Star in AI search.
Where it wins, and where it fades.

This split between product and brand is fed by source classes that structurally omit brands. Italian recipe platforms — Giallozafferano, Cookaround, Fatto in Casa — teach the model that ragù, brodo and dado are steps in a method, written as ingredients with no producer attached. Wikipedia and dictionary entries cement the common-noun reading: brodo and dado are defined as foods, not trademarks. Reddit and forum threads debate a good ready-made ragù generically. Retailer pages — Esselunga, Coop, Conad — do carry the Star name, but as a single line item buried under category navigation the model reads as taxonomy, not endorsement. The net effect: every authoritative source that explains the category strips the brand out of it. The opportunity is to flood the sources AI reads with branded, structured associations — 'Gran Ragù by Star', 'Il mio Dado by Star' — through schema-marked recipe content, branded how-to assets and review sources that bind the product back to the name. Star must re-attach its name to the noun before the model finishes the sentence without it.

In the Google-grounded sources Gemini reads — the same retrieval that feeds AI Overviews — Recipe publishers (giallozafferano.it, cookist.it, fattoincasadabenedetta.it, tavolartegusto.it) score the highest topical quality (0.68) with 340 citations and frequent brand mentions. They are the de-facto GEO winner for 'ragù', 'sugo pronto' and 'dado' answers. When an Italian owner asks how to make a quick ragù, Gemini lifts technique straight from these domains and answers in-line — no click, no Star. The catch: these recipes read as method ('a good dado', 'a passata'), not branded ingredient lines, so Star gets mentioned but rarely as the named base. The fix is brand-anchored recipe placement: paid or earned recipes on GialloZafferano and Cookist whose ingredient list literally reads 'Gran Ragù Star' and 'Il mio Dado Star', marked up with Recipe schema so Google's generative layer parses Star as the ingredient. Living inside these recipe pages IS the new SEO for this category — without it, Star stays an unbranded technique in the answer box.

star.it is Gemini's seventh-largest source (127 citations) and its highest-sentiment, most on-message input (sent 5, quality 0.7) — the owned site genuinely shows up in the sources Google and Gemini read, which is rare for a heritage FMCG brand. Yet investigation I01 shows Star named zero times across 52 unprompted 'what do I cook tonight' probes. The paradox: AI pulls the owned site for branded, navigational queries ('Star Gran Ragù'), not for the everyday-need queries where the weeknight decision actually happens. star.it answers the questions Star already wins and stays silent on the ones it loses. The fix is structural: build a 'what do I cook tonight' / 'dinner in 20 minutes' weeknight hub on star.it/ricette, with FAQ schema and question-phrased H2s ('How do I make a good ragù without spending 4 hours?') that mirror the exact everyday queries. Owned content already reaches the sources AI reads — redirect that authority from brand pages toward the unbranded-need layer, so Gemini has a Star-anchored answer to surface in the zero-click box, not just a product spec sheet.

News & lifestyle media is Gemini's second-largest source class (773 citations, 16 domains, top domain 178 hits) and skews neutral-to-critical: greenme.it (eco/health), ilfattoquotidiano.it, fanpage.it, starbene.it. These high-authority (auth 6) Italian outlets steer their food coverage toward additives, sodium in dado and ultra-processed framing — the exact axis where a stock-cube-and-jarred-sauce brand is most exposed. It is a zero-click reputation risk: when an owner asks whether dado is bad for you, Gemini grounds on these outlets and answers in-line, and Star gets judged by a category critique it never authored. Star has no counter-narrative in this class today. The move is earned editorial that reframes the story on Star's own terms — heritage since 1948, reduced-salt and clean-label reformulation, 'dado versus brodo made from scratch' comparisons — placed with these same Italian lifestyle desks. The goal isn't to win the health debate but to be a cited voice inside it, so the generative answer carries a Star data point instead of only the critical frame.

Wikipedia appears in both models' sources but is one of the smallest inputs (Gemini: 12 citations, single domain, auth 7, sent 0, quality 0.6). For a 1948 heritage brand owned by GBfoods Italia, that is a structural void: Wikipedia is the neutral fact-anchor LLMs lean on for founding date, ownership, product lines and provenance, and it disproportionately seeds the 'about the brand' scaffolding both Gemini and GPT-5 reuse. A thin or stub entry forces the models to reconstruct Star's heritage from fragmented long-tail blogs instead. Because Wikipedia is high-trust and frequently surfaced verbatim in zero-click panels, the cheapest GEO win here is editorial, not paid: stand up a well-sourced Italian-language Wikipedia entry covering Star's 1948 origin, the Gran Ragù / Il mio Dado / Star Brodo / sughi pronti lines and GBfoods ownership, each line backed by independent citations (trade press, La Cucina Italiana). This hardens the factual baseline every other model answer is built on and stops competitor framing from filling the gap.

The two models diverge sharply in their noisiest source class. Gemini's Other/long-tail (866 citations, 245 domains) stays at least food-adjacent — home.blog, bell-italia.com, affaritaliani.it. GPT-5's long-tail (674 citations, 303 domains) drifts into torrinomedica.it (a medical reference), laby.trieste.it, spice.alibaba.com (B2B commodity), fundo.be — domains with weak focal relevance (quality 0.38). GPT-5 is rebuilding fragments of Star's category from pharmaceutical and wholesale-spice contexts, risking off-key associations (sodium/health, commodity ingredient) that have nothing to do with the everyday-kitchen positioning. Because GPT-5 runs its own index — not Google's — winning Gemini's zero-click box does not fix GPT-5. The move is to flood GPT-5's preferred mid-authority surfaces with on-brand signal: La Cucina Italiana (already a GPT-5 recipe source), well-structured retailer PDPs and clean product schema, so the model has dense, correct food-context to outweigh the medical and B2B noise. Track the divergence: a query that surfaces Star warmly on Gemini can surface it incoherently on GPT-5.

Retailers & grocery is a large Gemini source class (386 citations, top domain 100) spanning carrefour.it, conad.it, dambros.it, cicalia.com — but the notes are damning: thin editorial and mildly positive framing (quality 0.55, sent 1). These PDPs prove availability and SKU names, so Gemini can confirm Star is real and buyable, but they carry no descriptive substance for the generative layer to lift into a recommendation. In a zero-click answer, a retailer page contributes a price and a pack size, not a reason to choose. The GEO opportunity is to upgrade retailer PDP content wherever Star can influence it: richer product descriptions, structured attributes (no-added-glutamate, recipe suggestions, usage occasions) and, crucially, seeded customer Q&A and reviews on the retailer pages and on opinioni.it / trovaprezzi.it. Star should treat retail PDPs as citation real-estate, not just transaction pages — because for many 'which is the best ready-made ragù' queries, the retailer page is the only commercial source the model can point to, and right now it gives the model nothing persuasive to repeat.

Unlike CPG brands plagued by skeptical forum chatter, Star's risk in the crowd-voice layer is silence, not hostility. Reddit & forums is tiny (Gemini: 35 citations, auth 2, quality 0.32, sent 0) and Consumer reviews & watchdog (opinioni.it, opinioncity.it, trovaprezzi.it) is moderate (175 citations) with mixed sentiment (sent 1). The read cuts both ways: no community veto suppresses Star, but almost no authentic consumer advocacy exists for the model to cite as social proof. When an owner asks whether people are happy with Gran Ragù, Gemini has thin, mixed material to work from. Because community signal earns trust precisely because it is independent, the move is to legitimately grow review density on opinioni.it and trovaprezzi.it and to take part in Italian cooking subreddits and forums (cookaround community) — not astroturf, but genuine prompts to satisfied buyers and recipe communities. A richer base of real, positive, specific reviews gives the zero-click layer corroborating voices, converting Star's 'under-surfaced but liked' status (I02: 35 endorsements of 44) into citable social proof.

YouTube is a meaningful Gemini source (159 citations, single platform, prev 6) yet scores low on authority and brand depth (auth 3, quality 0.45) — it is the cooking and review video the model pulls for technique, with variable brand-specific depth. For a category built on the act of cooking ('how to make ragù', 'quick dinner'), video is exactly where the unbranded-technique reflex from I01 gets reinforced: the assistant watches a from-scratch demo and answers with method, not a Star base. Because Google surfaces its own YouTube content prominently in generative answers, owned and creator video is a controllable GEO lever Star is currently ignoring. The move: a Star YouTube presence and Italian food-creator partnerships producing 'ragù in 15 minutes with Gran Ragù' and 'risotto with Dado Star' in short and long form, titled and transcribed with the exact everyday-query phrasing. Transcripts and descriptions are what Gemini parses — so captioned, keyword-aligned video gives the model branded weeknight technique to cite, planting Star inside the cooking-demo layer that today resolves entirely to generic method.