SEO de entidades

Como fazer os sistemas de IA identificarem e confiarem na sua marca com segurança — o Knowledge Graph, a desambiguação de entidades, o schema sameAs e por que menções de marca têm correlação mais forte com citações de IA do que backlinks.

Publicado pela primeira vez: 24 de jun. de 2026 · Última atualização: 3 de ago. de 2026 · Avançado
1 sinal de evidência nesta página

SEO de entidades ajuda os sistemas de IA a identificar com confiança quem ou o que você é — pesquisas da Ahrefs encontraram correlação entre menções de marca e citações no AI Overview mais forte do que entre backlinks e citações, embora correlação não prove que uma coisa cause a outra.

«> TL;DR — Entity SEO is practitioner terminology, not a documented Google

ranking system — but for AI the emphasis shifts hard toward disambiguation and corroboration. Google says the Knowledge Graph shed a large share of its entities in a June 2025 cleanup; the exact scale and its effect on any specific AI product aren’t things I can independently verify here, so treat that as directional, not gospel. In Ahrefs research, branded web mentions and YouTube presence correlated with AI-answer visibility far more strongly than backlinks did — correlation, not proof of cause. You also can’t shortcut trust with fake corroboration: an independent experiment seeding fictional experts into hundreds of press articles produced essentially no AI citations. And treat sameAs schema as most robust when it’s server-rendered — whether a given AI crawler executes JavaScript is provider- and version-specific, so don’t assume none of them do. » (Tradução) (Síntese localizada do trecho dezesseis, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«## Entity SEO is just SEO — with a different emphasis for AI » (Tradução) (Síntese localizada do trecho dezessete, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Start with the bounded definition: “Entity SEO” is practitioner terminology, not the name of a documented Google ranking system. Evidência desta afirmação Official or primary documentation supporting the adjacent article claim, with scope limited to the source's published description. Escopo: No ranking guarantee or undisclosed system mechanics are inferred beyond the cited source. Confiança: alta · Verificado: Schema.org: Thing It’s a useful lens on existing SEO practice — identity clarity, corroboration, and accurate structured data — not a separate discipline or a universal AI optimization layer with its own scoring model. Use those practices as clarity signals, not guaranteed ranking levers. Evidência desta afirmação Official or primary documentation supporting the adjacent article claim, with scope limited to the source's published description. Escopo: No ranking guarantee or undisclosed system mechanics are inferred beyond the cited source. Confiança: alta · Verificado: Google: Structured data introduction » (Tradução) (Síntese localizada do trecho dezoito, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«My standing framing hasn’t changed for traditional search: “the entity identification part is more on Google’s end than ours.” You don’t manually “do entity SEO” by stuffing entities into a page. You publish quality content, mark it up with accurate structured data, keep your signals consistent, and Google does the identification. » (Tradução) (Síntese localizada do trecho dezenove, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«What has changed is the weighting for AI search. When a model writes an answer, it pulls sentences and names, not link equity — so the brand-signal picture looks different from the link-driven world of rankings. The mechanics below are about making your entity unmistakable to that machine. » (Tradução) (Síntese localizada do trecho vinte, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«## The Knowledge Graph, and what we can and can’t say about AI » (Tradução) (Síntese localizada do trecho vinte e um, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Google launched the Knowledge Graph in 2012 with the pitch “things, not strings” — a model that “understands real-world entities and their relationships to one another.” It was built to solve disambiguation (which “Taj Mahal”?), summarization, and discovery, seeded from Freebase, Wikipedia, and the CIA World Factbook. Those 2012 numbers are historical; Google hasn’t published a current entity count. » (Tradução) (Síntese localizada do trecho vinte e dois, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Here’s where I want to be careful: it’s tempting to say “the Knowledge Graph now feeds AI Overviews, AI Mode, and Gemini, so being in it makes you eligible for AI-powered answers.” Google confirms the Knowledge Graph powers Search features like panels, drawing on hundreds of sources across the open web. But I don’t have product-specific primary documentation tying Knowledge Graph membership directly to eligibility for every AI surface, and I’m not going to assert a cross-product causal claim I can’t back with a citation. Treat “the KG is AI infrastructure” as a plausible industry read, not a confirmed mechanism. » (Tradução) (Síntese localizada do trecho vinte e três, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«What is publicly reported: journalist Jason Barnard (Kalicube) documented a large Knowledge Graph contraction in June 2025, with steep cuts concentrated in stale event entities and ambiguous “Thing”-category entities, alongside more precise person-entity typing. I’m not going to restate the exact percentages here — they come from one third-party analysis of Google’s data, not a source I can independently reproduce, and I’d rather point you to the original writeup than risk a stale or mistyped figure. The directional read holds up: as Barnard put it, “in the Knowledge Graph, clarity is the only point of entry.” Quality and disambiguation appear to matter more than raw signal volume — see the source for the full breakdown. » (Tradução) (Síntese localizada do trecho vinte e quatro, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«## The entity signal hierarchy — directionally, not exactly » (Tradução) (Síntese localizada do trecho vinte e cinco, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«This is the part most “entity SEO” advice gets wrong: it treats loosely related signals as if they were proven ranking levers. We ran correlation studies at Ahrefs across large samples of brands and AI answers, and the directional pattern is worth knowing even though I’m holding back the exact coefficients here pending independent verification against the original datasets (correlation studies are easy to mis-cite, and a correlation is not causation regardless of the number). » (Tradução) (Síntese localizada do trecho vinte e seis, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Directionally, across the studies: branded web mentions and YouTube presence/mentions correlated with AI-visibility metrics (like AI Overview brand mentions) noticeably more strongly than backlinks did — brand-mention and YouTube signals outranked link-based signals like Domain Rating and referring domains in the rankings we measured. Branded search volume also behaved as a leading indicator rather than a lagging one. As I summed up in my cross-platform AI brand-visibility correlations study: “Google favors brands and Perplexity seems to show some favoritism as well. What’s surprising is the weakness of ChatGPT here.” » (Tradução) (Síntese localizada do trecho vinte e sete, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«We also found meaningfully limited overlap between which sources get cited across Google, ChatGPT, and Perplexity — a small minority of sources showed up across all three simultaneously. Entity SEO is a multi-platform exercise, not a single-engine one. For the exact coefficients and sample sizes, see the linked study and the AI Overview brand-correlation study directly rather than relying on numbers restated here. » (Tradução) (Síntese localizada do trecho vinte e oito, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«## Disambiguation is the #1 priority » (Tradução) (Síntese localizada do trecho vinte e nove, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«AI cannot cite an entity it cannot confidently identify. If your brand shares a name with another company, has inconsistent NAP (name/address/phone) data, or has no external verification, you get skipped. The June 2025 cleanup’s whole point was raising the share of unambiguously typed entities — Google is optimizing for the brands it can be sure about. » (Tradução) (Síntese localizada do trecho trinta, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«So disambiguation is the job: a consistent canonical identity, declared once on your entity home (About page or homepage) via Organization JSON-LD with a stable @id, and confirmed against external references with sameAs. John Mueller’s one caveat: don’t point sameAs at a Knowledge Graph ID URL (kg:/m/...) because “that ID might change” — use stable URLs like Wikidata, Wikipedia, and social profiles instead. » (Tradução) (Síntese localizada do trecho trinta e um, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«## You can’t fake corroboration — the entities.org experiment » (Tradução) (Síntese localizada do trecho trinta e dois, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Here’s the most important guardrail in the whole field. Researchers at Entities.org describe AI systems wanting entity consensus — independent, high-confidence sources agreeing about the same claim before they’ll assert it. Below a threshold of independent corroboration, AI hedges; above it, AI asserts. » (Tradução) (Síntese localizada do trecho trinta e três, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«The headline finding: a controlled experiment seeded fictional experts into a large number of press articles and measured the result across nine models — essentially no AI recommendations resulted. Volume without independence fails; the researchers’ framing is that AI systems can detect synthetic consensus. I’m not restating their exact article count, model-recommendation threshold, or citation-share figures here — those are specific numbers from one third-party research project that I haven’t independently reproduced, so check the original research for the precise counts. What I’m comfortable asserting directionally: self-published content alone rarely carries a brand into an AI answer, and most AI answers appear to lean on third-party sources rather than brand-published ones. » (Tradução) (Síntese localizada do trecho trinta e quatro, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«This is the deep reason “entity stacking,” mass press-release seeding, and buy-in-bulk link campaigns don’t work for AI visibility. There’s no hidden entity score to game — earn the real mentions. » (Tradução) (Síntese localizada do trecho trinta e cinco, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«## sameAs, Wikidata, and Wikipedia » (Tradução) (Síntese localizada do trecho trinta e seis, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Neither Wikipedia nor Wikidata is a documented requirement for Google or AI recognition — Google names Wikipedia as one of many common sources it can draw on for knowledge panels, not a mandatory input. That said, they’re useful, accessible external references worth prioritizing in your sameAs list, roughly in this order: » (Tradução) (Síntese localizada do trecho trinta e sete, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«1. Wikidata — machine-readable and low barrier to entry (no Wikipedia-style notability gate); several AI systems and Google are reported to draw on it, though I don’t have primary documentation for exactly how every provider weights it. 2. Wikipedia — if you’re notable. Third-party research cites it as a large citation source for some AI systems, though I’m not restating the exact percentage here — check the source research for the current figure and its methodology. Wikipedia content is also used for real-time grounding by some systems, not just training. You can’t write your own page — earn the press coverage that establishes notability first. 3. LinkedIn / Crunchbase / official registrations, then social profiles. » (Tradução) (Síntese localizada do trecho trinta e oito, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«One cautionary tale from our own team: Ryan Law accidentally led Google to think he owned the Ahrefs website by adding schema to his personal site with an error in the sameAs property. Always double-check schema before you push it live. » (Tradução) (Síntese localizada do trecho trinta e nove, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«## Structured data: indirect lever, not a citation button » (Tradução) (Síntese localizada do trecho quarenta, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Be clear-eyed about what schema does. An Ahrefs observational study by Louise Linehan and Xibeijia Guan tracked a large set of pages that added JSON-LD over roughly seven months and compared citation volume before and after — see the schema/AI citations study for the dataset and exact percentages. The directional finding: adding schema to an already-cited page did not produce a clear positive lift in AI citation volume across the platforms tracked. » (Tradução) (Síntese localizada do trecho quarenta e um, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«So treat schema as not a reliable citation lever. Its value is entity disambiguation — connecting your brand to its Knowledge Graph identity — and correct indexing, not a direct bump in how often AI quotes you. Mueller’s own position lines up: structured data is not a direct ranking factor, though it helps machines understand entities and is genuinely useful for shopping data that’s “basically impossible to read in high fidelity… from a text page.” » (Tradução) (Síntese localizada do trecho quarenta e dois, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«For AI specifically, @id on every entity and an @graph connecting Organization, Person, and Article into one graph is a useful modeling pattern practitioners recommend for keeping a connected identity consistent — it isn’t a documented Google or AI-provider requirement, and using it doesn’t guarantee a ranking, panel, or citation effect. Isolated schema blocks for the same entities work too; a connected graph is about making maintenance and consistency easier, not unlocking a feature. » (Tradução) (Síntese localizada do trecho quarenta e três, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«## Server-render your schema — don’t assume any crawler runs your JS » (Tradução) (Síntese localizada do trecho quarenta e quatro, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«This is a technical mistake that can silently sink otherwise-good entity work, and it’s worth being precise about what’s actually documented rather than repeating a blanket claim. Google explicitly documents rendering JavaScript-generated structured data for Search. Whether a given AI crawler — GPTBot, ClaudeBot, PerplexityBot, or any other named provider — executes JavaScript is provider- and version-specific, and it changes over time; I don’t have current, per-provider primary documentation to assert a universal “AI crawlers don’t run JavaScript” rule, and I’d rather send you to verify the current behavior yourself (a dated fetch test against the provider you care about) than repeat an unverified generalization. » (Tradução) (Síntese localizada do trecho quarenta e cinco, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«The safe, provider-agnostic move either way: treat server-rendered schema in the initial HTML as the robust baseline. If a crawler you’re targeting turns out not to render JavaScript, only server-delivered schema is guaranteed visible to it; if it does render JS, server-rendering costs you nothing. Check your sameAs and @graph in the raw HTML response (not just the rendered DOM) so you know what’s actually being served before any rendering step. » (Tradução) (Síntese localizada do trecho quarenta e seis, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«## So what do you actually do? » (Tradução) (Síntese localizada do trecho quarenta e sete, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«This is practitioner guidance — a way I find useful to organize the work, not a documented AI-system scoring model. I think about it as three layers: identity, relationships, and independently verified corroboration. » (Tradução) (Síntese localizada do trecho quarenta e oito, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Identity. Nail your entity home — one canonical page, Organization/Person schema server-rendered as the robust baseline, and accurate sameAs. Keep NAP (name/address/phone) and descriptions consistent everywhere, including your Google Business Profile if you’re local — a well-documented Google data source, even though I can’t confirm it’s the primary local source for every AI system. » (Tradução) (Síntese localizada do trecho quarenta e nove, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Relationships. Connect your entities — sameAs to Wikidata (and Wikipedia if you’re notable), an @graph linking Organization, Person, and Article where it helps you keep things consistent. In your content, cover the entities and relationships your readers actually need explained — don’t chase a Cloud Natural Language salience score as if it were a ranking input; salience describes how a document-analysis API scores a piece of text, not a documented Google Search or AI-citation ranking factor. » (Tradução) (Síntese localizada do trecho cinquenta, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

«Corroboration. The layer that moves the needle most in my experience — earn genuine third-party mentions and a real YouTube presence, because those (not links alone) are strong signals of whether AI has independent, corroborated confidence you’re who you say you are. As I’ve said about the AI era, “SEO itself hasn’t drastically changed, but getting good results may now require closer collaboration with other teams like PR and partnerships.” » (Tradução) (Síntese localizada do trecho cinquenta e um, parte um: o texto-fonte foi preservado para conferência na revisão nativa.)

Adicionar uma nota de especialista

Fixar uma citação de especialista

É uma pessoa nova? Crie o perfil não reivindicado dela em /admin/experts/ → Fixar uma citação de especialista primeiro.