Facts→Bank→2–Review, Session, Run→5–Pipeline → Insights
Facilities & entity transitions
client-confirmed · 09 Jun 2026| Facility | Geo | Former name | Tier impact |
|---|---|---|---|
| Dallah Nakheel | Riyadh | - | CORE |
| Dallah Namar | Riyadh | - | CORE |
| Dallah Khobar | Khobar | Al Salam Hospital ⚠ transition | CORE · geo segment |
| Dallah Al Ahsa | Al-Ahsa | Al Ahsa Hospital ⚠ transition | CORE · geo segment |
| Makkah Medical Center 92% | Makkah | - | CORE · geo segment |
| Dallah Al Arid | Riyadh | under construction · 2027 | WATCH |
⚠ transitions: old names carry years of corpus + reviews; new names carry months. Both names live in the entity dictionary; SOV counts either, and the readout tracks the migration.
Group structure · affiliates ≠ competitors
| Entity | Stake | Counts as |
|---|---|---|
| Kingdom Hospital, Riyadh | 58.64% | GROUP SOV |
| Dr. Mohammed Al Fagih, Riyadh | 31.21% | AFFILIATE · flagged |
| IMC Jeddah | 27.18% | AFFILIATE · flagged |
Entity dictionary · aliases
feeds extractor v4dallah ← دلة · مستشفى دلة · Dallah Hospital · Dallah Health · مستشفى دلة نمار · Dallah Khobar · مستشفى السلام بالخبر · Al Salam Hospital Khobar
competitors confirmed ← الحبيب · السعودي الألماني · فقيه · التخصصي · SMC · المواساة
Prompt sources · two paths into the bank
PATH A · PRIMARY, UPLOAD YOUR OWN
XLSX / CSV import
Your prompts, call-center logs, GSC queries, app search terms. Column mapping on upload → dedup → classifier → review queue.
last upload: prompt_bank.xlsx · 2,240 rows · 09 Jun
PATH B · SECONDARY, GENERATE FOR GAPS
AI Generator, seeded by this page
Fills coverage the upload missed, e.g. no Khobar-dialect prompts arrived, the Generator drafts them from the confirmed facility list. Output always lands as unreviewed drafts, never straight to active.
detected gaps: Khobar dialect ×0 · Al-Ahsa OB-GYN ×2
Bank→Review→Session→Run→5–Read results
Active prompts
236
core 214 · canary 22
Awaiting review
38
imported from XLSX v2.1
Duplicates flagged
1,847
from 2,240 imported → 293 intents
Bank version
v3.0
frozen 01 Jul 2026 · immutable
Prompts
All tiersCoreCanaryExpansion (quarantined)PediatricsOB-GYNDialect onlyShare-of-voiceCitation probe
| ID | Prompt | Specialty | Intent | Register | Measures | Tier | Source | Status |
|---|---|---|---|---|---|---|---|---|
| PED-014 | وين أفضل طوارئ أطفال ٢٤ ساعة بالرياض؟ Where's the best 24h pediatric ER in Riyadh? |
Pediatrics | Local / urgent | Dialect | SOV | CORE | call-center | active |
| PED-021 | أبغى دكتور قلب أطفال شاطر، مين تنصحون؟ Need a good pediatric cardiologist, who do you recommend? |
Pediatrics | Recommendation | Dialect | SOV | CORE | call-center | active |
| OBG-003 | ما أفضل مستشفى خاص للولادة القيصرية في الرياض؟ Best private hospital for C-section in Riyadh? |
OB-GYN | Recommendation | MSA | SOV | CORE | WS1 v2.1 | active |
| OBG-047 | Dallah vs Al Habib for delivery, which is better and why? | OB-GYN | Comparison | English | SOV | CORE | GSC | active |
| CRD-009 | هل قسطرة القلب في مستشفى دلة موثوقة؟ وكم تكلفتها تقريباً؟ Is cardiac cath at Dallah reliable? Rough cost? |
Cardiology | Brand + cost | MSA | SOV | CORE | WS1 v2.1 | active |
| INF-112 | ما أعراض سكري الحمل ومتى يكون خطراً؟ Symptoms of gestational diabetes, when is it dangerous? |
Diabetes | Informational | MSA | CITE | CORE | WS1 v2.1 | active |
| CAN-002 | أفضل مستشفى خاص في الرياض head query, weekly canary | Cross | Recommendation | MSA | SOV | CANARY | curated | active |
| EXP-201 | أفضل مستشفى نساء وولادة في الخبر؟ Best OB-GYN hospital in Khobar?, Dallah has no Khobar facility |
OB-GYN | Recommendation | MSA | SOV | EXPANSION | WS1 v2.1 | quarantined |
| DUP-1148 | ما هو أحسن مستشفى للولادة القيصرية بالرياض؟ 98% similar to OBG-003, merged |
OB-GYN | Recommendation | MSA | - | - | WS1 v2.1 | duplicate |
showing 9 of 1,412 · 236 active
Bank→Review→Session→Run→5–Read results
PED-052 · draft 38 of 38
imported · XLSX v2.1ولدي عمره ٣ سنوات وعنده ربو، أي مستشفى أروح له الحين الساعة ٢ بالليل؟
My 3-year-old has asthma, which hospital do I go to right now at 2am?
| Specialty | Pediatrics conf 0.97 |
| Intent | Local / urgent conf 0.94 |
| Register | Saudi dialect conf 0.99 |
| Measures | Share-of-voice conf 0.91 |
| Geo | Riyadh conf 0.88 |
| Suggested tier | CORE high-intent, real patient voice |
| Near-duplicates | PED-014 · 71% similar, kept, different intent shape |
Queue summary
| High-confidence (auto-approvable) | 29 |
| Needs human eyes (conf < 0.85) | 6 |
| Flagged duplicate | 3 |
Classifier
extractor v4 · haiku-class
Every override you make is logged and becomes few-shot material for the next classifier version. Classification prompts are versioned like everything else, re-runs over history stay comparable.
Bank→Review→Session→Run→5–Read results
1 · Prompt set
Bank v3.0 (frozen snapshot)
Core · 214
Canary · 22
Expansion · 118
214 prompts selected · 96 dialect / 82 MSA / 36 English
2 · Models & modes
| Model | Search ON | Search OFF | $/1K calls | |
|---|---|---|---|---|
| ☑ | Claude Sonnet 4.6 | ☑ retrieval visibility | ☑ parametric strength | ~9.40 |
| ☑ | GPT-5 mini | ☑ | ☑ | ~7.10 |
| ☑ | Gemini 2.5 Flash + grounding | ☑ | n/a via API | ~5.80 |
| ☐ | Perplexity Sonar | - | - | - |
3 · Sampling
N = 5N = 10 · recommendedN = 20
At N=10, the 95% confidence interval on a mention rate is roughly ±9 points per prompt, tight enough at category level (55+ prompts pooled), too wide to report per-prompt movement.
Session estimate
JUL-2026 monthly| API calls | 10,700 |
| 214 × 10 × (3 on + 2 off) | |
| Extraction pass | 10,700 calls |
| Est. runtime | ~4.5 hrs |
| Est. cost | $96–128 |
Raw responses stored permanently. Model version strings logged per call, discontinuities get annotated on every chart.
Session→Run→Pipeline→6–Results
READY
Session JUL-2026 configured, bank v3.0 · 214 prompts · 3 models · N=10 · 10,700 calls
Session
JUL-2026
bank v3.0 · frozen snapshot
Progress
0%
0 / 10,700 calls
Failures → retried
0
0 dead after 5 attempts
Cost so far
$0.00
est. $96–128
Overall
Claude Sonnet 4.6 claude-sonnet-4-6 · 6 rps0 / 4,280
GPT-5 mini gpt-5-mini-2026-05 · 8 rps0 / 4,280
Gemini 2.5 Flash grounding-on only · 10 rps0 / 2,140
Live capture
raw responses + fan-out queries streaming to store| Time | Prompt | Model / mode | Fan-out queries captured | Brands seen |
|---|---|---|---|---|
| start the run to stream captures — | ||||
Pipeline→Sessions→Dashboard
Sessions completed
6
FEB–JUL 2026 · monthly
Total runs stored
61.4K
raw, permanent, re-extractable
Engagement spend
$641
API total · 6 sessions
SOV journey
27→41%
FEB → JUL · search-on
All sessions
click a row → that month's dashboard · select two → compare| Session | Bank | Models · N | Runs | SOV ±CI | Δ | Flags | Cost |
|---|---|---|---|---|---|---|---|
| JUL-2026 | v3.0 | CL · G5m · GEM · N10 | 10,700 | 41% ±3 | +6 ★ | ⚠ gemini 12 Jul action ×1 | $112 |
| JUN-2026 | v3.0 | CL · G5m · GEM · N10 | 10,700 | 35% ±3 | +2 · noise | action ×2 | $109 |
| MAY-2026 | v2.2 | CL · G5m · GEM · N10 | 10,450 | 33% ±3 | +2 · noise | ⚠ gpt-5m 12 May bank v2.1→2.2 | $107 |
| APR-2026 | v2.1 | CL · G5m · GEM · N10 | 10,450 | 31% ±3 | +2 · noise | - | $104 |
| MAR-2026 | v2.1 | CL · G5m · N10 | 8,560 | 29% ±3 | +2 · noise | gemini added next | $96 |
| FEB-2026 | v2.0 | CL · G5m · N10 | 8,560 | 27% ±3 | BASELINE | audit delivered | $93 |
Note the honesty pattern: four consecutive "+2 noise" months compound into +14 significant FEB→JUL. No single month was reportable; the trend is.
Canary status
22 prompts · weekly · discontinuity tripwireCLAUDE
41 days stable
GPT-5 MINI
63 days stable
GEMINI ⚠
2 days, discontinuity 12 Jul
version string changed · JUL Gemini deltas annotated, not attributed
Sessions→Dashboard→8–Competitors, Citations, Evidence
CANARY
Claude 41d ✓
GPT-5m 63d ✓
Gemini ⚠ 2d, discontinuity 12 Jul
JUL session · all runs in
sessions →
All specialties
Pediatrics
OB-GYN
Cardiology
Dermatology
Orthopedics
MENTION · all specialties
41%
Wilson 95% CI ±3 · n = 6,420
▲ +6 pts vs JUN · significant
appears in the answer at all
TOP-3
24%
Wilson 95% CI ±3
+2 pts · within noise
among the first three options
RECOMMENDED
11%
Wilson 95% CI ±2
+1 pt · within noise
framed as a pick, not just listed
Share of voice · per model
shaded = 95% CI
Dialect vs MSA mention
63 / 49%
±4 each
+14 pt dialect advantage · significant
Search-on vs search-off
41 / 18%
the two-channel split
retrieval ≫ parametric · expected
Own-site citation share
2.1%
dallah-hospital.com
vs hmg.com at 6.9% · 3.3× gap
Mention rate · specialty × intent
0%100%
⚑ Pediatrics × Best-of 67% mentioned, 13% top-3, the listed-not-recommended pattern, auto-flagged for the JUL readout. Click any cell for the raw responses behind it.
JUL vs JUN · movement by category
greyed = inside noise band · do not report| Category | JUN | JUL | Δ | 95% CI | Verdict |
|---|---|---|---|---|---|
| OB-GYN · dialect | 88.2% | 96.4% | +8.2 | ±3.1 | significant ▲ |
| Pediatrics · best-of | 65.1% | 67.0% | +1.9 | ±4.6 | within noise |
| Cardiology · comparison | 62.3% | 64.1% | +1.8 | ±3.9 | within noise |
| Dermatology · best-of | 57.2% | 51.0% | −6.2 | ±4.4 | significant ▼ |
| Orthopedics · best-of | 59.1% | 58.0% | −1.1 | ±4.1 | within noise |
Run→Pipeline→Dashboard→Insights
Analysis pipeline · raw answers to readout
two AIs · one deterministic math layer between them · human at the endSTAGE 1 · MODELS UNDER TEST
Raw answers
10,700 responses + fan-out queries + cited URLs. Stored forever, never edited.
→
STAGE 2 · AI, EXTRACTOR
Per answer · ×10,700
Haiku-class. Text → structured rows: brands, position, framing, domains, claims. Extraction only, no interpretation.
→
STAGE 3 · CODE, AGGREGATION
Pure math, no AI
Rates, CIs, deltas, significance flags, noise suppression. Significance and noise flags are computed here.
→
STAGE 4 · AI, ANALYST
Per session · ×1
Sonnet-class. Reads metrics, not raw text + few examples for color. Drafts the readout, maps gaps → playbook.
→
STAGE 5 · HUMAN
You, editor
Every claim traces: sentence → metric → extractor rows → raw answer. Nothing ships unreviewed.
Extractor QA · golden set gate
extractor v4 vs 100 hand-labeled responses| Check | Arabic | English | Gate ≥ |
|---|---|---|---|
| Brand mention recall | 97.2% | 99.1% | 95% |
| Alias resolution (دلة/Dallah/مستشفى دلة) | 98.4% | 99.5% | 97% |
| Recommended-vs-listed framing | 91.8% | 95.3% | 90% |
| Cited-domain capture | 99.0% | 99.2% | 98% |
v4 passed, deployed 30 Jun. Arabic framing at 91.8% is the weak edge: dialect "ينصحون فيه" cases. 14 new labeled examples queued for v5. A version that fails any gate cannot deploy, extractor errors are data errors.
One answer through the pipeline
run #4417 · extractor v4 output{
"run_id": 4417, "extractor_v": 4,
"brands": [
{"entity":"kfshrc","pos":1,"framing":"recommended"},
{"entity":"habib","pos":2,"framing":"recommended"},
{"entity":"ksmc","pos":3,"framing":"listed"},
{"entity":"dallah","pos":4,"framing":"listed"}],
"cited_domains": ["vezeeta.com","almrsal.com",
"expatriates-guide.com"],
"client_claims": [{"claim":"has pediatric dept",
"issue":"omits dedicated children's hospital"}]
}
"run_id": 4417, "extractor_v": 4,
"brands": [
{"entity":"kfshrc","pos":1,"framing":"recommended"},
{"entity":"habib","pos":2,"framing":"recommended"},
{"entity":"ksmc","pos":3,"framing":"listed"},
{"entity":"dallah","pos":4,"framing":"listed"}],
"cited_domains": ["vezeeta.com","almrsal.com",
"expatriates-guide.com"],
"client_claims": [{"claim":"has pediatric dept",
"issue":"omits dedicated children's hospital"}]
}
Aggregation then computes: PED mention 67% ±5, top-3 13% ±4 → flags listed-not-recommended → analyst receives the flag, not the 165 raw texts.
Dashboard→Competitors→Citations
All specialties
Pediatrics
OB-GYN
Cardiology
Dermatology
Share of voice · all brands
JUL-2026 · search-on · all specialties · n=6,420| Brand | SOV | Top-3 | Recommended* | Δ vs JUN |
|---|
مستشفى المملكة is 58.64% Dallah-owned (per Step 0), shown for completeness, counted toward group SOV, never as a competitor. *Recommended = framed as a pick, not merely listed. الملك فيصل التخصصي: only brand whose recommended-rate ≈ its top-3 rate, authority framing. فقيه Riyadh +9 pts in one month = new entrant momentum, watch item.
Who wins each category
brand with highest recommended-rate| Category | Winner | Their edge (from citations) |
|---|---|---|
| Pediatrics | الحبيب 58% | branch density + "book same day" framing in 3 directories |
| OB-GYN | دلة 54% | delivery-package pages cited directly, the model to copy |
| Cardiology (complex) | التخصصي 61% | research halo; "for complex cases" framing near-universal |
| Dermatology | السعودي الألماني 39% | weak winner, lowest-consensus category, most takeable |
What sources say about الحبيب vs دلة
framing phrases · extractor field · JUL| الحبيب, top framings | × |
|---|---|
| حجز سهل وسريع عبر التطبيق | 311 |
| فروع متعددة في الرياض | 287 |
| أسعاره مرتفعة نسبياً | 142 |
| دلة, top framings | × |
|---|---|
| مستشفى عريق وموثوق | 176 |
| جيد للولادة ومتابعة الحمل | 148 |
| لا ذكر لمستشفى الأطفال المتخصص | 0 |
The children's hospital appears in zero framing phrases across 6,420 answers, the consensus gap, quantified.
Competitors→Citations→Evidence
Cited-domain league table
All categoriesPediatricsOB-GYNArabic answers
| # | Domain | Citation share | Δ vs JUN | Dallah present? | Action |
|---|---|---|---|---|---|
| 1 | vezeeta.com | 18.2% | +0.4 | yes · profile thin | enrich profile: subspecialties, AR bios |
| 2 | altibbi.com | 14.7% | −0.2 | no | placement target, highest-value gap |
| 3 | expatriates-guide.com | 11.3% | +0.8 | yes · 2019 copy | outreach: update stale entry |
| 4 | sehatuk.sa | 8.1% | NEW ★ | yes | May placement, entered answers in 6 wks, driver of the +6 |
| 5 | webteb.com | 7.6% | +0.1 | no | placement target |
| 6 | hmg.com (الحبيب own site) | 6.9% | −0.3 | - | competitor's site cited directly, extractable service pages work |
| 7 | almrsal.com | 5.2% | −0.6 | yes | hold |
| 8 | dallah-hospital.com | 2.1% | +0.2 | - | own site barely cited: service pages not extractable, rewrite per playbook |
Concentration
why this list is shortTop 8 domains carry 74% of all citations across 6,420 answers. The comparison layer for Riyadh healthcare is ~a dozen pages. Presence on 6 specific URLs ≈ presence in most AI answers. That's the entire placement strategy, and it's measurably finite.
hmg.com vs dallah-hospital.com
own-site citation rate, the extractability scoreboard6.9%
hmg.com
2.1%
dallah-hospital.com
3.3× gap
Models cite الحبيب's own service pages 3.3× more often. Their pages state prices, doctor names, booking steps, extractable. Dallah's pages describe philosophy. Same finding as the framing table, from a second independent angle.
…→Dashboard→Evidence→Insights
Pediatrics · listed-not-recommended · 165 runs · session JUL-2026
Runs behind this cell
PED · 55 prompts × N10 × 3 models| Run | Prompt | Model | Result |
|---|---|---|---|
| #4417 | PED-021 | claude | rank 4 |
| #4418 | PED-021 | claude | rank 3 |
| #4423 | PED-021 | gpt-5m | not mentioned |
| #4426 | PED-014 | gemini | rank 5 |
| #4431 | PED-014 | claude | rank 2 |
Run #4417 · raw response
claude-sonnet-4-6 · search-on · 14:02:11FAN-OUT QUERIES
"pediatric cardiologist Riyadh" · "أفضل دكتور قلب أطفال الرياض" · "children heart hospital Riyadh"
RESPONSE · extracted entities highlighted
لأمراض قلب الأطفال في الرياض، أبرز الخيارات: مستشفى الملك فيصل التخصصي للحالات المعقدة، مستشفى د. سليمان الحبيب (فرع العليا) لسهولة الحجز، ومدينة الملك سعود الطبية. يمكن أيضاً النظر في مستشفى دلة الذي يضم قسم أطفال…
EXTRACTION
| Dallah mentioned | yes · position 4 · framing: "also consider", listed, not recommended |
| Cited domains | expatriates-guide.com · almrsal.com · vezeeta.com |
| Claim check | response omits Dallah's dedicated children's hospital, consensus gap |
Evidence→Insights→loop → bank v3.1
Draft assembled 13 Jul, 07:00 AST
5 insights · 2 significant · 1 accuracy issue
▸ Reference, gap taxonomy (the legend behind the chips above)
| Gap chip | What the signal means | Standard fix |
|---|---|---|
| Presence · rank | your pages don't rank for the fan-out queries | SEO, build/optimize pages for the captured queries |
| Presence · omission | the cited third-party pages don't include you | placement/outreach on those specific URLs |
| Presence · surface | answer built from Maps, not web (Gemini local) | Business Profile + review volume |
| Prominence | mentioned but buried, weak consensus strength | rise within sources · add sources · authority markers |
| Reason | listed, not recommended, no differentiating claim | publish concrete differentiators + 2–3 third parties repeating them |
| Entity | alias confusion, false facts, split identity | consistency fixes, schema, GBP, Wikipedia, site FAQ |
| Parametric | weak with search OFF, the model's memory itself | slow channel: sustained corpus over years; don't sell quarterly fixes |
Shipped work → timeline markers
why movement can be attributedEvery shipped action gets a dated marker on the trend charts, the same treatment as model updates. When a metric moves, the timeline shows what preceded it: the May placement sits 6 weeks before the JUL +6, inside the known 4–8 week retrieval lag. Attribution with dates, not vibes.
| Shipped | Action | Outcome so far |
|---|---|---|
| 14 May | sehatuk.sa placement | ✓ cited JUL · driver of +6 |
| 02 Jun | vezeeta profile, AR bios | watching · in lag window |
| AUG plan | pediatric sub-site AR-first | marker added on ship |
Loop back · this readout changes next month's questions
NEW FAN-OUT QUERIES → DRAFT PROMPTS
| مستشفى ولادة طبيعية بدون تدخل الرياض | ×41 | |
| pediatric ER wait time Riyadh | ×28 | |
| تكلفة الولادة القيصرية ٢٠٢٦ | ×24 | |
| أفضل دكتورة نساء وولادة في مستشفى دلة | ×22 | |
| NICU level 3 private hospital Riyadh | ×19 | |
| مستشفى يقبل تأمين بوبا للولادة الرياض | ×17 | |
| children hospital Riyadh north | ×17 | |
| مستشفى دلة الخبر أم السلام سابقاً؟ تقييم | ×14 | |
| IVF success rates Riyadh hospitals | ×12 |
queries recur ≥ 2 sessions · sorted by frequency
BANK v3.1 PROPOSAL, REBALANCING
| Dermatology fell significantly, investigate + weight up | +8 |
| Khobar entity transition (Al Salam → Dallah) tracking | +6 |
| Retire: 5 prompts silent & redundant 3 sessions | −5 |
THE LOOP
readout → new prompts + timeline markers + rebalancing → bank v3.1 → AUG session
back to step 0 →