client citation tracking data from Magnent's AEO programmes shows one consistent pattern across Indian brands: interventions targeting Perplexity produce the most volatile post-intervention numbers, while schema and entity-level work targeting Gemini tends to show the most stable uplift over rolling 8-week windows. ChatGPT falls between the two, responsive to content authority signals but with a lag of several weeks before changes manifest in citation data.
The AEO and AI visibility trends tracked across Indian brand programmes from January to April 2026 provides the baseline pattern data that informs how long to wait before evaluating a given intervention.
The broader context: AI-assisted research is now embedded in over 60% of enterprise purchasing workflows globally, according to McKinsey's 2025 State of AI research (McKinsey, 2025). For Indian B2B brands, the measurement rigour applied to AEO visibility directly affects whether AI citation translates into actual commercial reach.
Frequently Asked Questions
What is the formula for calculating AEO citation rate?
Citation Rate = (number of queries where the brand was cited / total queries tested) x 100. A citation means the AI engine linked to or directly sourced the brand's content, not merely named it in the response text. Tracking citation rate separately from brand mentions is important because the two measure different things: sourcing trust versus awareness.
Is 15% citation rate good for an Indian B2B brand?
Directional benchmarks from aggregate AEO data suggest 10-15% is a working baseline for competitive B2B SaaS categories, with market leaders exceeding 30%. For fintech and regulated sectors, 8-12% can represent a strong position given stricter AI sourcing standards. The category's overall citation density matters as much as the absolute number. The relevant question is whether the brand is performing above or below its category average, not whether it has crossed a universal threshold.
How is AEO math AI visibility different from traditional SEO metrics?
SEO metrics measure position in ranked lists: rank 1, rank 5, page 2. AEO math AI visibility measures how often AI engines choose to surface a brand as a source or named entity across a defined set of intent queries. There is no rank in AI search, only citation frequency, share of voice, and citation share. A brand can hold a strong position in traditional search for a keyword while maintaining a near-zero citation rate in AI search for the same intent.
How long before an AEO intervention shows up in citation data?
Schema and entity-level interventions targeting Gemini typically manifest in citation data within 4-8 weeks. Perplexity can reflect changes faster but is more volatile, meaning early post-intervention lifts may not hold. ChatGPT shows the longest lag (6-10 weeks is common), but the resulting uplift tends to be more durable once it appears. These patterns are drawn from client tracking data across Magnent's AEO programmes and represent directional guidance, not guarantees.
Do ChatGPT, Gemini, and Perplexity need separate measurement programmes?
For brands that depend on AI citation for lead generation or brand credibility, yes. The three models use different source retrieval mechanisms, weight content types differently, and respond to interventions on different timelines. Aggregating them into a single score produces a number that misrepresents performance on each individual platform. Magnent's prompt engineering and brand monitoring framework is built around model-specific query sets for this reason: one combined score obscures more than it reveals.