AI Search for Tattoo Studios in Berlin (2026):same city, new vertical — and the baseline snaps back
TL;DR: This is the series’ first same-city replication: Berlin, already measured for yoga, re-measured for tattoo studios with identical languages and proxies. Two things moved sharply. Perplexity’s booking-platform share — 31% for Berlin yoga via Urban Sports Club and Eversports — fell to 4% for tattoo, because no comparable booking layer exists for tattoos. And ChatGPT’s own-website share came back to 35%, right where the service-vertical cases (Paris yoga 32%, Berlin yoga 32%) sit, after Tokyo bookstores (8%) and Marseille coffee (10%) had broken it. A third surprise: asking in German vs English barely changes the answer — the control prompt’s top-5 overlap is 67%, by far the highest we’ve measured (prior best 25%). The engines’ personalities didn’t move; the vertical’s infrastructure did.
Executive Summary
Holding the city constant and swapping the vertical is the cleanest experiment this series has run — and on this evidence the vertical predicts the source mix better than the city does.
After Tokyo (8%) and Marseille (10%) had me wondering whether ChatGPT was drifting away from business websites altogether, Berlin tattoo lands at 35% own-website citations — in line with Paris yoga (32%), Berlin yoga (32%) and Amsterdam bikes (42%). The pattern that looked like decay was a property of food-and-retail scenes with thin owned webs. Tattoo studios, like yoga studios, run real websites (388 of the 531 seed studios list one on their Google profile — 531 is the filtered seed, before the 20 studios later recovered via Places brought the registry to 551), and the engines use them.
The sharper result is what happened to Perplexity. In the Berlin yoga study, booking marketplaces — Urban Sports Club, Eversports, ClassPass — carried 31% of its citations. Same city, same engine, same languages, tattoo prompts: 4%. The tattoo world’s closest equivalents (Fresha, Treatwell, Tattoodo) barely register — Fresha peaks at 3.6% of Perplexity’s citations, and Tattoodo, the vertical’s own marketplace, appears 3 times in 5,376 cited URLs. Yoga classes are a subscription product with an aggregator industry; a tattoo is a one-off commission, and the citation graph knows it. With the aggregators gone, Perplexity becomes the #2 entity engine here (64% studio websites) — though at 63% Gemini is level with it inside the noise, so read them as tied behind Copilot (95%) rather than ranked.
One layer the registry can’t see: Berlin’s famous private artists. Chaim Machlev (DotsToLines) was recommended in 25 answers and returns no Google listing on a name-and-city lookup; the same holds for Peter Aurisch and Valentin Hirsch. For tattoo, the artist is the brand, and a chunk of AI’s actual recommendations lives outside the map-based universe every local-SEO tool measures. Section 7 quantifies it — and separates two things we originally conflated:not resolved and not listed.
Reading note: citations = URLs an engine used as support; mentions = studios named in the visible answer text. Citation score = a plain count of times a studio was pointed at rather than named: one point per cited URL on its own domain (any engine), plus one per appearance by name in ChatGPT’s map panel. It is unweighted, and it depends on the registry knowing the studio’s domain — which is exactly why NAJS scores 0 (section 4). Source-mix sections count citations; the leaderboard counts mentions. Keeping the two apart is what catches counting artifacts (see the NAJS case, section 4).
Source mix by platform
Each of the 5,376 cited URLs was bucketed into a nine-part taxonomy. Berlin tattoo produces the most website-heavy mix this series has recorded: 47% of all citations point at a studio’s own site.
95% — the entity engine returns
Copilot had been sliding down the series — 95% in Berlin yoga, 89% in Tokyo, 83% in Marseille. For Berlin tattoo it’s back at 95%. The slide tracked the owned-web density of each scene — the engine itself never moved.
A one-two of websites and city media
35% studio websites, then 33% local editorial — dominated by top10berlin.de (93 cites, 18% of ChatGPT’s total) and the-berliner.com (49). Reddit takes another 20%. Third parties still matter to ChatGPT; they just sit beside the studios’ own sites instead of replacing them.
No trade press to reach for
In Marseille, Gemini gave a third of its citations to Barista Magazine. Berlin tattoo has no equivalent authority outlet Gemini trusts — tattoo magazines took under 1% — so it behaves like a normal entity engine: 63% studio websites, 20% Berlin city media.
The booking-platform collapse: 31% → 4%
This was the hypothesis the study was designed to test. When Berlin yoga showed Urban Sports Club, Eversports and ClassPass swallowing 31% of Perplexity’s citations, there were two possible readings: something about Berlin, or something about yoga. Re-running Berlin with a vertical that has no class-pass economy settles it.
The booking layer didn’t just shrink for Perplexity — it’s near-absent across all five engines: 126 booking citations in 5,376 total (2.3%). And the composition is telling. The generic beauty-appointment platforms carry what little exists: Fresha (105 cites) and Treatwell (17). Tattoodo — the marketplace built specifically for this industry — was cited 3 times. Booksy: once.
Why the difference? A yoga studio’s inventory is recurring classes, which is exactly what aggregators index, price and resell — giving engines a data-rich page to cite for “book a class near me.” A tattoo is a custom commission negotiated with an artist. There’s no timetable to aggregate, so the aggregator layer never grew the content depth that earns citations. The engines aren’t choosing differently; there’s nothing there to choose.
ChatGPT’s own-website share: back to 35%
Six verticals in, the own-website series now reads 32, 32, 42, 8, 10, 35 — and the split is no longer mysterious. Where businesses sell a service they book or explain on their own site (yoga, bike shops, tattoo studios), ChatGPT grounds a third or more of its answer there. Where the scene documents itself on third parties (bookstores, cafés), the share collapses.
Instagram got zero ChatGPT citations
Going in, my bet was the opposite — tattoo culture lives on Instagram, and 73 of the seed studios literally list an Instagram URL as their Google-profile website. Yet of ChatGPT’s 507 citations, instagram.com appears 0 times. The 125 Instagram citations that do exist come from AI Mode (86), Copilot (24) and Perplexity (15). Marseille’s 237-cite Instagram signal turns out to be local to Marseille rather than a property of visual verticals.
Reddit stays a ChatGPT habit
reddit.com drew 237 citations — only google.com’s AI Mode self-references and top10berlin.de rank higher — and 101 of them (43%) sit in ChatGPT, where Reddit alone is 20% of the engine’s citations. Perplexity, which indexes the same threads, used it 8 times (0.6%). Same open web, very different appetites.
Who the engines actually recommend
Ranked by answer mentions — the number of captures (of 520) whose visible answer names the brand. OMEN Tattoo — three locations across Mitte, Kreuzberg and Friedrichshain — tops both metrics at once: 177 answer mentions and the highest citation score (246). Marseille — the only earlier study to publish both rankings — had them split by an artifact; here they simply agree.
Per-engine mention matrix
| # | Studio | AI Mode104 prompts | ChatGPT104 prompts | Perplexity104 prompts | Gemini104 prompts | Copilot104 prompts |
|---|---|---|---|---|---|---|
| 1 | OMEN Tattoo | 44.2%(46) | 40.4%(42) | 14.4%(15) | 35.6%(37) | 35.6%(37) |
| 2 | Berlin Ink Tattooing | 25%(26) | 31.7%(33) | 21.2%(22) | 7.7%(8) | 43.3%(45) |
| 3 | Good Old Times Tattoo Berlin | 39.4%(41) | 16.3%(17) | 27.9%(29) | 19.2%(20) | 25%(26) |
| 4 | OEK FACTORY TATTOO | 17.3%(18) | 45.2%(47) | 34.6%(36) | 2.9%(3) | 22.1%(23) |
| 5 | Ivory Tattoo Berlin | 16.3%(17) | 14.4%(15) | 33.7%(35) | 6.7%(7) | 44.2%(46) |
| 6 | ETERNITY TATTOO BERLIN | 26%(27) | 22.1%(23) | 18.3%(19) | 7.7%(8) | 30.8%(32) |
| 7 | Iron City Tattoo | 17.3%(18) | 34.6%(36) | 12.5%(13) | 0(0) | 38.5%(40) |
| 8 | NAJS Tattoo Studio | 17.3%(18) | 23.1%(24) | 31.7%(33) | 0(0) | 28.8%(30) |
| 9 | Mugshot Tattoo | 26%(27) | 20.2%(21) | 5.8%(6) | 16.3%(17) | 14.4%(15) |
| 10 | Black Smoke Tattoo Berlin | 24%(25) | 19.2%(20) | 0(0) | 2.9%(3) | 23.1%(24) |
| 11 | NOIR Berlin | 24%(25) | 20.2%(21) | 0(0) | 16.3%(17) | 7.7%(8) |
| 12 | Inklabs Tattoo Studio | 3.8%(4) | 10.6%(11) | 13.5%(14) | 10.6%(11) | 22.1%(23) |
| # | Brand | Answer mentions | Citation score | Engines (of 5) |
|---|---|---|---|---|
| 1 | OMEN Tattoo | 177 | 246 | 5 |
| 2 | Berlin Ink Tattooing | 134 | 125 | 5 |
| 3 | Good Old Times Tattoo Berlin | 133 | 168 | 5 |
| 4 | OEK FACTORY TATTOO | 127 | 85 | 5 |
| 5 | Ivory Tattoo Berlin | 120 | 113 | 5 |
| 6 | ETERNITY TATTOO BERLIN | 109 | 100 | 5 |
| 7 | Iron City Tattoo | 107 | 79 | 4 |
| 8 | NAJS Tattoo Studio | 105 | 0 | 4 |
| 9 | Mugshot Tattoo | 86 | 82 | 5 |
| 10 | Black Smoke Tattoo Berlin | 72 | 40 | 4 |
| 11 | NOIR Berlin | 71 | 17 | 4 |
| 12 | Inklabs Tattoo Studio | 63 | 109 | 5 |
One listicle out-cites every studio in Berlin
The most-cited domain in the study isn’t a studio, Reddit or Google — it’s top10berlin.de, a Berlin best-of listicle site, with 256 citations from all five engines. The strongest studio domain (omentattoo.de, 198) doesn’t catch it.
| Domain | Engines | Cites | What it is |
|---|---|---|---|
| top10berlin.de | 5/5 | 256 | Berlin best-of listicle site — the study’s top domain |
| reddit.com | 4/5 | 237 | Community threads; 43% of its cites are ChatGPT |
| omentattoo.de | 5/5 | 198 | OMEN Tattoo — strongest studio domain |
| goodoldtimestattoo.com | 5/5 | 148 | Good Old Times Tattoo |
| the-berliner.com | 4/5 | 140 | Berlin city magazine (ex-Exberliner) |
| instagram.com | 3/5 | 125 | Studio profiles; zero of these come from ChatGPT |
| fresha.com | 4/5 | 105 | Beauty-appointment platform — the whole booking layer, basically |
| berlinink.de | 4/5 | 104 | Berlin Ink Tattooing |
| ivory-tattoo-berlin.de | 5/5 | 95 | Ivory Tattoo Berlin |
| eternitytattoo.com | 5/5 | 84 | Eternity Tattoo Berlin |
Full table: top_domains.csv (google.com excluded above — its 1,234 cites are AI Mode self-references, covered in section 10).
German and English mostly agree here — a series first
On the control prompt, ChatGPT’s EN and DE top-5 share 4 of 6 distinct studios — a 67% overlap where Berlin yoga and Paris yoga measured 25% and Marseille 11%. My best explanation: tattoo’s vocabulary is globalised. “Fine line,” “blackwork” and “walk-in” are the same words in German, the consensus studios publish in both languages, and the international guide layer covers the same names the German one does.
| Prompt template | Top-5 overlap |
|---|---|
| control (best tattoo studios) | 67% |
| style: minimalist | 67% |
| price: premium | 67% |
| district: Kreuzberg | 67% |
| style: blackwork | 43% |
| vegan-friendly | 43% |
| walk-in | 43% |
| price: cheap | 25% |
| booking (online booking) | 25% |
| persona: tourist | 11% |
| vibe: most famous artists | 11% |
| style: Japanese | 11% |
| style: realism | 0% |
Overlap = Jaccard similarity of the top-5 entity sets (shared ÷ combined). The table shows 13 of the 26 base templates; all 26: lang_overlap.csv.
The convergence isn’t uniform, and the exceptions carry the signal. Consensus prompts (control, minimalist, premium, Kreuzberg) converge at 67%. Prompts that route through different information ecosystems still split hard: realism lands at 0% — the German prompt surfaces German-forum favourites, the English one an entirely different set — and tourist and “most famous artists” prompts sit at 11%.
TLD coupling, the pattern Tokyo made famous (Japanese prompts citing .jp domains 5× more), is gone in this dataset: German prompts cite .de studio domains at 0.87× the English rate — effectively neutral — while .com studio domains skew mildly English (0.54×). When both language communities read the same local websites, the TLD split has nothing to bite on.
The artists our map layer can’t see
7.3% of extracted recommendations (253 of 3,444) never resolve to any Google Maps place — and the top of that unresolved pile isn’t noise. It’s Berlin’s most famous tattoo artists, working from private, appointment-only studios. Two different failures used to sit in here. A reader correction separated them, and one of the two has since been fixed.
| Artist | Answers naming them | Context |
|---|---|---|
| Chaim Machlev / DotsToLines | 25 | counted across 3 name variants; appointment-only; no listing found |
| Peter Aurisch | 6 | neo-traditional; private atelier; no listing found |
| Valentin Hirsch | 4 | engraving-style blackwork; private studio; no listing found |
Correction, 28 August 2026. This section originally listed a fourth artist here — Mo Ganji, the most-recommended independent name at 28 answers — and said none of them had a Google listing at all. That was wrong about him. He has one — Mo Ganji – One-Line Tattoo Artist Berlin, rated 5.0 from 3 reviews and categorised as a tattoo shop — and it was already sitting in our own entity table with coordinates and his website. What failed was the join: engines recommend him as “Mo Ganji”, the listing carries the long form, and none of his 28 mentions ever matched it.
That is now fixed in the pipeline, not just in this page. The resolver gained a containment tier — a short mention resolves to a longer listing name when exactly one candidate contains all of its words, and stays unresolved when several do. Mo Ganji’s 28 mentions resolve, he is out of the table above, and the study’s resolution rate moves from 90.4% to 92.7% (unresolved 332 → 253). Two leaderboard totals shifted with it: OMEN Tattoo 173 → 177 and Good Old Times 132 → 133. No rank changed, and at 28 mentions Mo Ganji still sits well below the 63 that 12th place required. The same fix recovered 431 mentions across nine of these studies.
For the other three, a name-and-city lookup against Google Places returns nothing. We no longer read that as proof of absence, because a Places lookup returned nothing for Mo Ganji either and he is plainly listed. The defensible claim is the narrower one: these artists are invisible to our Maps-based resolution, and to any tool built the same way. Whether a listing exists is a separate question we cannot settle from a null Places response.
Their web presence is a personal site or an Instagram account, and engines recommend them anyway. Note this does not put Mo Ganji on the leaderboard: 28 answers is well short of the 63 that 12th place required. This is still a structural feature of the tattoo vertical the other five studies never encountered — the artist can outrank the shop while being invisible to every Maps-based measurement, including ours.
Honest limit, stated plainly: the leaderboard in section 4 covers the Google-Maps-resolved universe. A Berlin artist deciding where the AI engines send first-timers should read both tables — and any tool that audits “AI visibility” purely off Maps data will miss this layer entirely.
551 studios, 12 winners, one map
The registry dots show how evenly tattooing covers Berlin — every Bezirk has studios. The engines’ favourites cluster anyway: of the top 12, most sit in the Kreuzberg / Friedrichshain / Neukölln belt plus Mitte.
All 551 registry tattoo studios in Berlin
Top 12 most-recommended brands — popups show per-engine answer mentions
District-targeting accuracy (“best studios in Kreuzberg” → are the returned studios in Kreuzberg?) is not measurable in this run: the Apify seed labels every studio’s district as “Berlin,” so in-district matches cannot register at all. Berlin yoga and Amsterdam hit the same wall. We report no figure rather than a misleading one; the caveats list carries the detail.
Six studies, one scoreboard
The recurring metrics, Berlin tattoo in bold at the end. Sibling values are re-read from each published article’s data on this site. Two caveats belong here rather than in the method, because they bear on these specific comparisons: the sibling studies ran on different dates, so a model update between runs cannot be fully separated from a vertical effect (Copilot 83%→95% is the row most exposed to this), and this run’s names were extracted by a Claude NER pass where the siblings used Gemini. Extraction is deterministic name-spotting with no scoring or ranking, and every extracted name was mechanically verified against its source answer, so the extractor cannot favour any engine’s studios — but it is a difference, and it sits on these rows. Sources — Berlin yoga, Marseille coffee, Tokyo bookstores, Paris yoga and Amsterdam bikes.
| Metric | Prior studies | Berlin tattoo | Verdict |
|---|---|---|---|
| ChatGPT own-website % | Paris yoga 32 · Berlin yoga 32 · Amsterdam bikes 42 · Tokyo 8 · Marseille 10 | 35% | Service baseline replicated |
| Copilot entity % | Berlin yoga 95 · Tokyo 89 · Marseille 83 | 95% | Slide reversed |
| Perplexity booking % | Berlin yoga 31 | 4% | Broken — tracks the vertical’s infrastructure |
| AI Mode google.com % | Berlin yoga 59 · Tokyo 61 · Marseille 80 | 53% | Series low, pattern intact |
| Top domain | Reddit (Tokyo) · USC blog (Berlin yoga) · Instagram (Marseille) | top10berlin.de (256) | Local listicle layer wins |
| EN vs local top-5 overlap | Paris 25 · Berlin yoga 25 · Marseille 11 | 67% | Broken upward — series high |
| Language→TLD coupling | Tokyo .jp 5.0× · Berlin yoga .de 1.5× | .de 0.87× | Coupling absent |
| Mentions #1 vs cites #1 | Marseille: split (Deep vs Nua artifact) | OMEN Tattoo — both | First clean dual-metric #1 |
Engine behaviour notes
ChatGPT: 104/104 searches, 104/104 map panels
Every single ChatGPT capture triggered web search and attached a Google-Maps-style venue panel — 1,031 panel entries, roughly 10 studios per answer. For local intent, ChatGPT’s answer is now effectively a mini maps result page with prose on top.
AI Mode: 53% self-citation
1,234 of AI Mode’s 2,350 citations point back at google.com surfaces. That’s its lowest self-cite share in our data (Marseille 80%, Tokyo 61%, Berlin yoga 59%) — the strong studio-website layer pulls even AI Mode outward, to 23% entity citations.
Gemini: the refusal engine
13 captures produced an answer recommending nothing — 11 of them Gemini, i.e. about 11% of its 104 captures. Source-mix shares are computed over cited URLs rather than answers, so a refusal contributes no entity citations either way. Gemini periodically declines tattoo recommendations outright (age-restriction caution, “consult a professional”). No other engine refused more than twice. If your category is 18+-adjacent, expect engine-level variance in whether you can be recommended at all.
Perplexity: fan-out on 82/104
Perplexity issued visible sub-queries on 82 of 104 captures — the only engine besides ChatGPT exposing its search decomposition here. Its answers also had the largest “other” source tail (10%), spread across small German directories.
If you ink in Berlin
- A real website earns citations here. 47% of all citations in this study point at studio sites; Copilot, Perplexity and Gemini are majority-website engines for this vertical. An Instagram-only presence forfeits that channel — and ChatGPT cited Instagram exactly zero times.
- Get into Berlin’s listicle layer. top10berlin.de out-cited every studio in the city, and the-berliner.com, tip-berlin and Mit Vergnügen carry ChatGPT’s third-party third. One placement there can move you more than months of on-site tweaks.
- Skip the booking-platform anxiety. The aggregator channel that dominates yoga barely exists for tattoo (2% of citations). A Fresha profile is fine housekeeping; as a visibility lever it does little here.
- Claim a Maps listing even if you’re appointment-only. The famous-artist tier proves you can be recommended without one — but every Maps-based surface (and ChatGPT’s ever-present venue panel) can’t show you. That panel appeared in 104 of 104 ChatGPT answers.
- Style pages are your language hedge. EN/DE results converge on the consensus names but split to 0–11% on realism, Japanese and famous-artist prompts — publish your style specialities in both languages to be in both answer sets.
The city was never the variable
Five cities in, it was still possible to argue that each place had its own AI-search climate — Berlin the aggregator city, Marseille the Instagram city, Tokyo the local-TLD city. Running a second vertical through Berlin closes that door. Every “Berlin” signature from the yoga study that had an infrastructural cause vanished the moment the infrastructure did, and every engine personality that was supposed to be stable stayed stable. The weight of that claim rests on a single same-city pair — Berlin yoga against Berlin tattoo, one city, two verticals. It is the cleanest test in the series and it is still n=1. Paris yoga is already published, so Paris tattoo would settle it.
What actually predicts an AI answer’s sources, on this evidence: does the vertical transact online (websites get cited), does it have marketplace inventory (aggregators get cited), does the city have a guide industry (listicles get cited), and does the community argue about it on Reddit (ChatGPT will be there). Tattoo scores yes / no / very yes / yes — and that profile, not anything about Berlin, is what this page documents.
The genuinely new thing tattoo contributes is the map-invisible tier — a class of heavily recommended businesses a Maps-derived dataset struggles to enumerate. The Mo Ganji correction sharpened what that means: the barrier is not always a missing listing, it is sometimes just a name the engines and the map file differently. Both leave the business unmeasurable by the same tools. I don’t yet know how big that tier is in other craft verticals (barbers? framers? luthiers?). It’s a good reason to keep going.
Study design
Data collection
- 26 prompt templates × 2 languages (EN/DE) × 2 proxy countries (US/DE) × 5 AI engines = 520 theoretical — and 520 captured. All 10 platform×proxy batches landed; this is the series’ first complete capture grid (AI Mode × DE included, consistent with Berlin yoga; the FR-proxy rejection seen in Paris/Marseille did not occur).
- Engines: ChatGPT, Perplexity, Gemini, Copilot, Google AI Mode — raw prose mode, captured 2026-07-15 via Bright Data
- 520 captures · 5,376 cited URLs · 1,031 map-panel entities · 3,444 extracted studio mentions
- Registry: 701-row Apify Google Maps seed (“tattoo studio,” Berlin) → 531 real tattoo businesses after category filtering (piercing-only, PMU, removal and beauty salons excluded) + 20 recovered via Google Places = 551 studios
- Scrapes ran through the pipeline’s in-process fallback runner (Modal was unreachable from this sandbox); identical payloads, tables and parsing as the standard Modal path. The NER pass likewise used a Claude extractor instead of the usual Gemini one, with the same rule set, the same deterministic resolution and the same output tables — both deviations changed which machine ran a step, not what was computed
What we measured
- Studios named per answer (brand-aggregated, dual-ranked by mentions and citation score)
- Cited URLs bucketed into a nine-part source taxonomy
- Perplexity booking share vs the Berlin yoga control
- EN vs DE top-5 overlap per prompt template
- .de vs .com citation balance by prompt language
- Unresolved-mention analysis (the unlisted-artist tier)
From prose answers to countable studios
Engines answer in free text (“OMEN in Kreuzberg is the safe pick; for fine line try NOIA or maybe Mo Ganji if you can get in…”), so recommendation counts require an extraction step. The pipeline’s NER pass reads each of the 520 answers with an LLM under a fixed rule set (named businesses only, recommendation context required, rank preserved, no inference), then a deterministic resolver matches each extracted name against the registry — exact normalised match first, then fuzzy similarity at a 0.86 cutoff. Frequently recommended names that fail to resolve are checked against Google Places by name + city: real studios the seed missed get added to the registry and re-counted (20 studios entered this way); the rest stay flagged as unresolved.
Extractor disclosure: this run’s extraction was performed by a Claude model rather than the pipeline’s usual Gemini extractor — the sandbox this study ran in could not reach the service holding the Gemini key. Same prompt rules, same deterministic resolution and storage. As a check, every one of the 3,444 extracted names was mechanically verified to appear in its source answer text (two contaminated captures were caught by this check and re-extracted by hand). Extraction differences between LLMs are a real, if small, cross-study variance source, so we flag it rather than hide it.
Caveats
- The leaderboard covers the Maps-resolved universe. Private artists whose mentions do not resolve to a Google Maps place (DotsToLines, Peter Aurisch, Valentin Hirsch) are quantified separately in section 7 — not because they aren’t real, and not necessarily because they lack a listing, but because they can’t be place-resolved. Mo Ganji was in this group until the 28 August resolver fix joined his mentions to his listing; at 28 mentions he still falls short of the 63 that 12th place required, as do the other three.
- District-targeting accuracy is unmeasurable here — the seed labels every studio’s district “Berlin,” so the 0–11% figures are a granularity artifact, matching the Berlin yoga and Amsterdam situations.
- NAJS Tattoo Studio’s citation score of 0 is a domain-matching blind spot (no website in the registry), not evidence of weak visibility — it ranks #8 by answer mentions.
- NER resolution is precision-first: mentions that can’t be confidently matched are dropped, so brand counts are conservative lower bounds.
- Citation counting over-weights AI Mode’s google.com self-references; cross-engine comparisons in this article always use per-engine percentages, never raw counts.
- Single-city, single-run snapshot (July 2026); engine behaviour shifts with model updates.
- Disclosure: no affiliation with any Berlin tattoo studio, and no tattoos were acquired in the making of this study.
Frequently Asked Questions
Summarize with AI
Continue Reading
The control study, and the rest of the series.