Share of Voice in AI

When AI answers,
is your brand the answer?

Your customers stopped reading ten links. They ask ChatGPT, Gemini, and Claude — and get a shortlist of two or three names. SOVA measures whether you're on it, why not, and what moved after you acted.

Arabic and English · confidence intervals on every number · formula published
ChatGPTGeminiClaudesearch on + offMSA · dialect · EN
live measurement · run 4 of 10 · gpt · search on
أبغى أفضل مستشفى أطفال بالرياض، وش تنصحون؟
Mention
0%
Top-3
0%
Recommended
0%
every number traces to answers like this one · extractor v4 · golden-set gated
Why now

The click disappeared

Classic search ended at a click — your page got the visit and made its case. In an AI answer there is no visit. The page's only reader is a machine that quotes selectively, and whoever it quotes wins the customer.

~70%
of the recommendation comes from what others publish

Models trust consensus over self-description. Your own site is necessary, never sufficient.

3.3×
measured citation gap between two comparable sites

Pages that state facts get quoted. Pages that state philosophy get skipped.

weeks
to change what AI retrieves and says about you

Versus years to change what it remembers from training. All action lives in retrieval.

What we measure

Three numbers, three different problems

A single "AI visibility score" hides the diagnosis. SOVA separates presence from prominence from endorsement — because each failure has a different owner and a different fix.

MENTION
Do you appear in the answer at all?
Low = presence gap

Decomposes into three causes — your pages don't rank, the cited pages omit you, or the answer came from Maps — each with its own fix. Never fix blind.

TOP-3
Are you in the first three options?
Low = prominence gap

Present but buried. Models order by consensus strength: how many sources name you, how early, how authoritatively.

RECOMMENDED
Are you the pick, or just a listing?
Low = reason gap

The model has no differentiating claim to endorse. Publish the reason; get third parties repeating it until it reads as fact.

Why the split matters: one hospital we measured appears in 67% of pediatric answers — but makes the shortlist in 13%. AI knows it exists; AI doesn't send it patients. A single score calls that "good visibility." Three numbers call it what it is: a reason gap, with a known fix.
The platform

A measurement operation, not a monthly PDF

Behind the dashboard is a running instrument: your category's full question bank, fired across the major engines every month, in both languages, with every answer stored and every number traceable. The dashboard is just where it surfaces.

10,700 AI answers measured monthly 214 questions × 10 runs × search on/off 3 engines, versions tripwired 2 languages + Saudi dialect Every competitor measured free, from the same answers
sova.sa · your dashboard · JUL session
MENTION
41%
95% CI ±3 · n=6,420
▲ +6 vs JUN · significant
TOP-3
24%
95% CI ±3
+2 · within noise, not reported
RECOMMENDED
11%
95% CI ±2
the gap that costs customers
0 30 60 ▲ placement shipped 14 May ⚠ engine update · annotated
per-model trend · shaded = 95% CI · shipped work becomes dated markers, so movement has a cause
BRANDMENTIONTOP-3REC.
د. سليمان الحبيب
79%64%
السعودي الألماني
48%31%
Your brand
24%11%
التخصصي
33%30%
every competitor measured from the same answers, at zero extra cost · drill any row to the specialty level
⚑ SERIOUS

Pediatrics: known, not chosen

REASON GAP

67% mentioned, 13% top-3. Models know the department exists; no retrieved source says dedicated children's hospital — the claim appears in zero framing phrases across 6,420 answers. The asset is real; the web doesn't know it.

Playbook → children's-hospital page, Arabic first · seed 3 comparison-layer placements repeating the differentiator · re-measure next session.
drafted by AI from the metrics · verified by a human editor · every claim traces to raw answers
PROMPTTAGS
وين أحسن مستشفى ولادة بالرياض؟OB-GYNbest-ofdialectCORE✓ approved
pediatric ER wait time RiyadhPEDlocal/urgentENCORE✓ approved
دلة أو الحبيب للولادة؟ أيهم أفضلOB-GYNcomparisondialectCANARY✓ approved
كم تكلفة الولادة القيصرية ٢٠٢٦OB-GYNcostMSACORE✓ approved
ولدي حرارته مرتفعة وين أوديه الحين؟PEDsymptom→wheredialectCORE✓ approved
your customers' real questions — imported, de-duplicated to distinct intents, tagged, human-approved, versioned · new questions arrive from what the engines themselves search
Session JUL-2026 running · 6,412 / 10,700 answers · est. $112 total
Claude · search on + off2,690 / 4,280
GPT · search on + off2,825 / 4,280
Gemini · grounded897 / 2,140
PED-021 · claude · captured fan-out: "pediatric cardiologist Riyadh" · "أفضل دكتور قلب أطفال"
KHB-002 · gpt · old name resolved: مستشفى السلام → دلة الخبر
OBG-047 · claude · brands seen: دلة, الحبيب · framing: recommended #1
every answer stored permanently with the queries the engine actually searched — those queries become next month's content targets
لأفضل تجربة ولادة في الرياض: ١. مستشفى دلة — تقييمات ممتازة لقسم النساء والولادة. ٢. مستشفى الحبيب فرع الريان. ٣. مستشفى المواساة.
{ brand: "دلة", position: 1, framing: "recommended", valence: "positive", cited: ["dallah-hospital.com", "sehatuk.sa"] }
what the extractor read from that answer — one structured row per brand per run, 10,700 times a month · this is what your metrics are made of
Monthly readoutWhat moved, what didn't, and what's noise — each gap named and routed to its owner with the fix attached.
Competitor leagueEvery brand in your category, three metrics each, drillable to the specialty level, from the same measurement.
Evidence on demandClick any number and read the actual AI answers behind it, Arabic and English, run by run.
AttributionYour shipped work becomes dated markers on the trend. When a number moves, you can point at why.
A curated question bankBuilt from how your customers actually ask — call-center language, dialect and formal — versioned, human-approved, growing from what the engines themselves search.
False-claims registerWhen an engine states something untrue about you, it's logged, traced to its source, fixed, and re-verified next session.
The score

The SOVA Index

One number with a real-world meaning: of all the recommendation-weight AI handed out in your category this month, the share your brand captured. Every appearance earns credit for how it appeared — recommended #1 is worth 6× a buried listing — weighted by what the question is worth.

Index = Σ weighted credits ÷ Σ possible credits × 100 · blend 70/30 search-on/off · formula v1.0, frozen & published
Market leader
58
High-volume peer
29
Authority brand
27
Your brand
21
New entrant
11
Sample category league · listed 4th in 90% of answers loses to recommended #1 in 30% — the Index sees what raw counting hides.
Methodology

Built like an instrument, not a dashboard

AI at the edges, math in the middle, a human at the end. Every number in a SOVA readout traces back to a raw model answer you can open.

Measure

Your question bank — real customer language, Arabic first — runs across the engines, ten times each, search on and off.

Diagnose

Extraction to structured rows, then statistics decide significance before any AI narrates. Gaps get named: presence, prominence, reason.

Route

Each gap maps to an owner: your record (SEO, content, entity) or the consensus (placements, rankings, reviews).

Attribute

Shipped work becomes dated markers on the trend. When a number moves, you see what preceded it.

Every number carries a confidence interval. Movements inside the noise band are greyed and never reported as wins. If we can't distinguish your +2 from dice, we say so.
Model updates are tripwired. Canary prompts detect engine version changes; deltas across a discontinuity are annotated, not attributed. Nobody takes credit for Google's release schedule.
Arabic-first

Your customers don't ask in English

SOVA measures formal Arabic, Saudi dialect, and English separately — because the answers differ, and the difference is strategy. Global tools don't see this at all.

Saudi dialect
وين أحسن مستشفى ولادة بالرياض؟
63%
mention rate · how patients actually ask
Formal Arabic (MSA)
ما هو أفضل مستشفى للولادة في الرياض؟
49%
mention rate · same question, formal register

A 14-point gap on the same question, invisible to any tool that treats Arabic as one language. Which register of the Arabic web you invest in is a measurable decision.

SOVA INDEX · QUARTERLY

Riyadh Private Healthcare
AI Visibility Index

01  Market leader 58
02  High-volume peer 29
03  Authority brand 27
04  ··· your brand?
Published quarterly

The ranking your market will be reading

  • Same methodology for every brand. Nobody pays to be ranked. The formula is versioned and public.
  • Free public rankings, paid drill-downs. The league is open; the why-and-what-to-do behind your position is the product.
  • New verticals each quarter. Healthcare first, then banking, telecom, real estate — Saudi market, both languages.
Start with the truth

See where you stand before your competitors do

A baseline measurement: your category's question bank across the major engines, three metrics with confidence intervals, the competitor league, and the three gaps that matter most — before you spend a riyal on fixes.

Response within two working days · Riyadh-based · AR/EN