We put 18 real purchase questions to the 5 major AI engines, two weeks in a row. In 58% of the answers, no major bookstore is recommended.
Original study · WebsemEpisode 231 July 202616 min
More and more readers no longer open Google when they look for “books for children aged 6-8”. They ask ChatGPT, Gemini or Perplexity — and get a single answer, not ten blue links. We wanted to know: who gets the customer inside that answer?
We put 18 real purchase questions to the 5 major AI engines, two weeks in a row, without naming a single brand in the question. We analysed 180 answers and counted 247 mentions. The short answer: in more than half of the cases, nobody gets the customer.
180
AI answers analysed
18 questions × 5 engines × 2 runs
247
brand mentions counted
21 brands tracked
58%
answers with no major brand
only 75 of 180 name one
422
distinct domains cited
across 172 answers with sources
This is the second study in our series on the AI visibility of Romanian brands, after the luxury jewelry study. This time we changed the methodology: not a single prompt photographed on one day, but a panel of 18 questions, run weekly, across 5 engines, with positions, citations and sentiment measured systematically.
We measured the visibility of book-market brands in the answers of 5 AI engines, using a fixed panel of questions, run repeatedly, in Romanian, from Romania.
| Window | 18–31 July 2026, with two full runs of the panel (20 and 27 July) |
|---|---|
| Questions | 18 purchase prompts, with no brand name in the text |
| AI engines | ChatGPT, Google Gemini, Perplexity, Google AI Mode, Google AI Overviews |
| Answers analysed | 180 (36 per engine) |
| Brands tracked | 21 — online bookstores, marketplaces and publishers |
| Localisation | Romania, Romanian language |
| What we measured | brand mentions, the position of the mention in the answer, the domains cited as sources, sentiment |
| Measurement platform | LLM Pulse (automated monitoring, weekly runs) |
Each of the 18 questions was put to each of the 5 engines, across 2 weekly runs — hence the 180 answers analysed.
The 5 engines: ChatGPT (OpenAI) · Google Gemini (Google) · Perplexity (Perplexity AI) · Google AI Mode (Google) · Google AI Overviews (Google).
The questions cover real purchase territories: children's books, discounts and low budgets, personal development, psychology, business, parenting, holiday fiction, English-language books, fantasy, cooking, nutrition, the preparatory school year. They are phrased the way people ask, not the way an SEO specialist would. The full list is below.
The study does not measure sales, traffic, real market share or service quality. It measures one thing: the extent to which a brand's name appears in the answer an AI engine gives a buyer. A brand invisible in AI can be a market leader; a very visible brand in AI can be small. The overlap between the two hierarchies is precisely the subject of study, not an assumption.
An important limit, stated up front: we track 21 main brands, not the whole market. Smaller players also appear in the answers (librex.ro, clb.ro, librariadelfin.ro, booknation.ro); we see them in the citations, but we do not count their brand mentions. The rest of the limits, in section 09.
247 brand mentions across 180 answers. The mention rate = the share of the 180 answers in which the brand appears.
Brand mentions · the percentage = share of the 180 answers analysed · Websem study, July 2026
| # | Brand | Type | Mentions | Rate | SoV 27 Jul |
|---|---|---|---|---|---|
| 1 | Librislibris.ro | online bookstore | 37 | 20.6% | 13.4% |
| 2 | Literalitera.ro | publisher | 30 | 16.7% | 10.9% |
| 3 | Cărtureșticarturesti.ro | bookstore chain + online | 29 | 16.1% | 11.8% |
| 4 | BookZonebookzone.ro | online bookstore + publisher | 24 | 13.3% | 9.2% |
| 5 | elefant.roelefant.ro | marketplace | 22 | 12.2% | 10.9% |
| 6 | Editura Treiedituratrei.ro | publisher | 17 | 9.4% | 4.2% |
| 7 | Humanitashumanitas.ro | publisher + bookstores | 13 | 7.2% | 6.7% |
| 7 | Editura Arthureditura-arthur.ro | publisher | 13 | 7.2% | 6.7% |
| 9 | Nemiranemira.ro | publisher | 10 | 5.6% | 4.2% |
| 9 | Curtea Vechecurteaveche.ro | publisher | 10 | 5.6% | 2.5% |
| 9 | Editura Corintedituracorint.ro | publisher | 10 | 5.6% | 6.7% |
| 12 | eMAGemag.ro | marketplace | 9 | 5.0% | 5.0% |
| 13 | Books Expressbooks-express.ro | online bookstore | 7 | 3.9% | 2.5% |
| 14 | Okianokian.ro | online bookstore | 4 | 2.2% | 1.7% |
| 14 | Târgul Cărțiitargulcartii.ro | online second-hand bookstore | 4 | 2.2% | 0.8% |
| 16 | Polirompolirom.ro | publisher | 3 | 1.7% | 0.8% |
| 17 | Editura ARTeditura-art.ro | publisher | 2 | 1.1% | 0.8% |
| 18 | RAOraobooks.com | publisher | 1 | 0.6% | — |
| 18 | Librăria Românălibrariaromana.ro | online bookstore | 1 | 0.6% | 0.8% |
| 18 | Librarullibrarul.ro | online bookstore | 1 | 0.6% | — |
| 21 | Editura Aramisedituraaramis.ro | publisher | 0 | 0.0% | — |
SoV = share of voice, the brand's share of all mentions in the 27 July run
Litera has more mentions (30) than Cărturești (29), BookZone (24) or elefant.ro (22). Overall, 44% of mentions are publishers, even though all 18 questions are about buying books. When the AI recommends “where to get the book”, it often answers with who published it.
The market leader appears in just one answer out of five. In classic SEO, a brand ranked first gets clicks every single time somebody searches. In AI, “position 1” is a weighted lottery replayed at every question — and in 58% of the draws nobody from the top wins.
Each brand's mention rate, per engine — a percentage of that engine's 36 answers. The top 10 brands:
| Brand | ChatGPT | Gemini | Perplexity | AI Mode | AI Overviews |
|---|---|---|---|---|---|
| Libris | 13.9% | 30.6% | 16.7% | 13.9% | 27.8% |
| Litera | 13.9% | 25.0% | 5.6% | 27.8% | 11.1% |
| Cărturești | 11.1% | 25.0% | 8.3% | 16.7% | 19.4% |
| BookZone | 8.3% | 8.3% | 11.1% | 22.2% | 16.7% |
| elefant.ro | 11.1% | 25.0% | 11.1% | 5.6% | 8.3% |
| Editura Trei | 8.3% | 5.6% | 11.1% | 11.1% | 11.1% |
| Humanitas | 5.6% | 19.4% | 5.6% | 5.6% | — |
| Editura Arthur | 8.3% | 11.1% | — | 11.1% | 5.6% |
| Nemira | 2.8% | 16.7% | 5.6% | 2.8% | — |
| Curtea Veche | — | 5.6% | 2.8% | 13.9% | 5.6% |
| Brands / answer | 0.9 | 2.2 | 0.8 | 1.6 | 1.4 |
A mention in position 1 (“the most recommended bookstore is…”) and one in position 8 (inside a footnote list) count the same in a mention leaderboard — but not for the customer. This is where the distance between the leader and the rest of the market becomes dramatic.
Libris doesn't just appear most often — it appears first two times out of three. In the position-weighted visibility score (position 1 = 100%, position 2 = 50%, position 3 = 33% and so on), the market hierarchy is rewritten:
| # | Brand | Weighted visibility | Rank by mentions |
|---|---|---|---|
| 1 | Libris | 2,857 | 1 |
| 2 | Cărturești | 1,503 | 3▲1 |
| 3 | Editura Trei | 1,410 | 6▲3 |
| 4 | Litera | 1,399 | 2▼2 |
| 5 | BookZone | 970 | 4▼1 |
| 6 | elefant.ro | 868 | 5▼1 |
Sum of scores across the 5 engines, the full 14-day window
Editura Trei is the textbook case: 6th by mention count, 3rd by weighted visibility. When it shows up — in psychology, personal development, fiction — it shows up high. Few territories, but owned. It is the opposite strategy to a marketplace's: not wide, shallow visibility, but ownership of a handful of questions.
Every question in the panel is a sales category. We counted each brand's mentions on each question (~10 answers per question). The result: the book market lives in three different realities.
Which is the best online bookstore for children's books?
Top online bookstores with book discounts in Romania
Books under 20 lei for gifts and light reading
How do I find good books for children aged 6-8?
How do I order cheap books without losing on quality?
Alternatives to the big online bookstores for English-language books
Which fantasy books for children do you recommend?
The best fiction books to read on holiday
How do I choose the right books for the preparatory school year?
Top psychology books for beginners
I'm looking for parenting books that are easy to apply day to day
Which personal-development book should I pick to build my self-confidence?
Which cookbook is worth it if I want simple, fast recipes?
I need books on nutrition and health for a better lifestyle
What are the best English-language books for easy reading?
What are the best business books for entrepreneurs?
Which books should I get for a teenager who barely reads?
How do I choose between a motivational book and a psychology one?
The questions about where you buy — “the best online bookstore for children”, “top bookstores with discounts”, “books under 20 lei” — gather 125 of the 247 mentions on their own. This is where the real battle is, and where the bookstores show up.
On “alternatives for English-language books”, the runaway winner is Books Express (6 mentions), followed by Okian (4) — two brands nearly absent from the rest of the study dominate an entire niche because they are specialised in exactly that. On “the preparatory year”, the podium is publishers only: Corint 3, Arthur 3 — not a single bookstore.
6 questions out of 18 gathered at most 3 brand mentions in total, across every engine, in both weeks: business (1 mention in the whole market), teenagers (1), motivational vs. psychology (1), easy English reading (2), cooking (3), nutrition (3). The AI recommends titles and authors, but never says where to buy.
A third of the book market's purchase territories currently belong to nobody. The same conclusion holds for the study as a whole: out of 180 answers, only 75 (42%) mention at least one major brand. And of those, only 19 recommend a single brand — the rest share the stage between 3.3 brands on average.
Engines with live search build their answers from web pages they cite. We counted the domains: 422 distinct domains were cited across the 180 answers.
34 answers, above Facebook (11), above YouTube (6), above any media publication. When somebody asks an AI which bookstore to choose, the answer comes, in part, from conversations between readers — threads no brand controls, but in which any brand can be the subject.
clb.ro, librex.ro, librariadelfin.ro, booknation.ro, printrecarti.ro — sites with little public recognition show up constantly in the AI's sources. Models don't cite by fame; they cite what they can crawl and understand.
Cărturești is cited in 46 answers but named as a brand in 29 — the site works harder than the name collects. The extreme cases, though, deserve their own chapter.
Cross-referencing citations with mentions reveals two species of brand, with diametrically opposite problems.
| Brand | Cited | Gap | Named | Diagnosis |
|---|---|---|---|---|
| Târgul Cărții | 18 | 4 | anonymous information supplier | |
| Librarul | 14 | 1 | anonymous information supplier | |
| eMAG | 15 | 9 | under-named relative to citations | |
| Editura Trei | 5 | 17 | lives off the model's memory | |
| Nemira | 1 | 10 | lives off the model's memory | |
| Humanitas | 0 | 13 | lives off the model's memory |
The AI crawls their site, uses their listings and prices as raw material for the answer — and then recommends somebody else. Their content is considered credible; their entity is not. Typically this points to weak brand signals: incomplete structured data, no consolidated identity (sameAs, profiles, reviews), a generic name that is hard to tie to the site. Librarul.ro is the boundary case: 14 citations, exactly one mention.
They are named without being cited, out of the models' parametric knowledge — reputation accumulated over years of cultural presence, press, reviews. Humanitas is the purest case: 13 mentions, zero citations of humanitas.ro across the entire measured window. A brand built over decades is recommended purely from the models' memory, without its site ever being used as a source. It is a flattering position, but a fragile one: wherever the answer is built from crawlable sources and your site isn't among them, you drop out of the conversation.
Have both: be cited as a source and named as a recommendation. No brand in the study achieves that on all five engines at once — the closest are Libris and Cărturești.
The same 18 questions, the same 5 engines, 7 days apart.
| Brand | 20 Jul | Movement | 27 Jul | Change |
|---|---|---|---|---|
| Editura Corint | 2 | 8 | ×4 | |
| elefant.ro | 9 | 13 | +44% | |
| Humanitas | 5 | 8 | +60% | |
| Editura Arthur | 5 | 8 | +60% | |
| eMAG | 3 | 6 | ×2 | |
| Litera | 17 | 13 | -24% | |
| Curtea Veche | 7 | 3 | -57% | |
| Libris | 21 | 16 | -24% | |
| Editura Trei | 12 | 5 | -58% | |
| Total mentions, whole market | 128 | 119 | -7% |
In share of voice, elefant.ro rose from 7.0% to 10.9% in a single week — a jump that in organic Google would take months. Editura Trei lost more than half its mentions in seven days.
The correct reading is not “chaos” but plasticity: AI answers are regenerated at every question, from re-crawled sources and constantly updated models. For today's leaders, that means the advantage requires maintenance. For everybody else, it means the window to enter is permanently open — unlike classic SEO, where positions 1–3 are set in concrete by years of authority.
We declare them ourselves, before anyone else finds them.
Enough to see the volatility, not enough for long-term trends. The study continues; data accumulates weekly.
The panel covers the big territories, but not exhaustively — textbooks, technical books, comics and others are missing.
Smaller players (librex.ro, clb.ro, librariadelfin.ro, booknation.ro) show up in citations, but we do not count their mentions. The leaderboard covers the main brands, not the whole market.
At the level of an individual question, the difference between 3 and 4 mentions is statistical noise. The solid conclusions are the aggregate ones.
We measure presence in answers, not the sales it generates. The link between the two is not demonstrated by this study.
The figures describe engine behaviour in the 18–31 July 2026 window, not a permanent truth.
The source percentages are computed against the 172 answers that contain citations, not all 180 — the remaining answers were generated without links.
Queries from Romania, in Romanian, with no history and no personalisation. A logged-in user with a purchase history may get different answers.
58% of answers recommend no major brand, and a third of the purchase territories are empty. The first brand that builds citable content for “business books”, “books for teenagers” or “cooking” will compete with nobody — it will be the only answer.
The Libris case shows real dominance is “first recommended”, not “among those mentioned”. Position-weighted visibility should be the KPI, not the raw mention count.
Litera dominates AI Mode, Libris dominates Gemini, Books Express dominates the English-language niche. The “one good site for all engines” strategy is not supported by the data — sources, retrieval logic and mention generosity differ radically.
Reddit is the market's second source, above any media publication. Presence in organic conversations — Reddit, forums, groups — is no longer optional PR, it is direct input into AI answers.
If your site is cited but your name is missing from recommendations, the problem isn't the content — it's the entity signals: structured data, a consolidated identity, coherence between name, domain and profiles. It's the fastest category of win in all of AEO.
Weekly volatility shows no position is set in concrete. The book market in AI is where SEO was in 2005: those who build systematically first will be the “default answer” when search behaviour shifts for good.
Every figure on this page comes from these files. They are published under a CC BY 4.0 licence: you may use, republish and dispute them, provided you cite the source.
Raw data
Charts for republication (SVG)
Vector files, with the source written into the chart. Free to use in articles and presentations.
Want to know how your brand looks inside AI answers? Websem monitors AI-engine visibility for Romanian brands and builds AEO/GEO strategies on data, not intuition — get in touch.
Study by Websem · data collected 18–31 July 2026 · published 31 July 2026. The study measures AI-engine behaviour, not the quality of the companies mentioned.