Zara holds 92% of AI answers
without its own site being cited once.
Seven months of weekly tracking. Both tracked prompts are pure market questions — no brand name — and in 386 stored answers, zara.com is never the source. The visibility rides entirely on third parties.
The climb.
Share of voice, monthly averages of weekly scans.
Market questions only — and zero own-site citations.
Both prompts are brand-free by design: this is discovery territory. The brand wins the answer without ever being the source.
| Tracked prompt | Type | Stored answers | Own site cited | Rate |
|---|---|---|---|---|
| “Dónde comprar ropa de moda en Madrid?” | Market | 236 | 0 | 0% |
| “donde comprar ropa en madrid?” | Market | 150 | 0 | 0% |
The citation ledger — before and after.
The raw accounting, market questions only, branded prompts excluded. Every number counted from the stored answers, not estimated.
| Metric — market questions only | Before (Nov–Jan) | After (Feb–Jun) | Change |
|---|---|---|---|
| Stored answers analysed | 23 | 518 | — |
| Total citations inside those answers | 482 | 12,183 | ×25 |
| Unique domains cited | 95 | 419 | ×4.4 |
| Citations of zara.com | 0 (0.0%) | 0 (0.0%) | zero — blocked |
The market’s source graph.
The domains the AI cites when answering “where to shop in Madrid” — the market view. Travel and city content dominates; Zara rides in it.
| Source | Before | After |
|---|---|---|
| 14 | 253 | |
| 178 | 87 | |
 instagram.com | 20 | 87 |
| 12 | 81 | |
| 12 | 51 | |
| 14 | 50 | |
| 3 | 47 | |
| 7 | 45 | |
| 9 | 37 |
The cause is in the citations.
Same windows, every citation counted from the stored runs: 516 before, 16,421 after (the early window is small — 31 runs). The answer changed genre.
| Source | Share before | Share after | Change |
|---|---|---|---|
| 5.6% | 3.0% | -2.6 pts | |
| 2.9% | 1.5% | -1.4 pts | |
| 3.1% | 1.9% | -1.2 pts | |
| 1.6% | 0.4% | -1.2 pts | |
| 1.4% | 3.5% | +2.1 pts | |
| 5.4% | 8.6% | +3.2 pts | |
| instagram.com | 3.9% | 5.2% | +1.3 pts |
| 0.0% | 1.0% | +1.0 pts |
The proof: which pages put Zara in the answer.
858 stored runs on the Madrid shopping questions; the brand appears in 88% of them. For every stored run we check two things in the same answer: which sources it cites, and whether the brand’s name appears in the answer text. Comparing runs where a source is present vs absent isolates each source’s effect. Correlation measured across runs — not a causal claim — sources with enough runs to matter only.
| Source cited in the run | Runs | Brand in answer when cited | When absent | Effect |
|---|---|---|---|---|
| 90 | 100% | 86% | +14 pts | |
| 135 | 99% | 85% | +14 pts | |
| 68 | 100% | 86% | +14 pts | |
| 172 | 98% | 85% | +13 pts | |
| 72 | 100% | 86% | +14 pts | |
| 52 | 42% | 90% | -48 pts | |
| 56 | 50% | 90% | -40 pts | |
| instagram.com/columbus_vintage | 21 | 33% | 89% | -56 pts |
| 26 | 38% | 89% | -51 pts |
The fan-out layer: where the gaps live.
Every prompt a user asks explodes into sub-queries the engines run silently. We track each one. Most appear once and vanish — nobody owns them yet. That long tail is the gap map.
| Sub-query the engines actually ran | Times seen | Engine consensus |
|---|---|---|
| “mejores zonas comprar ropa moda madrid 2025 2026” | 69× | 1 engines |
| “dónde comprar ropa de moda en madrid 2026” | 67× | 1 engines |
| “mejores tiendas ropa moda madrid centro” | 31× | 1 engines |
| “dónde comprar moda española auténtica en madrid” | 24× | 1 engines |
The playbook this case proves.
Zara’s position was built on earned media. In the suite, that channel is systematic.
| The move | What it does | In the suite |
|---|---|---|
| Press & city-guide presence | The outlets AI reads — find, pitch, track placement | Get Cited |
| Buy into the domains AI already cites | timeout.es-tier sources → a shopping list with measured lift | AI Links |
| Own the unclaimed tail | 706 one-off sub-queries → a publishing queue | Content Hub |
The technical reason for zero.
We tested zara.com with three user-agents — a normal browser, GPTBot (OpenAI’s crawler) and PerplexityBot. All three got the same response, even for robots.txt:
GET zara.com/robots.txt [GPTBot/1.0] → 403 Forbidden
GET zara.com/robots.txt [PerplexityBot/1.0] → 403 Forbidden
More cases, same measurement.
The full split: 57% own-site citations in brand questions, 0.7% in market questions — and what closes that gap.
The review-ecosystem case: in market questions, apple.com is cited in under 1% of answers — the tech press carries it.
The decline case: −23 points in one week when the engines dropped Yelp from the answer. Borrowed visibility, gone overnight.
Run the same measurement on your brand.
Free scan, live in ~15 seconds. Your visibility, your gaps, the sources AI cites instead of you.
Scan my brand free → Book a demoObservational study of public AI answers, measured weekly by AI Rankia (Nov 2025 – Jun 2026). Per-engine presence rates; early-window samples are smaller. Brand names are trademarks of their respective owners, referenced for identification only; no affiliation or endorsement implied.